diff --git a/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/submission.json b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/submission.json new file mode 100644 index 0000000000..5188cd4c1d --- /dev/null +++ b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/submission.json @@ -0,0 +1,14 @@ +{ + "val_bpb": 1.06287, + "val_loss": 2.326951, + "seeds": [42, 0, 1234], + "per_seed": { + "42": {"steps": 4989, "pre_quant_bpb": 1.06749, "quantized_bpb": 1.07678, "post_ttt_bpb": 1.06366, "artifact_bytes": 15909254}, + "0": {"steps": 4974, "pre_quant_bpb": 1.06685, "quantized_bpb": 1.07608, "post_ttt_bpb": 1.06311, "artifact_bytes": 15904209}, + "1234": {"steps": 4973, "pre_quant_bpb": 1.06578, "quantized_bpb": 1.07509, "post_ttt_bpb": 1.06183, "artifact_bytes": 15909401} + }, + "hardware": "8xH100 SXM 80GB", + "region": "NA-1", + "base_pr": 1787, + "git_commit": "1d12cb6" +} diff --git a/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_gpt.py b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_gpt.py new file mode 100644 index 0000000000..ec750ef327 --- /dev/null +++ b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_gpt.py @@ -0,0 +1,3999 @@ +import base64, collections, copy, fcntl, glob, hashlib, io, lzma, math, os +from pathlib import Path +import random, re, subprocess, sys, time, uuid, numpy as np, sentencepiece as spm, torch, torch.distributed as dist, torch.nn.functional as F +from torch import Tensor, nn +from flash_attn_interface import ( + flash_attn_func as flash_attn_3_func, + flash_attn_varlen_func, +) +from concurrent.futures import ThreadPoolExecutor +import triton +import triton.language as tl +from triton.tools.tensor_descriptor import TensorDescriptor + + +# ===== Fused softcapped cross-entropy (Triton) — training-only path ===== +_FUSED_CE_LIBRARY = "pg035dfusedce" +_FUSED_CE_BLOCK_SIZE = 1024 +_FUSED_CE_NUM_WARPS = 4 + + +@triton.jit +def _softcapped_ce_fwd_kernel( + logits_ptr, losses_ptr, lse_ptr, targets_ptr, + stride_logits_n, stride_logits_v, + n_rows, n_cols, softcap, + block_size: tl.constexpr, +): + row_idx = tl.program_id(0).to(tl.int64) + logits_row_ptr = logits_ptr + row_idx * stride_logits_n + max_val = -float("inf") + sum_exp = 0.0 + A = 2.0 * softcap + inv_C = 2.0 / softcap + for off in range(0, n_cols, block_size): + cols = off + tl.arange(0, block_size) + mask = cols < n_cols + val = tl.load( + logits_row_ptr + cols * stride_logits_v, + mask=mask, other=-float("inf"), + ).to(tl.float32) + z = A * tl.sigmoid(val * inv_C) + z = tl.where(mask, z, -float("inf")) + curr_max = tl.max(z, axis=0) + new_max = tl.maximum(max_val, curr_max) + sum_exp = sum_exp * tl.exp(max_val - new_max) + tl.sum(tl.exp(z - new_max), axis=0) + max_val = new_max + lse = max_val + tl.log(sum_exp) + tl.store(lse_ptr + row_idx, lse) + target = tl.load(targets_ptr + row_idx).to(tl.int32) + target_val = tl.load(logits_row_ptr + target * stride_logits_v).to(tl.float32) + target_z = A * tl.sigmoid(target_val * inv_C) + tl.store(losses_ptr + row_idx, lse - target_z) + + +@triton.jit +def _softcapped_ce_bwd_kernel( + grad_logits_ptr, grad_losses_ptr, lse_ptr, logits_ptr, targets_ptr, + stride_logits_n, stride_logits_v, + stride_grad_n, stride_grad_v, + n_rows, n_cols, softcap, + block_size: tl.constexpr, +): + row_idx = tl.program_id(0).to(tl.int64) + logits_row_ptr = logits_ptr + row_idx * stride_logits_n + grad_row_ptr = grad_logits_ptr + row_idx * stride_grad_n + lse = tl.load(lse_ptr + row_idx) + grad_loss = tl.load(grad_losses_ptr + row_idx).to(tl.float32) + target = tl.load(targets_ptr + row_idx).to(tl.int32) + A = 2.0 * softcap + inv_C = 2.0 / softcap + dz_dx_scale = A * inv_C + for off in range(0, n_cols, block_size): + cols = off + tl.arange(0, block_size) + mask = cols < n_cols + val = tl.load( + logits_row_ptr + cols * stride_logits_v, + mask=mask, other=0.0, + ).to(tl.float32) + sigmoid_u = tl.sigmoid(val * inv_C) + z = A * sigmoid_u + probs = tl.exp(z - lse) + grad_z = grad_loss * (probs - tl.where(cols == target, 1.0, 0.0)) + grad_x = grad_z * (dz_dx_scale * sigmoid_u * (1.0 - sigmoid_u)) + tl.store(grad_row_ptr + cols * stride_grad_v, grad_x, mask=mask) + + +def _validate_softcapped_ce_inputs( + logits: Tensor, targets: Tensor, softcap: float, +) -> tuple[Tensor, Tensor]: + if logits.ndim != 2: + raise ValueError(f"Expected logits.ndim=2, got {logits.ndim}") + if targets.ndim != 1: + raise ValueError(f"Expected targets.ndim=1, got {targets.ndim}") + if logits.shape[0] != targets.shape[0]: + raise ValueError( + f"Expected matching rows, got logits={tuple(logits.shape)} targets={tuple(targets.shape)}" + ) + if not logits.is_cuda or not targets.is_cuda: + raise ValueError("softcapped_cross_entropy requires CUDA tensors") + if softcap <= 0.0: + raise ValueError(f"softcap must be positive, got {softcap}") + if logits.dtype not in (torch.float16, torch.bfloat16, torch.float32): + raise ValueError(f"Unsupported logits dtype: {logits.dtype}") + logits = logits.contiguous() + targets = targets.contiguous() + if targets.dtype != torch.int64: + targets = targets.to(dtype=torch.int64) + return logits, targets + + +@torch.library.custom_op(f"{_FUSED_CE_LIBRARY}::softcapped_ce", mutates_args=()) +def softcapped_ce_op(logits: Tensor, targets: Tensor, softcap: float) -> tuple[Tensor, Tensor]: + logits, targets = _validate_softcapped_ce_inputs(logits, targets, float(softcap)) + n_rows, n_cols = logits.shape + losses = torch.empty((n_rows,), device=logits.device, dtype=torch.float32) + lse = torch.empty((n_rows,), device=logits.device, dtype=torch.float32) + _softcapped_ce_fwd_kernel[(n_rows,)]( + logits, losses, lse, targets, + logits.stride(0), logits.stride(1), + n_rows, n_cols, float(softcap), + block_size=_FUSED_CE_BLOCK_SIZE, num_warps=_FUSED_CE_NUM_WARPS, + ) + return losses, lse + + +@softcapped_ce_op.register_fake +def _(logits: Tensor, targets: Tensor, softcap: float): + if logits.ndim != 2 or targets.ndim != 1: + raise ValueError("softcapped_ce fake impl expects 2D logits and 1D targets") + if logits.shape[0] != targets.shape[0]: + raise ValueError( + f"Expected matching rows, got logits={tuple(logits.shape)} targets={tuple(targets.shape)}" + ) + n_rows = logits.shape[0] + return ( + logits.new_empty((n_rows,), dtype=torch.float32), + logits.new_empty((n_rows,), dtype=torch.float32), + ) + + +@torch.library.custom_op(f"{_FUSED_CE_LIBRARY}::softcapped_ce_backward", mutates_args=()) +def softcapped_ce_backward_op( + logits: Tensor, targets: Tensor, lse: Tensor, grad_losses: Tensor, softcap: float, +) -> Tensor: + logits, targets = _validate_softcapped_ce_inputs(logits, targets, float(softcap)) + lse = lse.contiguous() + grad_losses = grad_losses.contiguous().to(dtype=torch.float32) + if lse.ndim != 1 or grad_losses.ndim != 1: + raise ValueError("Expected 1D lse and grad_losses") + if lse.shape[0] != logits.shape[0] or grad_losses.shape[0] != logits.shape[0]: + raise ValueError( + f"Expected row-aligned lse/grad_losses, got logits={tuple(logits.shape)} " + f"lse={tuple(lse.shape)} grad_losses={tuple(grad_losses.shape)}" + ) + grad_logits = torch.empty_like(logits) + n_rows, n_cols = logits.shape + _softcapped_ce_bwd_kernel[(n_rows,)]( + grad_logits, grad_losses, lse, logits, targets, + logits.stride(0), logits.stride(1), + grad_logits.stride(0), grad_logits.stride(1), + n_rows, n_cols, float(softcap), + block_size=_FUSED_CE_BLOCK_SIZE, num_warps=_FUSED_CE_NUM_WARPS, + ) + return grad_logits + + +@softcapped_ce_backward_op.register_fake +def _(logits: Tensor, targets: Tensor, lse: Tensor, grad_losses: Tensor, softcap: float): + if logits.ndim != 2 or targets.ndim != 1 or lse.ndim != 1 or grad_losses.ndim != 1: + raise ValueError("softcapped_ce_backward fake impl expects 2D logits and 1D row tensors") + if ( + logits.shape[0] != targets.shape[0] + or logits.shape[0] != lse.shape[0] + or logits.shape[0] != grad_losses.shape[0] + ): + raise ValueError("softcapped_ce_backward fake impl expects row-aligned tensors") + return logits.new_empty(logits.shape) + + +def _softcapped_ce_setup_context( + ctx: torch.autograd.function.FunctionCtx, inputs, output, +) -> None: + logits, targets, softcap = inputs + _losses, lse = output + ctx.save_for_backward(logits, targets, lse) + ctx.softcap = float(softcap) + + +def _softcapped_ce_backward( + ctx: torch.autograd.function.FunctionCtx, grad_losses: Tensor, grad_lse: "Tensor | None", +): + del grad_lse + logits, targets, lse = ctx.saved_tensors + grad_logits = torch.ops.pg035dfusedce.softcapped_ce_backward( + logits, targets, lse, grad_losses, ctx.softcap + ) + return grad_logits, None, None + + +softcapped_ce_op.register_autograd( + _softcapped_ce_backward, setup_context=_softcapped_ce_setup_context, +) + + +def softcapped_cross_entropy( + logits: Tensor, targets: Tensor, softcap: float, reduction: str = "mean", +) -> Tensor: + losses, _lse = torch.ops.pg035dfusedce.softcapped_ce( + logits, targets, float(softcap) + ) + if reduction == "none": + return losses + if reduction == "sum": + return losses.sum() + if reduction == "mean": + return losses.mean() + raise ValueError(f"Unsupported reduction={reduction!r}") + + +class Hyperparameters: + data_dir = os.environ.get("DATA_DIR", "./data/") + seed = int(os.environ.get("SEED", 1337)) + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_frac = float(os.environ.get("WARMDOWN_FRAC", 0.75)) + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786432)) + fused_ce_enabled = bool(int(os.environ.get("FUSED_CE_ENABLED", "1"))) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 6e2)) + val_batch_tokens = int(os.environ.get("VAL_BATCH_TOKENS", 524288)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 8192)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 11)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 4.0)) + skip_gates_enabled = bool(int(os.environ.get("SKIP_GATES_ENABLED", "1"))) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 3e1)) + rope_base = float(os.environ.get("ROPE_BASE", 1e4)) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + rope_train_seq_len = int(os.environ.get("ROPE_TRAIN_SEQ_LEN", 2048)) + rope_yarn = bool(int(os.environ.get("ROPE_YARN", "0"))) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 5.0)) + num_loops = int(os.environ.get("NUM_LOOPS", 2)) + loop_start = int(os.environ.get("LOOP_START", 3)) + loop_end = int(os.environ.get("LOOP_END", 5)) + enable_looping_at = float(os.environ.get("ENABLE_LOOPING_AT", 0.35)) + loop_depth_upgrade_at = float(os.environ.get("LOOP_DEPTH_UPGRADE_AT", 0.0)) + parallel_start_layer = int(os.environ.get("PARALLEL_START_LAYER", 8)) + parallel_final_lane = os.environ.get("PARALLEL_FINAL_LANE", "mean") + min_lr = float(os.environ.get("MIN_LR", 0.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.03)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.026)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.02)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.97)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float( + os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92) + ) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + muon_row_normalize = bool(int(os.environ.get("MUON_ROW_NORMALIZE", "1"))) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-08)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + adam_wd = float(os.environ.get("ADAM_WD", 0.02)) + muon_wd = float(os.environ.get("MUON_WD", 0.095)) + embed_wd = float(os.environ.get("EMBED_WD", 0.085)) + ema_decay = float(os.environ.get("EMA_DECAY", 0.9965)) + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "1"))) + ttt_lora_rank = int(os.environ.get("TTT_LORA_RANK", 96)) + ttt_lora_alpha = int(os.environ.get("TTT_LORA_ALPHA", 96)) + ttt_lora_lr = float(os.environ.get("TTT_LORA_LR", 0.0001)) + ttt_chunk_size = int(os.environ.get("TTT_CHUNK_SIZE", 48)) + ttt_eval_seq_len = int(os.environ.get("TTT_EVAL_SEQ_LEN", 2048)) + ttt_batch_size = int(os.environ.get("TTT_BATCH_SIZE", 64)) + ttt_grad_steps = int(os.environ.get("TTT_GRAD_STEPS", 1)) + ttt_weight_decay = float(os.environ.get("TTT_WEIGHT_DECAY", 0.5)) + ttt_beta1 = float(os.environ.get("TTT_BETA1", 0)) + ttt_beta2 = float(os.environ.get("TTT_BETA2", 0.999)) + ttt_k_lora = bool(int(os.environ.get("TTT_K_LORA", "1"))) + ttt_mlp_lora = bool(int(os.environ.get("TTT_MLP_LORA", "1"))) + ttt_o_lora = bool(int(os.environ.get("TTT_O_LORA", "1"))) + ttt_optimizer = os.environ.get("TTT_OPTIMIZER", "adam") + ttt_eval_batches = os.environ.get("TTT_EVAL_BATCHES", "") + val_doc_fraction = float(os.environ.get("VAL_DOC_FRACTION", 1.0)) + compressor = os.environ.get("COMPRESSOR", "brotli") + gptq_calibration_batches = int(os.environ.get("GPTQ_CALIBRATION_BATCHES", 16)) + gptq_reserve_seconds = float(os.environ.get("GPTQ_RESERVE_SECONDS", 4.0)) + phased_ttt_prefix_docs = int(os.environ.get("PHASED_TTT_PREFIX_DOCS", 2000)) + phased_ttt_num_phases = int(os.environ.get("PHASED_TTT_NUM_PHASES", 1)) + global_ttt_lr = float(os.environ.get("GLOBAL_TTT_LR", 0.001)) + global_ttt_momentum = float(os.environ.get("GLOBAL_TTT_MOMENTUM", 0.9)) + global_ttt_epochs = int(os.environ.get("GLOBAL_TTT_EPOCHS", 1)) + global_ttt_chunk_tokens = int(os.environ.get("GLOBAL_TTT_CHUNK_TOKENS", 32768)) + global_ttt_batch_seqs = int(os.environ.get("GLOBAL_TTT_BATCH_SEQS", 32)) + global_ttt_warmup_start_lr = float(os.environ.get("GLOBAL_TTT_WARMUP_START_LR", 0.0)) + global_ttt_warmup_chunks = int(os.environ.get("GLOBAL_TTT_WARMUP_CHUNKS", 0)) + global_ttt_grad_clip = float(os.environ.get("GLOBAL_TTT_GRAD_CLIP", 1.0)) + global_ttt_respect_doc_boundaries = bool(int(os.environ.get("GLOBAL_TTT_RESPECT_DOC_BOUNDARIES", "1"))) + matrix_bits = int(os.environ.get("MATRIX_BITS", 6)) + embed_bits = int(os.environ.get("EMBED_BITS", 8)) + matrix_clip_sigmas = float(os.environ.get("MATRIX_CLIP_SIGMAS", 12.85)) + embed_clip_sigmas = float(os.environ.get("EMBED_CLIP_SIGMAS", 2e1)) + mlp_clip_sigmas = float(os.environ.get("MLP_CLIP_SIGMAS", 10.0)) + attn_clip_sigmas = float(os.environ.get("ATTN_CLIP_SIGMAS", 13.0)) + # AttnOutGate (per-head multiplicative output gate, PR #1667 MarioPaerle). + # Zero-init weight: 2*sigmoid(0)=1 -> transparent at start. Source defaults to + # block input x ('proj'); 'q' uses raw Q projection output. + attn_out_gate_enabled = bool(int(os.environ.get("ATTN_OUT_GATE_ENABLED", "0"))) + attn_out_gate_src = os.environ.get("ATTN_OUT_GATE_SRC", "proj") + # SmearGate (input-dependent forward-1 token smear, modded-nanogpt @classiclarryd + # via PR #1667). x_t <- x_t + lam * sigmoid(W*x_t[:gate_window]) * x_{t-1}. + # lam=0 + W=0 -> transparent at init. + smear_gate_enabled = bool(int(os.environ.get("SMEAR_GATE_ENABLED", "0"))) + # Window: first GATE_WINDOW dims of the source feed the gate projection. + gate_window = int(os.environ.get("GATE_WINDOW", 12)) + # Spec 015 Recur-Alpha (port #1714 Anakintano). Learnable scalar per + # non-first recurrence pass per looped block. At init all zero -> pure + # passthrough on extra passes (equivalent to NUM_LOOPS=0 effective behavior). + # Model learns how much to commit per pass via gradient descent. + # Default off -> byte-identical to #1736 baseline. + recur_alpha_enabled = bool(int(os.environ.get("RECUR_ALPHA_ENABLED", "0"))) + # Diagnostics: compute/store pass-to-pass cosine similarity on block deltas. + # Answers "is cross-pass XSA the right follow-up?" informational side-channel. + recur_diag_p2p_cos = bool(int(os.environ.get("RECUR_DIAG_P2P_COS", "0"))) + # Gated Attention (Qwen, NeurIPS 2025 Best Paper, arXiv:2505.06708; + # qiuzh20/gated_attention). Per-head sigmoid gate on SDPA output, BEFORE + # out_proj. Gate input = full block input x (paper's headwise G1 variant + # driven from hidden_states). W_g shape (num_heads, dim), plain sigmoid. + # Near-zero init gives g~0.5 at step 0 (half attention output); per-block + # attn_scale (init 1.0) compensates during training. Name contains + # "attn_gate" so CONTROL_TENSOR_NAME_PATTERNS routes it to scalar AdamW. + gated_attn_enabled = bool(int(os.environ.get("GATED_ATTN_ENABLED", "0"))) + gated_attn_init_std = float(os.environ.get("GATED_ATTN_INIT_STD", 0.01)) + sparse_attn_gate_enabled = bool(int(os.environ.get("SPARSE_ATTN_GATE_ENABLED", "0"))) + sparse_attn_gate_init_std = float(os.environ.get("SPARSE_ATTN_GATE_INIT_STD", 0.0)) + sparse_attn_gate_scale = float(os.environ.get("SPARSE_ATTN_GATE_SCALE", 1.0)) + # Dedicated int8-per-row quantization for `attn_gate_w` tensors. These are + # small ((num_heads, dim) = (8, 512) = 4096 params) and bypass GPTQ via the + # numel<=65536 passthrough branch -> stored as fp16 (8 KB/layer, ~65 KB total + # compressed). int8-per-row cuts the raw tensor in half with negligible BPB + # impact: scales per head (8 values), symmetric quant over [-127, 127]. + # No Hessian needed (gate weights not in collect_hessians()). + gated_attn_quant_gate = bool(int(os.environ.get("GATED_ATTN_QUANT_GATE", "0"))) + # SpinQuant V1 port from PR #1695 (X-Abhishek-X). Disabled by default so this + # file still reproduces #1736 bit-identically when the env var is absent. + # When enabled, four Hadamard rotations are applied online in the forward + # pass (pre-QKV, pre-attn-proj, pre-fc, pre-mlp-proj) and correspondingly + # baked into the state_dict + Hessian before GPTQ. F.linear(x @ R, W @ R) + # == F.linear(x, W) exactly (float-invariant), but GPTQ sees the rotated + # basis where outliers are more evenly spread → lower quantization error. + # See research/specs/010-port-1695-online-rotation.md for the integration + # plan; installation happens in deserialize() and the 4 forward-pass hooks + # live in CausalSelfAttention.forward, MLP.forward, and the TTT mirrors. + spinquant_enabled = bool(int(os.environ.get("SPINQUANT_ENABLED", "0"))) + spinquant_seed = int(os.environ.get("SPINQUANT_SEED", "42")) + # Comma-separated subset of the 4 rotation sites to apply. Default is "all + # 4" so existing SPINQUANT_ENABLED=1 runs (spec 010) behave identically. + # Valid tag names: attn_in, attn_proj_in, mlp_in, mlp_proj_in. Spec 010b + # uses this to ablate which sites are responsible for the long/short-doc + # regime split observed in spec 010. + spinquant_sites = os.environ.get( + "SPINQUANT_SITES", "attn_in,attn_proj_in,mlp_in,mlp_proj_in" + ) + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + is_main_process = rank == 0 + grad_accum_steps = 8 // world_size + # CaseOps integration: optional override of dataset root + tokenizer path. + # When CASEOPS_ENABLED=1, the wrapper loads a per-token byte sidecar + # (fineweb_val_bytes_*.bin, identical shard layout to val_*.bin) and uses + # it as the canonical raw-byte budget for BPB accounting. The sidecar + # REPLACES the build_sentencepiece_luts byte-counting path entirely. + caseops_enabled = bool(int(os.environ.get("CASEOPS_ENABLED", "0"))) + _default_caseops_data = os.path.join( + data_dir, + "datasets", + "fineweb10B_sp8192_caseops", + "datasets", + "datasets", + "fineweb10B_sp8192_lossless_caps_caseops_v1_reserved", + ) + _default_caseops_tok = os.path.join( + data_dir, + "datasets", + "fineweb10B_sp8192_caseops", + "datasets", + "tokenizers", + "fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model", + ) + if caseops_enabled: + datasets_dir = os.environ.get("DATA_PATH", _default_caseops_data) + tokenizer_path = os.environ.get("TOKENIZER_PATH", _default_caseops_tok) + else: + datasets_dir = os.environ.get( + "DATA_PATH", + os.path.join(data_dir, "datasets", f"fineweb10B_sp{vocab_size}"), + ) + tokenizer_path = os.environ.get( + "TOKENIZER_PATH", + os.path.join(data_dir, "tokenizers", f"fineweb_{vocab_size}_bpe.model"), + ) + train_files = os.path.join(datasets_dir, "fineweb_train_*.bin") + val_files = os.path.join(datasets_dir, "fineweb_val_*.bin") + val_bytes_files = os.path.join(datasets_dir, "fineweb_val_bytes_*.bin") + artifact_dir = os.environ.get("ARTIFACT_DIR", "") + logfile = ( + os.path.join(artifact_dir, f"{run_id}.txt") + if artifact_dir + else f"logs/{run_id}.txt" + ) + model_path = ( + os.path.join(artifact_dir, "final_model.pt") + if artifact_dir + else "final_model.pt" + ) + quantized_model_path = ( + os.path.join(artifact_dir, "final_model.int6.ptz") + if artifact_dir + else "final_model.int6.ptz" + ) + + +_logger_hparams = None + + +def set_logging_hparams(h): + global _logger_hparams + _logger_hparams = h + + +def log(msg, console=True): + if _logger_hparams is None: + print(msg) + return + if _logger_hparams.is_main_process: + if console: + print(msg) + if _logger_hparams.logfile is not None: + with open(_logger_hparams.logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + + +class ValidationData: + def __init__(self, h, device): + self.sp = spm.SentencePieceProcessor(model_file=h.tokenizer_path) + if int(self.sp.vocab_size()) != h.vocab_size: + raise ValueError( + f"VOCAB_SIZE={h.vocab_size} does not match tokenizer vocab_size={int(self.sp.vocab_size())}" + ) + self.val_tokens = load_validation_tokens(h.val_files, h.eval_seq_len) + ( + self.base_bytes_lut, + self.has_leading_space_lut, + self.is_boundary_token_lut, + ) = build_sentencepiece_luts(self.sp, h.vocab_size, device) + # CaseOps: when enabled, load per-token byte sidecar and stash it as a + # CPU tensor aligned 1:1 with self.val_tokens. eval_val/eval_val_ttt + # branches use this as the canonical raw-byte budget per token. + self.caseops_enabled = bool(getattr(h, "caseops_enabled", False)) + self.val_bytes = None + if self.caseops_enabled: + self.val_bytes = load_validation_byte_sidecar( + h.val_bytes_files, h.eval_seq_len, self.val_tokens.numel() + ) + + +def build_sentencepiece_luts(sp, vocab_size, device): + sp_vocab_size = int(sp.vocab_size()) + assert ( + sp.piece_to_id("▁") != sp.unk_id() + ), "Tokenizer must have '▁' (space) as its own token for correct BPB byte counting" + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("▁"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + + +def load_validation_tokens(pattern, seq_len): + # Filter out CaseOps byte sidecar shards which share the val_*.bin glob. + files = [ + Path(p) + for p in sorted(glob.glob(pattern)) + if "_bytes_" not in Path(p).name + ] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = (tokens.numel() - 1) // seq_len * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] + + +def load_validation_byte_sidecar(pattern, seq_len, expected_len): + """Load CaseOps per-token byte sidecar(s). Same shard layout as token shards + (256 int32 header + uint16 array). Each entry = canonical raw-text byte + budget for that token in the corresponding val shard. Returns a CPU + int16 tensor sliced to match expected_len (i.e. val_tokens length).""" + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No byte sidecar files for pattern: {pattern}") + shards = [load_data_shard(file) for file in files] + # load_data_shard returns uint16 — that's exactly what the sidecar stores. + bytes_full = torch.cat(shards).contiguous() + if bytes_full.numel() < expected_len: + raise ValueError( + f"Byte sidecar too short: {bytes_full.numel()} < val_tokens {expected_len}" + ) + return bytes_full[:expected_len].to(torch.int32) + + +def load_data_shard(file): + header_bytes = 256 * np.dtype(" 0: + pos = start + while pos < end: + seg_starts.append(pos) + pos += max_doc_len + else: + seg_starts.append(start) + boundaries = seg_starts + [total_len] + padded_len = get_next_multiple_of_n(len(boundaries), bucket_size) + cu = torch.full((padded_len,), total_len, dtype=torch.int32, device=device) + cu[: len(boundaries)] = torch.tensor(boundaries, dtype=torch.int32, device=device) + seg_ends = seg_starts[1:] + [total_len] + max_seqlen = max(end - start for start, end in zip(seg_starts, seg_ends)) + return cu, max_seqlen + +class DocumentPackingLoader: + _shard_pool = ThreadPoolExecutor(1) + + def __init__(self, h, device, cu_bucket_size=64): + self.rank = h.rank + self.world_size = h.world_size + self.device = device + self.cu_bucket_size = cu_bucket_size + self.max_seq_len = h.train_seq_len + all_files = [Path(p) for p in sorted(glob.glob(h.train_files))] + if not all_files: + raise FileNotFoundError(f"No files found for pattern: {h.train_files}") + self.files = all_files + self.file_iter = iter(self.files) + self._init_shard(load_data_shard(next(self.file_iter))) + self._next_shard = self._submit_next_shard() + self._batch_pool = ThreadPoolExecutor(1) + self._next_batch = None + + def _init_shard(self, tokens): + global BOS_ID + self.tokens = tokens + self.shard_size = tokens.numel() + if BOS_ID is None: + BOS_ID = 1 + self.bos_idx = ( + (tokens == BOS_ID).nonzero(as_tuple=True)[0].to(torch.int64).cpu().numpy() + ) + if self.bos_idx.size == 0: + self.bos_idx = np.array([0], dtype=np.int64) + self.cursor = int(self.bos_idx[0]) + + def _submit_next_shard(self): + try: + path = next(self.file_iter) + return self._shard_pool.submit(load_data_shard, path) + except StopIteration: + return None + + def _advance_shard(self): + if self._next_shard is None: + self.file_iter = iter(self.files) + self._next_shard = self._shard_pool.submit( + load_data_shard, next(self.file_iter) + ) + self._init_shard(self._next_shard.result()) + self._next_shard = self._submit_next_shard() + + def _local_doc_starts(self, local_start, total_len): + lo = np.searchsorted(self.bos_idx, local_start, side="left") + hi = np.searchsorted(self.bos_idx, local_start + total_len, side="left") + return (self.bos_idx[lo:hi] - local_start).tolist() + + def _prepare_batch(self, num_tokens_local, max_seq_len): + per_rank_span = num_tokens_local + 1 + global_span = per_rank_span * self.world_size + while self.cursor + global_span > self.shard_size: + self._advance_shard() + local_start = self.cursor + self.rank * per_rank_span + buf = self.tokens[local_start : local_start + per_rank_span] + inputs = buf[:-1].to(dtype=torch.int64).pin_memory() + targets = buf[1:].to(dtype=torch.int64).pin_memory() + starts = self._local_doc_starts(local_start, inputs.numel()) + cu_seqlens, max_seqlen = _build_cu_seqlens( + starts, inputs.numel(), inputs.device, max_seq_len, self.cu_bucket_size + ) + cu_seqlens = cu_seqlens.pin_memory() + self.cursor += global_span + return inputs, targets, cu_seqlens, max_seqlen + + def next_batch(self, global_tokens, grad_accum_steps): + num_tokens_local = global_tokens // (self.world_size * grad_accum_steps) + if self._next_batch is not None: + inputs, targets, cu_seqlens, max_seqlen = self._next_batch.result() + else: + inputs, targets, cu_seqlens, max_seqlen = self._prepare_batch( + num_tokens_local, self.max_seq_len + ) + self._next_batch = self._batch_pool.submit( + self._prepare_batch, num_tokens_local, self.max_seq_len + ) + return ( + inputs[None].to(self.device, non_blocking=True), + targets[None].to(self.device, non_blocking=True), + cu_seqlens.to(self.device, non_blocking=True), + max_seqlen, + ) + + +class ShuffledSequenceLoader: + def __init__(self, h, device): + self.world_size = h.world_size + self.seq_len = h.train_seq_len + self.device = device + all_files = [Path(p) for p in sorted(glob.glob(h.train_files))] + if not all_files: + raise FileNotFoundError(f"No files found for pattern: {h.train_files}") + self.files = all_files[h.rank :: h.world_size] + self.rng = np.random.Generator(np.random.PCG64(h.rank)) + self.num_tokens = [_read_num_tokens(f) for f in self.files] + self.start_inds = [[] for _ in self.files] + for si in range(len(self.files)): + self._reset_shard(si) + + def _reset_shard(self, si): + max_phase = min( + self.seq_len - 1, max(0, self.num_tokens[si] - self.seq_len - 1) + ) + phase = int(self.rng.integers(max_phase + 1)) if max_phase > 0 else 0 + num_sequences = (self.num_tokens[si] - 1 - phase) // self.seq_len + sequence_order = self.rng.permutation(num_sequences) + self.start_inds[si] = (phase + sequence_order * self.seq_len).tolist() + + def next_batch(self, global_tokens, grad_accum_steps): + device_tokens = global_tokens // (self.world_size * grad_accum_steps) + device_batch_size = device_tokens // self.seq_len + remaining = np.array([len(s) for s in self.start_inds], dtype=np.float64) + x = torch.empty((device_batch_size, self.seq_len), dtype=torch.int64) + y = torch.empty((device_batch_size, self.seq_len), dtype=torch.int64) + for bi in range(device_batch_size): + total = remaining.sum() + if total <= 0: + for si in range(len(self.files)): + self._reset_shard(si) + remaining = np.array( + [len(s) for s in self.start_inds], dtype=np.float64 + ) + total = remaining.sum() + probs = remaining / total + si = int(self.rng.choice(len(self.files), p=probs)) + start_ind = self.start_inds[si].pop() + remaining[si] -= 1 + mm = _get_shard_memmap(self.files[si]) + window = torch.as_tensor( + np.array(mm[start_ind : start_ind + self.seq_len + 1], dtype=np.int64) + ) + x[bi] = window[:-1] + y[bi] = window[1:] + return x.to(self.device, non_blocking=True), y.to( + self.device, non_blocking=True + ) + + +class RMSNorm(nn.Module): + def __init__(self, eps=None): + super().__init__() + self.eps = eps + + def forward(self, x): + return F.rms_norm(x, (x.size(-1),), eps=self.eps) + + +class CastedLinear(nn.Linear): + # SpinQuant V1: class-level flag gates the forward-pass rotation hooks in + # CausalSelfAttention.forward, MLP.forward, _block_with_lora, and + # _parallel_block_with_lora. OFF during training (Dynamo constant-folds the + # branch away). Flipped to True by deserialize() after install_spinquant_ + # rotations() registers the R buffers on every attn/mlp module. + _sq_active: bool = False + + def forward(self, x): + w = self.weight.to(x.dtype) + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) + + +@triton.jit +def linear_leaky_relu_square_kernel( + a_desc, + b_desc, + c_desc, + aux_desc, + M, + N, + K, + BLOCK_SIZE_M: tl.constexpr, + BLOCK_SIZE_N: tl.constexpr, + BLOCK_SIZE_K: tl.constexpr, + NUM_SMS: tl.constexpr, + FORWARD: tl.constexpr, +): + dtype = tl.bfloat16 + start_pid = tl.program_id(axis=0) + num_pid_m = tl.cdiv(M, BLOCK_SIZE_M) + num_pid_n = tl.cdiv(N, BLOCK_SIZE_N) + k_tiles = tl.cdiv(K, BLOCK_SIZE_K) + num_tiles = num_pid_m * num_pid_n + tile_id_c = start_pid - NUM_SMS + for tile_id in tl.range(start_pid, num_tiles, NUM_SMS, flatten=True): + pid_m = tile_id // num_pid_n + pid_n = tile_id % num_pid_n + offs_am = pid_m * BLOCK_SIZE_M + offs_bn = pid_n * BLOCK_SIZE_N + accumulator = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32) + for ki in range(k_tiles): + offs_k = ki * BLOCK_SIZE_K + a = a_desc.load([offs_am, offs_k]) + b = b_desc.load([offs_bn, offs_k]) + accumulator = tl.dot(a, b.T, accumulator) + tile_id_c += NUM_SMS + offs_am_c = offs_am + offs_bn_c = offs_bn + acc = tl.reshape(accumulator, (BLOCK_SIZE_M, 2, BLOCK_SIZE_N // 2)) + acc = tl.permute(acc, (0, 2, 1)) + acc0, acc1 = tl.split(acc) + c0 = acc0.to(dtype) + c1 = acc1.to(dtype) + if not FORWARD: + pre0 = aux_desc.load([offs_am_c, offs_bn_c]) + pre1 = aux_desc.load([offs_am_c, offs_bn_c + BLOCK_SIZE_N // 2]) + c0 = c0 * tl.where(pre0 > 0, 2.0 * pre0, 0.5 * pre0) + c1 = c1 * tl.where(pre1 > 0, 2.0 * pre1, 0.5 * pre1) + c_desc.store([offs_am_c, offs_bn_c], c0) + c_desc.store([offs_am_c, offs_bn_c + BLOCK_SIZE_N // 2], c1) + if FORWARD: + aux0 = tl.where(c0 > 0, c0, 0.5 * c0) + aux1 = tl.where(c1 > 0, c1, 0.5 * c1) + aux_desc.store([offs_am_c, offs_bn_c], aux0 * aux0) + aux_desc.store([offs_am_c, offs_bn_c + BLOCK_SIZE_N // 2], aux1 * aux1) + + +def linear_leaky_relu_square(a, b, aux=None): + M, K = a.shape + N, K2 = b.shape + assert K == K2 + c = torch.empty((M, N), device=a.device, dtype=a.dtype) + forward = aux is None + if aux is None: + aux = torch.empty((M, N), device=a.device, dtype=a.dtype) + num_sms = torch.cuda.get_device_properties(a.device).multi_processor_count + BLOCK_SIZE_M, BLOCK_SIZE_N, BLOCK_SIZE_K = 128, 256, 64 + num_stages = 4 if forward else 3 + a_desc = TensorDescriptor.from_tensor(a, [BLOCK_SIZE_M, BLOCK_SIZE_K]) + b_desc = TensorDescriptor.from_tensor(b, [BLOCK_SIZE_N, BLOCK_SIZE_K]) + c_desc = TensorDescriptor.from_tensor(c, [BLOCK_SIZE_M, BLOCK_SIZE_N // 2]) + aux_desc = TensorDescriptor.from_tensor(aux, [BLOCK_SIZE_M, BLOCK_SIZE_N // 2]) + grid = lambda _meta: ( + min(num_sms, triton.cdiv(M, BLOCK_SIZE_M) * triton.cdiv(N, BLOCK_SIZE_N)), + ) + linear_leaky_relu_square_kernel[grid]( + a_desc, + b_desc, + c_desc, + aux_desc, + M, + N, + K, + BLOCK_SIZE_M=BLOCK_SIZE_M, + BLOCK_SIZE_N=BLOCK_SIZE_N, + BLOCK_SIZE_K=BLOCK_SIZE_K, + NUM_SMS=num_sms, + FORWARD=forward, + num_stages=num_stages, + num_warps=8, + ) + if forward: + return c, aux + return c + + +class FusedLinearLeakyReLUSquareFunction(torch.autograd.Function): + @staticmethod + def forward(ctx, x, w1, w2): + x_flat = x.reshape(-1, x.shape[-1]) + pre, post = linear_leaky_relu_square(x_flat, w1) + out = F.linear(post, w2) + ctx.save_for_backward(x, w1, w2, pre, post) + return out.view(*x.shape[:-1], out.shape[-1]) + + @staticmethod + def backward(ctx, grad_output): + x, w1, w2, pre, post = ctx.saved_tensors + x_flat = x.reshape(-1, x.shape[-1]) + grad_output_flat = grad_output.reshape(-1, grad_output.shape[-1]) + dw2 = grad_output_flat.T @ post + dpre = linear_leaky_relu_square(grad_output_flat, w2.T.contiguous(), aux=pre) + dw1 = dpre.T @ x_flat + dx = dpre @ w1 + return dx.view_as(x), dw1, dw2 + + +FusedLeakyReLUSquareMLP = FusedLinearLeakyReLUSquareFunction.apply + + +class Rotary(nn.Module): + def __init__(self, dim, base=1e4, train_seq_len=1024, rope_dims=0, yarn=True): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.yarn = yarn + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / base ** ( + torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims + ) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached = None + self._sin_cached = None + + def forward(self, seq_len, device, dtype): + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached < seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if self.yarn and seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * scale ** (rd / (rd - 2)) + inv_freq = 1.0 / new_base ** ( + torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd + ) + else: + inv_freq = self.inv_freq.float().to(device) + t = torch.arange(seq_len, device=device, dtype=torch.float32) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached[:, :seq_len].to(dtype=dtype), self._sin_cached[:, :seq_len].to(dtype=dtype) + + +def apply_rotary_emb(x, cos, sin, rope_dims=0): + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * -sin + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * -sin + x2 * cos), dim=-1) + + +class CausalSelfAttention(nn.Module): + def __init__( + self, dim, num_heads, num_kv_heads, rope_base, qk_gain_init, train_seq_len, yarn=True, + attn_out_gate=False, attn_out_gate_src="proj", gate_window=12, + gated_attn=False, gated_attn_init_std=0.01, + sparse_attn_gate=False, sparse_attn_gate_init_std=0.0, sparse_attn_gate_scale=1.0, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + if int(attn_out_gate) + int(gated_attn) + int(sparse_attn_gate) > 1: + raise ValueError( + "attn_out_gate, gated_attn, and sparse_attn_gate are mutually exclusive" + ) + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + self.q_gain = nn.Parameter( + torch.full((num_heads,), qk_gain_init, dtype=torch.float32) + ) + self.rope_dims = 0 + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=train_seq_len, yarn=yarn) + self.use_xsa = False + # AttnOutGate (PR #1667 MarioPaerle): per-head multiplicative gate on attention + # output. CastedLinear so restore_fp32_params casts back to fp32 for GPTQ. + # _zero_init -> 2*sigmoid(0)=1 -> transparent at init. + self.attn_out_gate = attn_out_gate + self.attn_out_gate_src = attn_out_gate_src + self.gate_window = gate_window + if attn_out_gate: + self.attn_gate_proj = CastedLinear(gate_window, num_heads, bias=False) + self.attn_gate_proj._zero_init = True + # Gated Attention (arXiv:2505.06708, Qwen, NeurIPS 2025). Per-head sigmoid + # gate on SDPA output, BEFORE out_proj. Gate projection W_g: (num_heads, dim). + # Name "attn_gate_w" contains "attn_gate" substring so it matches + # CONTROL_TENSOR_NAME_PATTERNS and routes to the scalar AdamW group. + # fp32 Parameter -> restore_fp32_params path covers it via the ndim<2 OR + # name-pattern check (name matches "attn_gate"). Cast to x.dtype on use. + self.gated_attn = gated_attn + if gated_attn: + W = torch.empty(num_heads, dim, dtype=torch.float32) + nn.init.normal_(W, mean=0.0, std=gated_attn_init_std) + self.attn_gate_w = nn.Parameter(W) + self.sparse_attn_gate = sparse_attn_gate + self.sparse_attn_gate_scale = sparse_attn_gate_scale + if sparse_attn_gate: + W = torch.empty(num_heads, gate_window, dtype=torch.float32) + if sparse_attn_gate_init_std > 0: + nn.init.normal_(W, mean=0.0, std=sparse_attn_gate_init_std) + else: + nn.init.zeros_(W) + self.attn_gate_w = nn.Parameter(W) + + def _xsa_efficient(self, y, v): + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) + vn = F.normalize(v, dim=-1).unsqueeze(-2) + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + + def forward(self, x, q_w, k_w, v_w, out_w, cu_seqlens=None, max_seqlen=0): + bsz, seqlen, dim = x.shape + # SpinQuant V1: rotate residual → Q/K/V input. Branch dies at Dynamo + # compile when _sq_active=False (training). Math: F.linear(x @ R, W @ R) + # == F.linear(x, W) exactly; the weight rotation lives in state_dict + # courtesy of _spinquant_rotate_sd_and_H() before GPTQ. + if CastedLinear._sq_active and hasattr(self, "_sq_R_attn_in"): + x_qkv = x @ self._sq_R_attn_in.to(x.dtype) + else: + x_qkv = x + # q_raw kept around as a tap point for attn_out_gate_src='q' (post-projection, + # pre-reshape, pre-RoPE). + q_raw = F.linear(x_qkv, q_w.to(x.dtype)) + q = q_raw.reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x_qkv, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x_qkv, v_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = self.rotary(seqlen, x.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, self.rope_dims) + k = apply_rotary_emb(k, cos, sin, self.rope_dims) + q = q * self.q_gain.to(dtype=q.dtype)[None, None, :, None] + if cu_seqlens is not None: + y = flash_attn_varlen_func( + q[0], + k[0], + v[0], + cu_seqlens_q=cu_seqlens, + cu_seqlens_k=cu_seqlens, + max_seqlen_q=max_seqlen, + max_seqlen_k=max_seqlen, + causal=True, + window_size=(-1, -1), + )[None] + else: + y = flash_attn_3_func(q, k, v, causal=True) + if self.use_xsa: + y = self._xsa_efficient(y, v) + # AttnOutGate inlined (PR #1667). Inline + .contiguous() barrier so torch.compile + # fullgraph=True is happy (this avoids the @torch.compiler.disable trap that + # crashed gates v3). Per-head gate on (B,T,H,D) tensor: g shape [B,T,H], broadcast + # over D via [..., None]. zero-init weight -> 2*sigmoid(0)=1 -> transparent. + if self.attn_out_gate: + gate_src = q_raw if self.attn_out_gate_src == "q" else x + gate_in = gate_src[..., : self.gate_window].contiguous() + g = 2.0 * torch.sigmoid(self.attn_gate_proj(gate_in)) + y = y * g[..., None] + # Gated Attention (arXiv:2505.06708 G1). Inline + .contiguous() barrier so + # torch.compile fullgraph=True is happy. Per-head gate on (B,T,H,D): g shape + # [B,T,H], broadcast over D via [..., None]. Paper: g = sigmoid(x @ W_g.T) + # where W_g: (H, dim). .to(x.dtype) on fp32 param before broadcast with bf16. + if self.gated_attn: + x_c = x.contiguous() + g = torch.sigmoid(F.linear(x_c, self.attn_gate_w.to(x.dtype))) + y = y * g[..., None] + if self.sparse_attn_gate: + gate_in = x[..., : self.gate_window].contiguous() + g = torch.sigmoid( + self.sparse_attn_gate_scale + * F.linear(gate_in, self.attn_gate_w.to(x.dtype)) + ) + y = y * g[..., None] + y = y.reshape(bsz, seqlen, dim) + self._last_proj_input = y.detach() if getattr(self, "_calib", False) else None + # SpinQuant V1: rotate attention output → proj input, matched by the + # input-col rotation of attn.proj.weight in state_dict. + if CastedLinear._sq_active and hasattr(self, "_sq_R_attn_proj_in"): + y = y @ self._sq_R_attn_proj_in.to(x.dtype) + return F.linear(y, out_w.to(x.dtype)) + + +class MLP(nn.Module): + def __init__(self, dim, mlp_mult): + super().__init__() + self.use_fused = True + + def forward(self, x, up_w, down_w): + # SpinQuant V1 forward rotations. Branches die at compile when _sq_active=False. + sq = CastedLinear._sq_active and hasattr(self, "_sq_R_mlp_in") + if sq: + x = x @ self._sq_R_mlp_in.to(x.dtype) + # Fused kernel cannot express mid-hidden rotation, so disable it when SQ is on. + # SQ is only active post-deserialize (eval/TTT) where fused is already typically + # off; this guard covers the TTT-train case if it ever arises. + if self.training and self.use_fused and not sq: + return FusedLeakyReLUSquareMLP(x, up_w.to(x.dtype), down_w.to(x.dtype)) + hidden = F.leaky_relu(F.linear(x, up_w.to(x.dtype)), negative_slope=0.5).square() + # Capture BEFORE the mlp_proj_in rotation so the Hessian stays on unrotated hidden. + self._last_down_input = hidden.detach() if getattr(self, "_calib", False) else None + if sq and hasattr(self, "_sq_R_mlp_proj_in"): + hidden = hidden @ self._sq_R_mlp_proj_in.to(x.dtype) + return F.linear(hidden, down_w.to(x.dtype)) + + +class Block(nn.Module): + def __init__( + self, + dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + train_seq_len, + layer_idx=0, + ln_scale=False, + yarn=True, + attn_out_gate=False, + attn_out_gate_src="proj", + gate_window=12, + gated_attn=False, + gated_attn_init_std=0.01, + sparse_attn_gate=False, + sparse_attn_gate_init_std=0.0, + sparse_attn_gate_scale=1.0, + ): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention( + dim, num_heads, num_kv_heads, rope_base, qk_gain_init, train_seq_len, yarn=yarn, + attn_out_gate=attn_out_gate, attn_out_gate_src=attn_out_gate_src, gate_window=gate_window, + gated_attn=gated_attn, gated_attn_init_std=gated_attn_init_std, + sparse_attn_gate=sparse_attn_gate, + sparse_attn_gate_init_std=sparse_attn_gate_init_std, + sparse_attn_gate_scale=sparse_attn_gate_scale, + ) + self.mlp = MLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter( + torch.stack((torch.ones(dim), torch.zeros(dim))).float() + ) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + + def forward(self, x, x0, q_w, k_w, v_w, out_w, up_w, down_w, cu_seqlens=None, max_seqlen=0): + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn( + self.attn_norm(x_in) * self.ln_scale_factor, + q_w, k_w, v_w, out_w, + cu_seqlens=cu_seqlens, + max_seqlen=max_seqlen, + ) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[ + None, None, : + ] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + return x_out + +class GPT(nn.Module): + def __init__(self, h): + super().__init__() + if h.logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {h.logit_softcap}") + self.tie_embeddings = h.tie_embeddings + self.tied_embed_init_std = h.tied_embed_init_std + self.logit_softcap = h.logit_softcap + self.fused_ce_enabled = bool(h.fused_ce_enabled) + self.tok_emb = nn.Embedding(h.vocab_size, h.model_dim) + self.num_layers = h.num_layers + head_dim = h.model_dim // h.num_heads + kv_dim = h.num_kv_heads * head_dim + hidden_dim = int(h.mlp_mult * h.model_dim) + self.qo_bank = nn.Parameter(torch.empty(2 * h.num_layers, h.model_dim, h.model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * h.num_layers, kv_dim, h.model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(h.num_layers, hidden_dim, h.model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(h.num_layers, h.model_dim, hidden_dim)) + self.num_encoder_layers = h.num_layers // 2 + self.num_decoder_layers = h.num_layers - self.num_encoder_layers + self.blocks = nn.ModuleList( + [ + Block( + h.model_dim, + h.num_heads, + h.num_kv_heads, + h.mlp_mult, + h.rope_base, + h.qk_gain_init, + h.train_seq_len, + layer_idx=i, + ln_scale=h.ln_scale, + yarn=h.rope_yarn, + attn_out_gate=h.attn_out_gate_enabled, + attn_out_gate_src=h.attn_out_gate_src, + gate_window=h.gate_window, + gated_attn=h.gated_attn_enabled, + gated_attn_init_std=h.gated_attn_init_std, + sparse_attn_gate=h.sparse_attn_gate_enabled, + sparse_attn_gate_init_std=h.sparse_attn_gate_init_std, + sparse_attn_gate_scale=h.sparse_attn_gate_scale, + ) + for i in range(h.num_layers) + ] + ) + if h.rope_dims > 0: + head_dim = h.model_dim // h.num_heads + for block in self.blocks: + block.attn.rope_dims = h.rope_dims + block.attn.rotary = Rotary( + head_dim, + base=h.rope_base, + train_seq_len=h.train_seq_len, + rope_dims=h.rope_dims, + yarn=h.rope_yarn, + ) + self.final_norm = RMSNorm() + self.lm_head = ( + None + if h.tie_embeddings + else CastedLinear(h.model_dim, h.vocab_size, bias=False) + ) + if self.lm_head is not None: + self.lm_head._zero_init = True + if h.xsa_last_n > 0: + for i in range(max(0, h.num_layers - h.xsa_last_n), h.num_layers): + self.blocks[i].attn.use_xsa = True + self.looping_active = False + self._num_loops = h.num_loops + if h.num_loops > 0: + loop_seg = list(range(h.loop_start, h.loop_end + 1)) + all_indices = list(range(h.loop_start)) + for _ in range(h.num_loops + 1): + all_indices.extend(loop_seg) + all_indices.extend(range(h.loop_end + 1, h.num_layers)) + num_enc = len(all_indices) // 2 + self.encoder_indices = all_indices[:num_enc] + self.decoder_indices = all_indices[num_enc:] + # depth curriculum: precompute intermediate (num_loops-1) indices for phase 2 + if h.loop_depth_upgrade_at > 0 and h.num_loops >= 2: + all_int = list(range(h.loop_start)) + for _ in range(h.num_loops): # num_loops-1+1 passes = num_loops passes + all_int.extend(loop_seg) + all_int.extend(range(h.loop_end + 1, h.num_layers)) + num_enc_int = len(all_int) // 2 + self._enc_idx_intermediate = all_int[:num_enc_int] + self._dec_idx_intermediate = all_int[num_enc_int:] + self.looping_depth = h.num_loops - 1 # start at intermediate after activation + else: + self.looping_depth = h.num_loops + else: + self.encoder_indices = list(range(self.num_encoder_layers)) + self.decoder_indices = list(range(self.num_encoder_layers, h.num_layers)) + self.looping_depth = 0 + self.num_skip_weights = min( + len(self.encoder_indices), len(self.decoder_indices) + ) + self.skip_weights = nn.Parameter( + torch.ones(self.num_skip_weights, h.model_dim, dtype=torch.float32) + ) + self.skip_gates = ( + nn.Parameter( + torch.zeros(self.num_skip_weights, h.model_dim, dtype=torch.float32) + ) + if h.skip_gates_enabled + else None + ) + self.parallel_start_layer = h.parallel_start_layer + self.parallel_final_lane = h.parallel_final_lane.lower() + self.parallel_post_lambdas = nn.Parameter( + torch.ones(h.num_layers, 2, 2, dtype=torch.float32) + ) + self.parallel_resid_lambdas = nn.Parameter( + torch.full((h.num_layers, 2), 1.1, dtype=torch.float32) + ) + # SmearGate (PR #1667 / modded-nanogpt @classiclarryd): + # x_t <- x_t + lam * sigmoid(W * x_t[:gate_window]) * x_{t-1}. + # Per-token forward-1 smear of the embedding lane. W zero-init + lam=0 -> + # transparent at init. Uses CastedLinear so restore_fp32_params handles dtype. + self.smear_gate_enabled = h.smear_gate_enabled + if self.smear_gate_enabled: + self.smear_window = h.gate_window + self.smear_gate = CastedLinear(self.smear_window, 1, bias=False) + self.smear_gate._zero_init = True + self.smear_lambda = nn.Parameter(torch.zeros(1, dtype=torch.float32)) + # Spec 015 Recur-Alpha (port #1714). All state attr-guarded so default=off + # is byte-identical to baseline (attr checks in forward look for None). + self.recur_alpha_enabled = bool(h.recur_alpha_enabled) and h.num_loops > 0 + self.recur_diag_p2p_cos = bool(h.recur_diag_p2p_cos) and h.num_loops > 0 + self.loop_start = h.loop_start + self.loop_end = h.loop_end + if self.recur_alpha_enabled: + num_looped = h.loop_end - h.loop_start + 1 + self.num_looped = num_looped + # Spec 036: sparse-gate 8xH promotion using updated recurrent carry + # from the sparse learnable-alpha/beta experiment, rounded to 2 + # decimal places before freezing. + self.register_buffer( + "recur_beta", + torch.tensor([1.56, 1.85, 2.13], dtype=torch.float32), + ) + self.register_buffer( + "recur_alpha", + torch.tensor( + [[0.23, 0.04, 0.03], + [0.13, -0.34, 0.01], + [0.06, 0.19, -0.02]], + dtype=torch.float32, + ), + ) + # Precompute alpha_info lists: parallel to encoder_indices and + # decoder_indices, indicating (pass_offset, local_idx) or None for + # each position. Pass counts span encoder + decoder (sequential). + visits = {} + self._encoder_alpha_info = [] + for idx in self.encoder_indices: + pi = visits.get(idx, 0) + visits[idx] = pi + 1 + if self.loop_start <= idx <= self.loop_end and pi > 0: + self._encoder_alpha_info.append((pi - 1, idx - self.loop_start)) + else: + self._encoder_alpha_info.append(None) + self._decoder_alpha_info = [] + for idx in self.decoder_indices: + pi = visits.get(idx, 0) + visits[idx] = pi + 1 + if self.loop_start <= idx <= self.loop_end and pi > 0: + self._decoder_alpha_info.append((pi - 1, idx - self.loop_start)) + else: + self._decoder_alpha_info.append(None) + # Depth curriculum: _dec_idx_intermediate is NOT a prefix of + # decoder_indices (enc/dec split shifts with one fewer loop pass), + # so _decoder_alpha_info indexed by dec_int step_idx is misaligned. + # Build separate alpha_info lists from the intermediate sequences. + if (h.loop_depth_upgrade_at > 0 and h.num_loops >= 2 + and hasattr(self, '_enc_idx_intermediate')): + visits_int = {} + self._encoder_alpha_info_int = [] + for idx in self._enc_idx_intermediate: + pi = visits_int.get(idx, 0) + visits_int[idx] = pi + 1 + if self.loop_start <= idx <= self.loop_end and pi > 0: + self._encoder_alpha_info_int.append((pi - 1, idx - self.loop_start)) + else: + self._encoder_alpha_info_int.append(None) + self._decoder_alpha_info_int = [] + for idx in self._dec_idx_intermediate: + pi = visits_int.get(idx, 0) + visits_int[idx] = pi + 1 + if self.loop_start <= idx <= self.loop_end and pi > 0: + self._decoder_alpha_info_int.append((pi - 1, idx - self.loop_start)) + else: + self._decoder_alpha_info_int.append(None) + else: + self._encoder_alpha_info_int = None + self._decoder_alpha_info_int = None + else: + self.recur_beta = None + self.recur_alpha = None + self.num_looped = 0 + self._encoder_alpha_info = None + self._decoder_alpha_info = None + self._encoder_alpha_info_int = None + self._decoder_alpha_info_int = None + # Diagnostic buffers for p2p cosine. Updated in forward, read by logger. + # Detached values only; no backprop through these. + if self.recur_diag_p2p_cos: + num_looped = h.loop_end - h.loop_start + 1 + # Mean cosine per (pass_pair, layer) — num_loops pass pairs × num_looped layers. + self.register_buffer( + "_diag_p2p_cos", + torch.zeros(h.num_loops, num_looped, dtype=torch.float32), + persistent=False, + ) + self._diag_prev_deltas = {} # layer_idx -> tensor (updated in forward) + self._init_weights() + + def _init_weights(self): + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) + nn.init.zeros_(self.qo_bank.data[n + i]) + self.qo_bank.data[n + i].mul_(proj_scale) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) + for i in range(n): + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) + nn.init.zeros_(self.mlp_down_bank.data[i]) + self.mlp_down_bank.data[i].mul_(proj_scale) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif ( + module.weight.ndim == 2 + and module.weight.shape[0] >= 64 + and module.weight.shape[1] >= 64 + ): + nn.init.orthogonal_(module.weight, gain=1.0) + + def _bank_weights(self, i): + n = self.num_layers + return ( + self.qo_bank[i], + self.kv_bank[i], + self.kv_bank[n + i], + self.qo_bank[n + i], + self.mlp_up_bank[i], + self.mlp_down_bank[i], + ) + + def _parallel_block( + self, block_idx, lane0, lane1, x0, + q_w, k_w, v_w, out_w, up_w, down_w, + cu_seqlens=None, max_seqlen=0, + ): + block = self.blocks[block_idx] + mix = block.resid_mix.to(dtype=lane0.dtype) + attn_read = mix[0][None, None, :] * lane0 + mix[1][None, None, :] * x0 + attn_out = block.attn( + block.attn_norm(attn_read) * block.ln_scale_factor, + q_w, k_w, v_w, out_w, + cu_seqlens=cu_seqlens, max_seqlen=max_seqlen, + ) + attn_out = block.attn_scale.to(dtype=attn_out.dtype)[None, None, :] * attn_out + mlp_read = lane1 + mlp_out = block.mlp_scale.to(dtype=lane1.dtype)[None, None, :] * block.mlp( + block.mlp_norm(mlp_read) * block.ln_scale_factor, up_w, down_w + ) + attn_resid = self.parallel_resid_lambdas[block_idx, 0].to(dtype=lane0.dtype) + attn_post = self.parallel_post_lambdas[block_idx, 0].to(dtype=lane0.dtype) + mlp_resid = self.parallel_resid_lambdas[block_idx, 1].to(dtype=lane0.dtype) + mlp_post = self.parallel_post_lambdas[block_idx, 1].to(dtype=lane0.dtype) + lane0 = attn_resid * lane0 + attn_post[0] * attn_out + mlp_post[0] * mlp_out + lane1 = mlp_resid * lane1 + attn_post[1] * attn_out + mlp_post[1] * mlp_out + return lane0, lane1 + + def _final_parallel_hidden(self, lane0, lane1): + if self.parallel_final_lane == "mlp": + return lane1 + if self.parallel_final_lane == "attn": + return lane0 + return 0.5 * (lane0 + lane1) + + def _forward_hidden(self, input_ids, cu_seqlens=None, max_seqlen=0): + x = self.tok_emb(input_ids) + # SmearGate (PR #1667). Inline gate compute with .contiguous() on the slice fed + # to the projection so torch.compile fullgraph is happy. lam=0 + W=0 -> identity + # at init. This block runs unconditionally on the smear path; the cat keeps + # position 0 untouched so causality holds. + if self.smear_gate_enabled: + sl = self.smear_lambda.to(dtype=x.dtype) + gate_in = x[:, 1:, : self.smear_window].contiguous() + g = sl * torch.sigmoid(self.smear_gate(gate_in)) + x = torch.cat([x[:, :1], x[:, 1:] + g * x[:, :-1]], dim=1) + x = F.rms_norm(x, (x.size(-1),)) + x0 = x + skips = [] + if self.looping_active: + if self.looping_depth < self._num_loops and hasattr(self, '_enc_idx_intermediate'): + enc_iter = self._enc_idx_intermediate + dec_iter = self._dec_idx_intermediate + else: + enc_iter = self.encoder_indices + dec_iter = self.decoder_indices + else: + enc_iter = range(self.num_encoder_layers) + dec_iter = range( + self.num_encoder_layers, + self.num_encoder_layers + self.num_decoder_layers, + ) + # Spec 015: use precomputed alpha_info only when Recur-Alpha enabled AND + # looping is active. When inactive, behavior is byte-identical to baseline. + # Spec 029: during depth curriculum (looping_depth < _num_loops), use + # the intermediate-sequence alpha_info to avoid misalignment. + if self.recur_alpha is not None and self.looping_active: + if (self.looping_depth < self._num_loops + and self._encoder_alpha_info_int is not None): + enc_alpha_info = self._encoder_alpha_info_int + else: + enc_alpha_info = self._encoder_alpha_info + else: + enc_alpha_info = None + carry = {} if enc_alpha_info is not None else None + for step_idx, i in enumerate(enc_iter): + q_w, k_w, v_w, out_w, up_w, down_w = self._bank_weights(i) + x_before = x + x_new = self.blocks[i](x_before, x0, q_w, k_w, v_w, out_w, up_w, down_w, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + if enc_alpha_info is not None and enc_alpha_info[step_idx] is not None: + pass_off, local_idx = enc_alpha_info[step_idx] + beta = self.recur_beta[local_idx].to(x_new.dtype) + x = beta * x_new + for j in range(self.num_looped): + x = x + self.recur_alpha[local_idx, j].to(x_new.dtype) * carry[self.loop_start + j] + # Diagnostic: p2p cosine similarity on block deltas (optional). + if self.recur_diag_p2p_cos: + delta_this = (x_new - x_before).detach() + prev = self._diag_prev_deltas.get(i, None) + if prev is not None: + flat_this = delta_this.reshape(-1, delta_this.size(-1)) + flat_prev = prev.reshape(-1, prev.size(-1)) + cos = F.cosine_similarity(flat_this, flat_prev, dim=-1).mean() + self._diag_p2p_cos[pass_off, local_idx] = cos + self._diag_prev_deltas[i] = delta_this + else: + x = x_new + if carry is not None and self.loop_start <= i <= self.loop_end: + carry[i] = x_new.detach() + if self.recur_diag_p2p_cos and self.loop_start <= i <= self.loop_end: + self._diag_prev_deltas[i] = (x_new - x_before).detach() + skips.append(x) + psl = self.parallel_start_layer + lane0 = None + lane1 = None + # Spec 015: alpha_info for decoder; shares visit-count state with encoder + # (see __init__ precompute). None when Recur-Alpha disabled or looping inactive. + if self.recur_alpha is not None and self.looping_active: + if (self.looping_depth < self._num_loops + and self._decoder_alpha_info_int is not None): + dec_alpha_info = self._decoder_alpha_info_int + else: + dec_alpha_info = self._decoder_alpha_info + else: + dec_alpha_info = None + for skip_idx, i in enumerate(dec_iter): + q_w, k_w, v_w, out_w, up_w, down_w = self._bank_weights(i) + if i >= psl and psl > 0: + if lane0 is None: + lane0 = x + lane1 = x + if skip_idx < self.num_skip_weights and skips: + skip = skips.pop() + w = self.skip_weights[skip_idx].to(dtype=lane0.dtype)[None, None, :] + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[skip_idx].to(dtype=lane0.dtype))[None, None, :] + lane0 = torch.lerp(w * skip, lane0, g) + else: + lane0 = lane0 + w * skip + lane0, lane1 = self._parallel_block( + i, lane0, lane1, x0, q_w, k_w, v_w, out_w, up_w, down_w, + cu_seqlens=cu_seqlens, max_seqlen=max_seqlen, + ) + else: + if skip_idx < self.num_skip_weights and skips: + scaled_skip = ( + self.skip_weights[skip_idx].to(dtype=x.dtype)[None, None, :] + * skips.pop() + ) + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[skip_idx].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + x_before = x + x_new = self.blocks[i](x_before, x0, q_w, k_w, v_w, out_w, up_w, down_w, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + if dec_alpha_info is not None and dec_alpha_info[skip_idx] is not None: + pass_off, local_idx = dec_alpha_info[skip_idx] + beta = self.recur_beta[local_idx].to(x_new.dtype) + x = beta * x_new + for j in range(self.num_looped): + x = x + self.recur_alpha[local_idx, j].to(x_new.dtype) * carry[self.loop_start + j] + if self.recur_diag_p2p_cos: + delta_this = (x_new - x_before).detach() + prev = self._diag_prev_deltas.get(i, None) + if prev is not None: + flat_this = delta_this.reshape(-1, delta_this.size(-1)) + flat_prev = prev.reshape(-1, prev.size(-1)) + cos = F.cosine_similarity(flat_this, flat_prev, dim=-1).mean() + self._diag_p2p_cos[pass_off, local_idx] = cos + self._diag_prev_deltas[i] = delta_this + else: + x = x_new + if carry is not None and self.loop_start <= i <= self.loop_end: + carry[i] = x_new.detach() + if self.recur_diag_p2p_cos and self.loop_start <= i <= self.loop_end: + self._diag_prev_deltas[i] = (x_new - x_before).detach() + if lane0 is not None: + x = self._final_parallel_hidden(lane0, lane1) + return self.final_norm(x) + + def _project_logits(self, hidden): + if self.tie_embeddings: + return F.linear(hidden, self.tok_emb.weight) + return self.lm_head(hidden) + + def forward_logits(self, input_ids, cu_seqlens=None, max_seqlen=0): + hidden = self._forward_hidden( + input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen + ) + logits_proj = self._project_logits(hidden) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward(self, input_ids, target_ids, cu_seqlens=None, max_seqlen=0): + hidden = self._forward_hidden(input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + logits_proj = self._project_logits(hidden) + flat_targets = target_ids.reshape(-1) + if self.fused_ce_enabled: + return softcapped_cross_entropy( + logits_proj.reshape(-1, logits_proj.size(-1)), + flat_targets, + self.logit_softcap, + reduction="mean", + ) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + flat_targets, + reduction="mean", + ) + + def forward_ttt(self, input_ids, target_ids, lora): + x = self.tok_emb(input_ids) + # SmearGate on the TTT path — same inline compute as forward_logits. + if self.smear_gate_enabled: + sl = self.smear_lambda.to(dtype=x.dtype) + gate_in = x[:, 1:, : self.smear_window].contiguous() + g = sl * torch.sigmoid(self.smear_gate(gate_in)) + x = torch.cat([x[:, :1], x[:, 1:] + g * x[:, :-1]], dim=1) + x = F.rms_norm(x, (x.size(-1),)) + x0 = x + skips = [] + enc_iter = ( + self.encoder_indices + if self.looping_active + else list(range(self.num_encoder_layers)) + ) + dec_iter = ( + self.decoder_indices + if self.looping_active + else list( + range( + self.num_encoder_layers, + self.num_encoder_layers + self.num_decoder_layers, + ) + ) + ) + # TTT α fix: apply the same blend as forward_logits. Without this, + # TTT adaptation sees the un-α-weighted forward pass, which mismatches + # training and leaves ~0.002 of TTT delta on the table (017-era bug). + enc_alpha_info = ( + self._encoder_alpha_info + if (self.recur_alpha is not None and self.looping_active) + else None + ) + carry = {} if enc_alpha_info is not None else None + slot = 0 + for step_idx, i in enumerate(enc_iter): + q_w, k_w, v_w, out_w, up_w, down_w = self._bank_weights(i) + x_before = x + x_new = self._block_with_lora(self.blocks[i], x_before, x0, lora, slot, q_w, k_w, v_w, out_w, up_w, down_w) + if enc_alpha_info is not None and enc_alpha_info[step_idx] is not None: + pass_off, local_idx = enc_alpha_info[step_idx] + beta = self.recur_beta[local_idx].to(x_new.dtype) + x = beta * x_new + for j in range(self.num_looped): + x = x + self.recur_alpha[local_idx, j].to(x_new.dtype) * carry[self.loop_start + j] + else: + x = x_new + if carry is not None and self.loop_start <= i <= self.loop_end: + carry[i] = x_new.detach() + slot += 1 + skips.append(x) + psl = self.parallel_start_layer + lane0 = None + lane1 = None + dec_alpha_info = ( + self._decoder_alpha_info + if (self.recur_alpha is not None and self.looping_active) + else None + ) + for skip_idx, i in enumerate(dec_iter): + q_w, k_w, v_w, out_w, up_w, down_w = self._bank_weights(i) + if i >= psl and psl > 0: + if lane0 is None: + lane0 = x + lane1 = x + if skip_idx < self.num_skip_weights and skips: + skip = skips.pop() + w = self.skip_weights[skip_idx].to(dtype=lane0.dtype)[None, None, :] + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[skip_idx].to(dtype=lane0.dtype))[None, None, :] + lane0 = torch.lerp(w * skip, lane0, g) + else: + lane0 = lane0 + w * skip + lane0, lane1 = self._parallel_block_with_lora( + i, lane0, lane1, x0, lora, slot, + q_w, k_w, v_w, out_w, up_w, down_w, + ) + else: + if skip_idx < self.num_skip_weights and skips: + scaled_skip = ( + self.skip_weights[skip_idx].to(dtype=x.dtype)[None, None, :] + * skips.pop() + ) + if self.skip_gates is not None: + g = torch.sigmoid(self.skip_gates[skip_idx].to(dtype=x.dtype))[None, None, :] + x = torch.lerp(scaled_skip, x, g) + else: + x = x + scaled_skip + x_before = x + x_new = self._block_with_lora(self.blocks[i], x_before, x0, lora, slot, q_w, k_w, v_w, out_w, up_w, down_w) + if dec_alpha_info is not None and dec_alpha_info[skip_idx] is not None: + pass_off, local_idx = dec_alpha_info[skip_idx] + beta = self.recur_beta[local_idx].to(x_new.dtype) + x = beta * x_new + for j in range(self.num_looped): + x = x + self.recur_alpha[local_idx, j].to(x_new.dtype) * carry[self.loop_start + j] + else: + x = x_new + if carry is not None and self.loop_start <= i <= self.loop_end: + carry[i] = x_new.detach() + slot += 1 + if lane0 is not None: + x = self._final_parallel_hidden(lane0, lane1) + x = self.final_norm(x) + if self.tie_embeddings: + logits = F.linear(x, self.tok_emb.weight) + else: + logits = self.lm_head(x) + logits = logits + lora.lm_head_lora(x) + logits = self.logit_softcap * torch.tanh(logits / self.logit_softcap) + bsz, sl, V = logits.shape + return F.cross_entropy( + logits.float().reshape(-1, V), target_ids.reshape(-1), reduction="none" + ).reshape(bsz, sl) + + def _block_with_lora(self, block, x, x0, lora, slot, q_w, k_w, v_w, out_w, up_w, down_w): + mix = block.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + n = block.attn_norm(x_in) * block.ln_scale_factor + attn = block.attn + bsz, seqlen, dim = n.shape + # SpinQuant V1 (TTT path): rotate n for base Q/K/V linears. LoRA continues + # to see unrotated n so the adapter output lives in the unrotated basis, + # which is the same basis the base path produces after R cancels. + if CastedLinear._sq_active and hasattr(attn, "_sq_R_attn_in"): + n_qkv = n @ attn._sq_R_attn_in.to(n.dtype) + else: + n_qkv = n + # Keep raw Q for AttnOutGate src='q' (matches forward path semantics). + q_raw = F.linear(n_qkv, q_w.to(n.dtype)) + lora.q_loras[slot](n) + q = q_raw.reshape(bsz, seqlen, attn.num_heads, attn.head_dim) + k = F.linear(n_qkv, k_w.to(n.dtype)) + if lora.k_loras is not None: + k = k + lora.k_loras[slot](n) + k = k.reshape(bsz, seqlen, attn.num_kv_heads, attn.head_dim) + v = (F.linear(n_qkv, v_w.to(n.dtype)) + lora.v_loras[slot](n)).reshape( + bsz, seqlen, attn.num_kv_heads, attn.head_dim + ) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = attn.rotary(seqlen, n.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, attn.rope_dims) + k = apply_rotary_emb(k, cos, sin, attn.rope_dims) + q = q * attn.q_gain.to(dtype=q.dtype)[None, None, :, None] + y = flash_attn_3_func(q, k, v, causal=True) + if attn.use_xsa: + y = attn._xsa_efficient(y, v) + # AttnOutGate (TTT path) — inline + .contiguous() barrier, same as the eval path. + if attn.attn_out_gate: + gate_src = q_raw if attn.attn_out_gate_src == "q" else n + gate_in = gate_src[..., : attn.gate_window].contiguous() + g = 2.0 * torch.sigmoid(attn.attn_gate_proj(gate_in)) + y = y * g[..., None] + # Gated Attention (TTT path). Gate input is n (post-norm block input), same + # as eval path. .to(n.dtype) on fp32 param before bf16 broadcast. + if attn.gated_attn: + n_c = n.contiguous() + g = torch.sigmoid(F.linear(n_c, attn.attn_gate_w.to(n.dtype))) + y = y * g[..., None] + # Sparse gate (TTT path). Match eval path semantics: gate uses the + # post-norm block input restricted to gate_window dims. + if attn.sparse_attn_gate: + gate_in = n[..., : attn.gate_window].contiguous() + g = torch.sigmoid( + attn.sparse_attn_gate_scale + * F.linear(gate_in, attn.attn_gate_w.to(n.dtype)) + ) + y = y * g[..., None] + y = y.reshape(bsz, seqlen, dim) + # SpinQuant V1 (TTT path): rotate attention output before out_proj. + if CastedLinear._sq_active and hasattr(attn, "_sq_R_attn_proj_in"): + y_proj = y @ attn._sq_R_attn_proj_in.to(n.dtype) + else: + y_proj = y + attn_out = F.linear(y_proj, out_w.to(n.dtype)) + if lora.o_loras is not None: + attn_out = attn_out + lora.o_loras[slot](n) + x_out = x_in + block.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + mlp_n = block.mlp_norm(x_out) * block.ln_scale_factor + mlp_out = block.mlp(mlp_n, up_w, down_w) + if lora.mlp_loras is not None: + mlp_out = mlp_out + lora.mlp_loras[slot](mlp_n) + x_out = x_out + block.mlp_scale.to(dtype=x_out.dtype)[None, None, :] * mlp_out + return x_out + + def _parallel_block_with_lora( + self, block_idx, lane0, lane1, x0, lora, slot, + q_w, k_w, v_w, out_w, up_w, down_w, + ): + block = self.blocks[block_idx] + mix = block.resid_mix.to(dtype=lane0.dtype) + attn_read = mix[0][None, None, :] * lane0 + mix[1][None, None, :] * x0 + n = block.attn_norm(attn_read) * block.ln_scale_factor + attn = block.attn + bsz, seqlen, dim = n.shape + # SpinQuant V1 (TTT parallel path): rotate n for base Q/K/V. LoRA uses unrotated n. + if CastedLinear._sq_active and hasattr(attn, "_sq_R_attn_in"): + n_qkv = n @ attn._sq_R_attn_in.to(n.dtype) + else: + n_qkv = n + q_raw = F.linear(n_qkv, q_w.to(n.dtype)) + lora.q_loras[slot](n) + q = q_raw.reshape(bsz, seqlen, attn.num_heads, attn.head_dim) + k = F.linear(n_qkv, k_w.to(n.dtype)) + if lora.k_loras is not None: + k = k + lora.k_loras[slot](n) + k = k.reshape(bsz, seqlen, attn.num_kv_heads, attn.head_dim) + v = (F.linear(n_qkv, v_w.to(n.dtype)) + lora.v_loras[slot](n)).reshape( + bsz, seqlen, attn.num_kv_heads, attn.head_dim + ) + q = F.rms_norm(q, (q.size(-1),)) + k = F.rms_norm(k, (k.size(-1),)) + cos, sin = attn.rotary(seqlen, n.device, q.dtype) + q = apply_rotary_emb(q, cos, sin, attn.rope_dims) + k = apply_rotary_emb(k, cos, sin, attn.rope_dims) + q = q * attn.q_gain.to(dtype=q.dtype)[None, None, :, None] + y = flash_attn_3_func(q, k, v, causal=True) + if attn.use_xsa: + y = attn._xsa_efficient(y, v) + # AttnOutGate (TTT parallel path) — inline + .contiguous() barrier. + if attn.attn_out_gate: + gate_src = q_raw if attn.attn_out_gate_src == "q" else n + gate_in = gate_src[..., : attn.gate_window].contiguous() + g = 2.0 * torch.sigmoid(attn.attn_gate_proj(gate_in)) + y = y * g[..., None] + # Gated Attention (TTT parallel path). Gate input is n (post-norm block input). + if attn.gated_attn: + n_c = n.contiguous() + g = torch.sigmoid(F.linear(n_c, attn.attn_gate_w.to(n.dtype))) + y = y * g[..., None] + # Sparse gate (TTT parallel path). Match eval path semantics. + if attn.sparse_attn_gate: + gate_in = n[..., : attn.gate_window].contiguous() + g = torch.sigmoid( + attn.sparse_attn_gate_scale + * F.linear(gate_in, attn.attn_gate_w.to(n.dtype)) + ) + y = y * g[..., None] + y = y.reshape(bsz, seqlen, dim) + # SpinQuant V1 (TTT parallel path): rotate attention output before out_proj. + if CastedLinear._sq_active and hasattr(attn, "_sq_R_attn_proj_in"): + y_proj = y @ attn._sq_R_attn_proj_in.to(n.dtype) + else: + y_proj = y + attn_out = F.linear(y_proj, out_w.to(n.dtype)) + if lora.o_loras is not None: + attn_out = attn_out + lora.o_loras[slot](n) + attn_out = block.attn_scale.to(dtype=attn_out.dtype)[None, None, :] * attn_out + mlp_read = lane1 + mlp_n = block.mlp_norm(mlp_read) * block.ln_scale_factor + mlp_out = block.mlp(mlp_n, up_w, down_w) + if lora.mlp_loras is not None: + mlp_out = mlp_out + lora.mlp_loras[slot](mlp_n) + mlp_out = block.mlp_scale.to(dtype=lane1.dtype)[None, None, :] * mlp_out + attn_resid = self.parallel_resid_lambdas[block_idx, 0].to(dtype=lane0.dtype) + attn_post = self.parallel_post_lambdas[block_idx, 0].to(dtype=lane0.dtype) + mlp_resid = self.parallel_resid_lambdas[block_idx, 1].to(dtype=lane0.dtype) + mlp_post = self.parallel_post_lambdas[block_idx, 1].to(dtype=lane0.dtype) + lane0 = attn_resid * lane0 + attn_post[0] * attn_out + mlp_post[0] * mlp_out + lane1 = mlp_resid * lane1 + attn_post[1] * attn_out + mlp_post[1] * mlp_out + return lane0, lane1 + + +class BatchedLinearLoRA(nn.Module): + def __init__(self, bsz, in_features, out_features, rank, alpha=96): + super().__init__() + self._bound = 1.0 / math.sqrt(in_features) + self._scale = alpha / rank + self.A = nn.Parameter( + torch.empty(bsz, rank, in_features).uniform_(-self._bound, self._bound) + ) + self.B = nn.Parameter(torch.zeros(bsz, out_features, rank)) + + def reset(self): + with torch.no_grad(): + # warm-start A: keep accumulated feature directions; only zero B so + # LoRA output = 0 at each batch start (per-document reset preserved) + self.B.zero_() + + def forward(self, x): + return (x @ self.A.transpose(1, 2)) @ self.B.transpose(1, 2) * self._scale + + +class BatchedTTTLoRA(nn.Module): + def __init__(self, bsz, model, rank, alpha=96, k_lora=True, mlp_lora=True, o_lora=True): + super().__init__() + self.bsz = bsz + dim = model.qo_bank.shape[-1] + vocab = model.tok_emb.num_embeddings + if getattr(model, "looping_active", False): + num_slots = len(model.encoder_indices) + len(model.decoder_indices) + else: + num_slots = len(model.blocks) + kv_dim = model.blocks[0].attn.num_kv_heads * ( + dim // model.blocks[0].attn.num_heads + ) + embed_dim = model.tok_emb.embedding_dim + self.lm_head_lora = BatchedLinearLoRA(bsz, embed_dim, vocab, rank, alpha) + self.q_loras = nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, dim, rank, alpha) for _ in range(num_slots)] + ) + self.v_loras = nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, kv_dim, rank, alpha) for _ in range(num_slots)] + ) + self.k_loras = ( + nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, kv_dim, rank, alpha) for _ in range(num_slots)] + ) + if k_lora + else None + ) + self.mlp_loras = ( + nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, dim, rank, alpha) for _ in range(num_slots)] + ) + if mlp_lora + else None + ) + self.o_loras = ( + nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, dim, rank, alpha) for _ in range(num_slots)] + ) + if o_lora + else None + ) + + def reset(self): + with torch.no_grad(): + self.lm_head_lora.reset() + for loras in [self.q_loras, self.v_loras, self.k_loras, + self.mlp_loras, self.o_loras]: + if loras is not None: + for lora in loras: + lora.reset() + + +# Polar Express per-iteration minimax Newton-Schulz coefficients (PR #1344, +# later reused by PR #1787). This replaces stock Muon's repeated fixed +# `(3.4445, -4.775, 2.0315)` tuple with 5 optimized per-step tuples while +# keeping `MUON_BACKEND_STEPS=5`. +_PE_COEFFS = ( + (8.156554524902461, -22.48329292557795, 15.878769915207462), + (4.042929935166739, -2.808917465908714, 0.5000178451051316), + (3.8916678022926607, -2.772484153217685, 0.5060648178503393), + (3.285753657755655, -2.3681294933425376, 0.46449024233003106), + (2.3465413258596377, -1.7097828382687081, 0.42323551169305323), +) + + +@torch.compile +def zeropower_via_newtonschulz5(G, steps=10, eps=1e-07): + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + coeffs = _PE_COEFFS[:steps] if steps <= len(_PE_COEFFS) else _PE_COEFFS + for a, b, c in coeffs: + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + + +class Muon(torch.optim.Optimizer): + def __init__( + self, + params, + lr, + momentum, + backend_steps, + nesterov=True, + weight_decay=0.0, + row_normalize=False, + ): + super().__init__( + params, + dict( + lr=lr, + momentum=momentum, + backend_steps=backend_steps, + nesterov=nesterov, + weight_decay=weight_decay, + row_normalize=row_normalize, + ), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + "p": p, + "B": B, + "padded_grad": torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + "shard": torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + "shard_mom": torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + "full_update": torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + "scale": max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + self._bank_meta.sort(key=lambda m: -m["p"].numel()) + self._built = True + + def launch_reduce_scatters(self): + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m["p"] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m["padded_grad"] + pg[: m["B"]].copy_(p.grad.bfloat16()) + if pg.shape[0] > m["B"]: + pg[m["B"] :].zero_() + fut = dist.reduce_scatter_tensor( + m["shard"], pg, op=dist.ReduceOp.AVG, async_op=True + ) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + if not self._built: + self._build() + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + row_normalize = group.get("row_normalize", False) + prev_ag_handle = None + prev_m = None + sharded = self._distributed and hasattr(self, "_rs_futures") + for idx, m in enumerate(self._bank_meta): + p = m["p"] + if p.grad is None: + continue + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m["p"] + upd = prev_m["full_update"][: prev_m["B"]] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m["scale"]) + if sharded and self._rs_futures[idx] is not None: + self._rs_futures[idx].wait() + g = m["shard"] + buf = m["shard_mom"] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + if row_normalize: + rn = update.float().norm(dim=-1, keepdim=True).clamp_min(1e-07) + update = update / rn.to(update.dtype) + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m["full_update"], update, async_op=True + ) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m["scale"]) + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m["p"] + upd = prev_m["full_update"][: prev_m["B"]] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m["scale"]) + if hasattr(self, "_rs_futures"): + del self._rs_futures + return loss + + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,skip_gates,parallel_post_lambdas,parallel_resid_lambdas,attn_gate_proj,attn_gate_w,smear_gate,smear_lambda", + ).split(",") + if pattern +) + + +PACKED_REPLICATED_GRAD_MAX_NUMEL = 1 << 15 + + +class Optimizers: + def __init__(self, h, base_model): + matrix_params = [ + base_model.qo_bank, + base_model.kv_bank, + base_model.mlp_up_bank, + base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for (name, p) in block_named_params + if p.ndim < 2 + or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + if base_model.skip_gates is not None and base_model.skip_gates.numel() > 0: + scalar_params.append(base_model.skip_gates) + if base_model.parallel_post_lambdas is not None: + scalar_params.append(base_model.parallel_post_lambdas) + if base_model.parallel_resid_lambdas is not None: + scalar_params.append(base_model.parallel_resid_lambdas) + # SmearGate params live on GPT root (not in .blocks), so add them by hand. + # Both are tiny (gate_window scalars + 1 lambda). Optimized via scalar Adam. + if getattr(base_model, "smear_gate_enabled", False): + scalar_params.append(base_model.smear_gate.weight) + scalar_params.append(base_model.smear_lambda) + # Spec 015 Recur-Alpha: 6 scalars (num_loops × num_looped), route to scalar AdamW. + # Not in .blocks so not picked up by block_named_params. ndim=2 but tiny — + # would be silly to send to Muon. Append by hand like SmearGate. + # Spec 021: recur_alpha is frozen (buffer or Parameter(requires_grad=False)), + # so DO NOT append to optimizer. Guard on requires_grad so the + # 015/016/017 learnable-Parameter form still works if we ever revert. + if getattr(base_model, "recur_alpha_enabled", False): + if base_model.recur_alpha is not None and base_model.recur_alpha.requires_grad: + scalar_params.append(base_model.recur_alpha) + if getattr(base_model, "recur_beta", None) is not None and base_model.recur_beta.requires_grad: + scalar_params.append(base_model.recur_beta) + token_lr = h.tied_embed_lr if h.tie_embeddings else h.embed_lr + tok_params = [ + {"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr} + ] + self.optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(h.beta1, h.beta2), + eps=h.adam_eps, + weight_decay=h.embed_wd, + fused=True, + ) + self.optimizer_muon = Muon( + matrix_params, + lr=h.matrix_lr, + momentum=h.muon_momentum, + backend_steps=h.muon_backend_steps, + weight_decay=h.muon_wd, + row_normalize=h.muon_row_normalize, + ) + for group in self.optimizer_muon.param_groups: + group["base_lr"] = h.matrix_lr + self.optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": h.scalar_lr, "base_lr": h.scalar_lr}], + betas=(h.beta1, h.beta2), + eps=h.adam_eps, + weight_decay=h.adam_wd, + fused=True, + ) + self.optimizers = [ + self.optimizer_tok, + self.optimizer_muon, + self.optimizer_scalar, + ] + self.replicated_params = list(tok_params[0]["params"]) + self.replicated_params.extend(scalar_params) + self.replicated_large_params = [] + self.replicated_packed_params = [] + for p in self.replicated_params: + if p.numel() <= PACKED_REPLICATED_GRAD_MAX_NUMEL: + self.replicated_packed_params.append(p) + else: + self.replicated_large_params.append(p) + + def __iter__(self): + return iter(self.optimizers) + + def zero_grad_all(self): + for opt in self.optimizers: + opt.zero_grad(set_to_none=True) + + def _all_reduce_packed_grads(self): + grads_by_key = collections.defaultdict(list) + for p in self.replicated_packed_params: + if p.grad is not None: + grads_by_key[(p.grad.device, p.grad.dtype)].append(p.grad) + for grads in grads_by_key.values(): + flat = torch.empty( + sum(g.numel() for g in grads), + device=grads[0].device, + dtype=grads[0].dtype, + ) + offset = 0 + for g in grads: + n = g.numel() + flat[offset : offset + n].copy_(g.contiguous().view(-1)) + offset += n + dist.all_reduce(flat, op=dist.ReduceOp.AVG) + offset = 0 + for g in grads: + n = g.numel() + g.copy_(flat[offset : offset + n].view_as(g)) + offset += n + + def step(self, distributed=False): + self.optimizer_muon.launch_reduce_scatters() + if distributed: + reduce_handles = [ + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG, async_op=True) + for p in self.replicated_large_params + if p.grad is not None + ] + self._all_reduce_packed_grads() + for handle in reduce_handles: + handle.wait() + self.optimizer_tok.step() + self.optimizer_scalar.step() + self.optimizer_muon.step() + self.zero_grad_all() + + +def restore_fp32_params(model): + for module in model.modules(): + if isinstance(module, CastedLinear): + module.float() + for name, param in model.named_parameters(): + if ( + param.ndim < 2 + or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ) and param.dtype != torch.float32: + param.data = param.data.float() + if hasattr(model, "qo_bank") and model.qo_bank is not None: + model.qo_bank.data = model.qo_bank.data.float() + model.kv_bank.data = model.kv_bank.data.float() + model.mlp_up_bank.data = model.mlp_up_bank.data.float() + model.mlp_down_bank.data = model.mlp_down_bank.data.float() + + +def collect_hessians(model, train_loader, h, device, n_calibration_batches=64): + hessians = {} + hooks = [] + for i, block in enumerate(model.blocks): + block.attn._calib = True + block.mlp._calib = True + block.mlp.use_fused = False + + def make_attn_hook(layer_idx): + def hook_fn(module, inp, out): + x = inp[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + for suffix in ["c_q", "c_k", "c_v"]: + name = f"blocks.{layer_idx}.attn.{suffix}.weight" + if name not in hessians: + hessians[name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(x.T, x) + y = module._last_proj_input + if y is not None: + y = y.float() + if y.ndim == 3: + y = y.reshape(-1, y.shape[-1]) + name = f"blocks.{layer_idx}.attn.proj.weight" + if name not in hessians: + hessians[name] = torch.zeros( + y.shape[1], y.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(y.T, y) + return hook_fn + + def make_mlp_hook(layer_idx): + def hook_fn(module, inp, out): + x = inp[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + name = f"blocks.{layer_idx}.mlp.fc.weight" + if name not in hessians: + hessians[name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(x.T, x) + h_act = module._last_down_input + if h_act is not None: + h_act = h_act.float() + if h_act.ndim == 3: + h_act = h_act.reshape(-1, h_act.shape[-1]) + name = f"blocks.{layer_idx}.mlp.proj.weight" + if name not in hessians: + hessians[name] = torch.zeros( + h_act.shape[1], h_act.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(h_act.T, h_act) + return hook_fn + + for i, block in enumerate(model.blocks): + hooks.append(block.attn.register_forward_hook(make_attn_hook(i))) + hooks.append(block.mlp.register_forward_hook(make_mlp_hook(i))) + + # Hessian hooks for embedding factorization projection layers + def make_linear_input_hook(weight_name): + def hook_fn(module, inp, out): + x = inp[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + if weight_name not in hessians: + hessians[weight_name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[weight_name].addmm_(x.T, x) + return hook_fn + + if model.tie_embeddings: + hook_module = model.final_norm + + def make_output_hook(name): + def hook_fn(module, inp, out): + x = out.detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + if name not in hessians: + hessians[name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(x.T, x) + return hook_fn + + hooks.append( + hook_module.register_forward_hook(make_output_hook("tok_emb.weight")) + ) + model.eval() + with torch.no_grad(): + for _ in range(n_calibration_batches): + x, _ = train_loader.next_batch(h.train_batch_tokens, h.grad_accum_steps) + model.forward_logits(x) + for hook in hooks: + hook.remove() + for i, block in enumerate(model.blocks): + block.attn._calib = False + block.mlp._calib = False + block.mlp.use_fused = True + for name in hessians: + hessians[name] = hessians[name].cpu() / n_calibration_batches + return hessians + + +def gptq_quantize_weight(w, H, clip_sigmas=3.0, clip_range=63, block_size=128): + W_orig = w.float().clone() + rows, cols = W_orig.shape + H = H.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = 0.01 * H.diag().mean() + H.diagonal().add_(damp) + perm = torch.argsort(H.diag(), descending=True) + invperm = torch.argsort(perm) + W_perm = W_orig[:, perm].clone() + W_perm[:, dead[perm]] = 0 + H = H[perm][:, perm] + Hinv = torch.cholesky_inverse(torch.linalg.cholesky(H)) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + row_std = W_orig.std(dim=1) + s = (clip_sigmas * row_std / clip_range).clamp_min(1e-10).to(torch.float16) + sf = s.float() + Q = torch.zeros(rows, cols, dtype=torch.int8) + W_work = W_perm.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + W_block = W_work[:, i1:i2].clone() + Hinv_block = Hinv[i1:i2, i1:i2] + Err = torch.zeros(rows, i2 - i1) + for j in range(i2 - i1): + w_col = W_block[:, j] + d = Hinv_block[j, j] + q_col = torch.clamp(torch.round(w_col / sf), -clip_range, clip_range) + Q[:, i1 + j] = q_col.to(torch.int8) + err = (w_col - q_col.float() * sf) / d + Err[:, j] = err + W_block[:, j:] -= err.unsqueeze(1) * Hinv_block[j, j:].unsqueeze(0) + if i2 < cols: + W_work[:, i2:] -= Err @ Hinv[i1:i2, i2:] + return Q[:, invperm], s + + +def _quantize_gate_int8_row(w): + # Symmetric int8-per-row quantization for small gate tensors. w shape + # (R, C) -> (R,) scales in fp16, int8 values in [-127, 127]. Single scale + # per row keeps accuracy high while halving storage vs fp16. + W = w.float().contiguous() + row_max = W.abs().amax(dim=1).clamp_min(1e-10) + s = (row_max / 127.0).to(torch.float16) + sf = s.float().view(-1, 1) + q = torch.clamp(torch.round(W / sf), -127, 127).to(torch.int8) + return q, s + + +def gptq_mixed_quantize(state_dict, hessians, h): + result = {} + meta = {} + quant_gate = bool(getattr(h, "gated_attn_quant_gate", False)) + for (name, tensor) in state_dict.items(): + t = tensor.detach().cpu().contiguous() + # Dedicated int8-per-row path for attn_gate_w (bypasses both GPTQ and + # fp16 passthrough). Applied BEFORE the numel<=65536 passthrough check + # so the gate tensor is routed here instead of to fp16. + if ( + quant_gate + and t.is_floating_point() + and t.ndim == 2 + and name.endswith(".attn_gate_w") + and 1024 <= t.numel() <= 8192 + ): + gq, gs = _quantize_gate_int8_row(t) + result[name + ".gq"] = gq + result[name + ".gs"] = gs + meta[name] = "gate_int8_row" + continue + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough (float16)" + continue + if "tok_emb" in name: + cs = h.embed_clip_sigmas + elif ".mlp." in name: + cs = h.mlp_clip_sigmas + elif ".attn." in name: + cs = h.attn_clip_sigmas + else: + cs = h.matrix_clip_sigmas + bits = h.embed_bits if "tok_emb" in name else h.matrix_bits + clip_range = 2 ** (bits - 1) - 1 + ret = gptq_quantize_weight( + t, hessians[name], clip_sigmas=cs, clip_range=clip_range + ) + q, s = ret + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = f"gptq (int{bits})" + categories = collections.defaultdict(set) + for (name, cat) in meta.items(): + short = re.sub("\\.\\d+$", "", re.sub("blocks\\.\\d+", "blocks", name)) + categories[cat].add(short) + log("Quantized weights:") + for cat in sorted(categories): + log(f" {cat}: {', '.join(sorted(categories[cat]))}") + return result, meta + +def dequantize_mixed(result, meta, template_sd): + out = {} + for (name, orig) in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if "passthrough" in info: + t = result[name] + if t.dtype == torch.float16 and orig_dtype in ( + torch.float32, + torch.bfloat16, + ): + t = t.to(orig_dtype) + out[name] = t + continue + if info == "gate_int8_row": + gq = result[name + ".gq"] + gs = result[name + ".gs"] + out[name] = (gq.float() * gs.float().view(-1, 1)).to(orig_dtype) + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + out[name] = ( + q.float() * s.float().view(q.shape[0], *[1] * (q.ndim - 1)) + ).to(orig_dtype) + else: + out[name] = (q.float() * float(s.item())).to(orig_dtype) + return out + + +_BSHF_MAGIC = b"BSHF" + + +def _byte_shuffle(data, stride=2): + if stride <= 1 or len(data) < stride: + return data + src = np.frombuffer(data, dtype=np.uint8) + n = len(src) + out = np.empty(n, dtype=np.uint8) + dest_off = 0 + for pos in range(stride): + chunk = src[pos::stride] + out[dest_off : dest_off + len(chunk)] = chunk + dest_off += len(chunk) + return _BSHF_MAGIC + bytes([stride]) + out.tobytes() + + +def _byte_unshuffle(data): + if len(data) < 5 or data[:4] != _BSHF_MAGIC: + return data + stride = data[4] + if stride < 2: + return data[5:] + payload = np.frombuffer(data, dtype=np.uint8, offset=5) + n = len(payload) + out = np.empty(n, dtype=np.uint8) + src_off = 0 + for pos in range(stride): + chunk_len = n // stride + (1 if pos < n % stride else 0) + out[pos::stride][:chunk_len] = payload[src_off : src_off + chunk_len] + src_off += chunk_len + return out.tobytes() + + +def _compress(data, compressor): + data = _byte_shuffle(data) + if compressor == "lzma": + return lzma.compress(data, preset=6) + elif compressor == "brotli": + import brotli + + return brotli.compress(data, quality=11) + raise ValueError(f"Unknown compressor: {compressor!r}") + + +def _decompress(data, compressor): + if compressor == "lzma": + raw = lzma.decompress(data) + elif compressor == "brotli": + import brotli + + raw = brotli.decompress(data) + else: + raise ValueError(f"Unknown compressor: {compressor!r}") + raw = _byte_unshuffle(raw) + return raw + + +def _unbank_state_dict(state_dict, num_layers): + sd = {} + n = num_layers + for k, v in state_dict.items(): + t = v.detach().cpu() if v is not None else None + if k == "qo_bank": + for i in range(n): + sd[f"blocks.{i}.attn.c_q.weight"] = t[i] + sd[f"blocks.{i}.attn.proj.weight"] = t[n + i] + elif k == "kv_bank": + for i in range(n): + sd[f"blocks.{i}.attn.c_k.weight"] = t[i] + sd[f"blocks.{i}.attn.c_v.weight"] = t[n + i] + elif k == "mlp_up_bank": + for i in range(n): + sd[f"blocks.{i}.mlp.fc.weight"] = t[i] + elif k == "mlp_down_bank": + for i in range(n): + sd[f"blocks.{i}.mlp.proj.weight"] = t[i] + else: + if t is not None: + sd[k] = t + return sd + + +def _rebank_state_dict(flat_sd, num_layers, model_dim, kv_dim, hidden_dim): + sd = {} + n = num_layers + sd["qo_bank"] = torch.zeros(2 * n, model_dim, model_dim) + sd["kv_bank"] = torch.zeros(2 * n, kv_dim, model_dim) + for i in range(n): + sd["qo_bank"][i] = flat_sd[f"blocks.{i}.attn.c_q.weight"] + sd["qo_bank"][n + i] = flat_sd[f"blocks.{i}.attn.proj.weight"] + sd["kv_bank"][i] = flat_sd[f"blocks.{i}.attn.c_k.weight"] + sd["kv_bank"][n + i] = flat_sd[f"blocks.{i}.attn.c_v.weight"] + sd["mlp_up_bank"] = torch.zeros(n, hidden_dim, model_dim) + sd["mlp_down_bank"] = torch.zeros(n, model_dim, hidden_dim) + for i in range(n): + sd["mlp_up_bank"][i] = flat_sd[f"blocks.{i}.mlp.fc.weight"] + sd["mlp_down_bank"][i] = flat_sd[f"blocks.{i}.mlp.proj.weight"] + for k, v in flat_sd.items(): + if not ( + k.startswith("blocks.") + and any( + p in k + for p in [ + ".attn.c_q.", ".attn.c_k.", ".attn.c_v.", + ".attn.proj.", ".mlp.fc.", ".mlp.proj.", + ] + ) + ): + sd[k] = v + return sd + + +# ============================================================================= +# SpinQuant V1 — Hadamard rotation primitives (port from PR #1695) +# ============================================================================= +# Zero serialized bytes: rotations are regenerated deterministically from +# (SPINQUANT_SEED, tag) at load time. Applied in two places: +# +# 1. Statically on the state_dict + Hessians, right before GPTQ quantizes: +# W_rot = W @ R (input-col rotation, so F.linear(x @ R, W_rot) == F.linear(x, W)) +# H_rot = R.T @ H @ R (matches the rotated activation covariance) +# +# 2. Online at forward time, gated by CastedLinear._sq_active: +# x_rotated = x @ R before each of the 4 target linear layers. +# +# Math: orthogonal R means R @ R.T == I, so the static + online rotations +# cancel at fp precision. Pre-quant forward is bit-identical to unrotated. +# Only GPTQ sees the rotated basis, where outliers are spread more evenly +# and quantization error is reduced. +# +# Four rotation sites: +# - R_attn_in (d_model) — residual → Q/K/V input +# - R_attn_proj_in (d_model) — attention output → proj input +# - R_mlp_in (d_model) — residual → fc input +# - R_mlp_proj_in (d_ff) — post-LeakyReLU² → proj input +# +# The MLP proj rotation is applied AFTER the nonlinearity, so the rotation +# never has to commute with LeakyReLU. Residual stream is untouched, so all +# per-channel multipliers (attn_scale, mlp_scale, resid_mix, skip_weights) +# operate in their trained basis. + +_SPINQUANT_CACHE: "dict[tuple[int, str, int], torch.Tensor]" = {} + + +def _stable_seed(seed: int, tag: str) -> int: + """SHA-256-derived seed. Deterministic across processes; Python's built-in + hash() varies with PYTHONHASHSEED and would desync train vs eval.""" + h = hashlib.sha256(f"{seed}:{tag}".encode("utf-8")).digest() + return int.from_bytes(h[:4], "big") + + +def _hadamard_rotation(n: int, seed: int, tag: str) -> torch.Tensor: + """Sylvester-Hadamard × random sign diagonal → QR re-orthonormalize. + Deterministic in (seed, tag, n). Returns orthogonal R of shape (n, n) + such that R.T @ R == I (to QR precision ~2e-6).""" + key = (seed, tag, n) + if key in _SPINQUANT_CACHE: + return _SPINQUANT_CACHE[key] + p = 1 + while p < n: + p *= 2 + H = torch.ones(1, 1) + while H.shape[0] < p: + H = torch.cat( + [torch.cat([H, H], dim=1), torch.cat([H, -H], dim=1)], + dim=0, + ) + H = H / math.sqrt(p) + g = torch.Generator().manual_seed(_stable_seed(seed, tag)) + D = torch.diag(torch.randint(0, 2, (p,), generator=g).float() * 2 - 1) + R = (D @ H)[:n, :n] + Q, _ = torch.linalg.qr(R) + _SPINQUANT_CACHE[key] = Q + return Q + + +def _parse_spinquant_sites(h) -> "frozenset[str]": + """Parse h.spinquant_sites (comma-separated) into a frozenset of tag names. + Unknown tags are silently dropped (with a log). Returns empty set if the + env var is empty or 'none'.""" + raw = getattr(h, "spinquant_sites", "attn_in,attn_proj_in,mlp_in,mlp_proj_in") + if raw.strip().lower() in ("", "none"): + return frozenset() + valid = {"attn_in", "attn_proj_in", "mlp_in", "mlp_proj_in"} + tags = {t.strip() for t in raw.split(",") if t.strip()} + unknown = tags - valid + if unknown: + log(f"spinquant:unknown_sites_ignored:{sorted(unknown)}") + return frozenset(tags & valid) + + +def install_spinquant_rotations(model, h, seed=None, log_fn=print) -> int: + """Install the global rotation buffers on every CausalSelfAttention and MLP + in `model`, restricted to sites in h.spinquant_sites. Buffers are + non-persistent (regenerated deterministically at load). Returns number of + modules touched. Does NOT flip CastedLinear._sq_active — caller does that + after banks have been loaded with rotated weights.""" + if seed is None: + seed = int(os.environ.get("SPINQUANT_SEED", "42")) + sites = _parse_spinquant_sites(h) + model_dim = h.model_dim + hidden_dim = int(h.mlp_mult * h.model_dim) + # Generate rotations only for sites we're actually installing. + R_attn_in = _hadamard_rotation(model_dim, seed, "attn_in") if "attn_in" in sites else None + R_attn_proj_in = _hadamard_rotation(model_dim, seed, "attn_proj_in") if "attn_proj_in" in sites else None + R_mlp_in = _hadamard_rotation(model_dim, seed, "mlp_in") if "mlp_in" in sites else None + R_mlp_proj_in = _hadamard_rotation(hidden_dim, seed, "mlp_proj_in") if "mlp_proj_in" in sites else None + try: + device = next(model.parameters()).device + except StopIteration: + device = torch.device("cpu") + touched = 0 + for m in model.modules(): + if isinstance(m, CausalSelfAttention): + if R_attn_in is not None: + m.register_buffer("_sq_R_attn_in", R_attn_in.to(device), persistent=False) + if R_attn_proj_in is not None: + m.register_buffer("_sq_R_attn_proj_in", R_attn_proj_in.to(device), persistent=False) + touched += 1 + elif isinstance(m, MLP): + if R_mlp_in is not None: + m.register_buffer("_sq_R_mlp_in", R_mlp_in.to(device), persistent=False) + if R_mlp_proj_in is not None: + m.register_buffer("_sq_R_mlp_proj_in", R_mlp_proj_in.to(device), persistent=False) + touched += 1 + log_fn( + f"spinquant:installed_rotations:{touched}_modules seed:{seed} " + f"sites:{sorted(sites)} model_dim:{model_dim} hidden_dim:{hidden_dim}" + ) + return touched + + +# Which globally-shared rotation applies to each flat state_dict key suffix. +# Only the 6 attn/mlp bank weights are rotated in V1. +_SQ_KEY_TO_TAG: "dict[str, str]" = { + ".attn.c_q.weight": "attn_in", + ".attn.c_k.weight": "attn_in", + ".attn.c_v.weight": "attn_in", + ".attn.proj.weight": "attn_proj_in", + ".mlp.fc.weight": "mlp_in", + ".mlp.proj.weight": "mlp_proj_in", +} + + +def _spinquant_rotate_sd_and_H(sd_cpu: dict, hessians: dict, h, log_fn=print) -> None: + """In-place: rotate the 6 canonical flat weights and their matching Hessians. + Must be called AFTER collect_hessians() returns (so H is collected on + unrotated activations) and BEFORE gptq_mixed_quantize() consumes them. + + Math: + x_rot = x @ R + W_rot = W @ R (W is [out, in], R is [in, in]) + H_rot = R.T @ (x.T @ x) @ R = R.T @ H @ R + + After this call, F.linear(x_rot, W_rot) == F.linear(x, W) exactly (to fp + precision), so GPTQ quantizing W_rot with H_rot is mathematically matched. + + h.spinquant_sites filters which of the 4 tags get rotated; tags not in the + set are skipped here AND in install_spinquant_rotations, keeping the two + code paths in sync.""" + seed = h.spinquant_seed + sites = _parse_spinquant_sites(h) + tag_to_R: "dict[str, torch.Tensor]" = {} + + def _R_for(tag: str, in_dim: int) -> torch.Tensor: + if tag not in tag_to_R: + tag_to_R[tag] = _hadamard_rotation(in_dim, seed, tag).float().cpu() + return tag_to_R[tag] + + baked_weights = 0 + baked_hessians = 0 + missing_hessian = 0 + skipped_by_sites = 0 + for name in list(sd_cpu.keys()): + tag = None + for suffix, t in _SQ_KEY_TO_TAG.items(): + if name.endswith(suffix) and name.startswith("blocks."): + tag = t + break + if tag is None: + continue + if tag not in sites: + skipped_by_sites += 1 + continue + W = sd_cpu[name] + if W.ndim != 2: + continue + in_dim = W.shape[1] + R = _R_for(tag, in_dim) + assert R.shape == (in_dim, in_dim), ( + f"spinquant: R shape {tuple(R.shape)} != ({in_dim},{in_dim}) " + f"for {name} tag={tag}" + ) + orig_dtype = W.dtype + sd_cpu[name] = (W.float() @ R).to(orig_dtype).contiguous() + baked_weights += 1 + + if name in hessians: + H = hessians[name] + assert H.shape == (in_dim, in_dim), ( + f"spinquant: H shape {tuple(H.shape)} != ({in_dim},{in_dim}) for {name}" + ) + H_dev = H.device + H32 = H.float().cpu() + hessians[name] = (R.T @ H32 @ R).to(H.dtype).to(H_dev) + baked_hessians += 1 + else: + missing_hessian += 1 + + log_fn( + f"spinquant:rotated_weights:{baked_weights} hessians:{baked_hessians} " + f"missing_hessians:{missing_hessian} skipped_by_sites:{skipped_by_sites} " + f"active_sites:{sorted(sites)}" + ) + + + +def _compressed_code_size(code): + code_raw = code.encode("utf-8") + minified = subprocess.run( + ["pyminify", "--no-rename-locals", "--no-hoist-literals", "--remove-literal-statements", "-"], + input=code_raw, capture_output=True, check=True, + ).stdout + compressed = lzma.compress(minified) + encoded = base64.b85encode(compressed) + wrapper = b'import lzma as L,base64 as B\nexec(L.decompress(B.b85decode("' + encoded + b'")))\n' + return len(code_raw), len(wrapper) + + +def serialize(h, base_model, code): + if h.is_main_process: + torch.save(base_model.state_dict(), h.model_path) + model_bytes = os.path.getsize(h.model_path) + log(f"Serialized model: {model_bytes} bytes") + try: + code_bytes_uncompressed, code_bytes = _compressed_code_size(code) + except (FileNotFoundError, subprocess.CalledProcessError) as e: + code_bytes_uncompressed, code_bytes = len(code.encode("utf-8")), 0 + log(f"pyminify unavailable ({type(e).__name__}); skipping compressed-code-size measurement") + if h.is_main_process: + log(f"Code size (uncompressed): {code_bytes_uncompressed} bytes") + log(f"Code size (compressed): {code_bytes} bytes") + sd_cpu = _unbank_state_dict(base_model.state_dict(), h.num_layers) + device = torch.device("cuda", h.local_rank) + t0 = time.perf_counter() + calib_loader = ShuffledSequenceLoader(h, device) + log("GPTQ:collecting Hessians from calibration data...") + hessians = collect_hessians( + base_model, + calib_loader, + h, + device, + n_calibration_batches=h.gptq_calibration_batches, + ) + log(f"GPTQ:collected {len(hessians)} Hessians in {time.perf_counter()-t0:.1f}s") + # SpinQuant V1 bake: rotate weights W ← W @ R and Hessians H ← R.T H R. + # Runs AFTER Hessian collection (so H was measured on unrotated activations) + # and BEFORE GPTQ (so the quantizer sees the rotated frame end-to-end). + if getattr(h, "spinquant_enabled", False): + _spinquant_rotate_sd_and_H(sd_cpu, hessians, h, log_fn=log) + quant_result, quant_meta = gptq_mixed_quantize(sd_cpu, hessians, h) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = _compress(quant_raw, h.compressor) + quant_file_bytes = len(quant_blob) + bytes_total = quant_file_bytes + code_bytes + if h.is_main_process: + with open(h.quantized_model_path, "wb") as f: + f.write(quant_blob) + log(f"Serialized model quantized+{h.compressor}: {quant_file_bytes} bytes") + log(f"Total submission size quantized+{h.compressor}: {bytes_total} bytes") + return bytes_total, quant_file_bytes + + +def deserialize(h, device): + eval_model = GPT(h).to(device).bfloat16() + restore_fp32_params(eval_model) + flat_template = _unbank_state_dict(eval_model.state_dict(), h.num_layers) + with open(h.quantized_model_path, "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(_decompress(quant_blob_disk, h.compressor)), map_location="cpu" + ) + deq_flat = dequantize_mixed(quant_state["w"], quant_state["m"], flat_template) + head_dim = h.model_dim // h.num_heads + kv_dim = h.num_kv_heads * head_dim + hidden_dim = int(h.mlp_mult * h.model_dim) + deq_state = _rebank_state_dict(deq_flat, h.num_layers, h.model_dim, kv_dim, hidden_dim) + eval_model.load_state_dict(deq_state, strict=True) + # SpinQuant V1: banks now hold rotated weights (W @ R). Install the matching + # R buffers and flip the class-level flag so the forward rotation hooks fire. + # Math: F.linear(x @ R, W @ R) == F.linear(x, W) exactly (to fp precision). + if getattr(h, "spinquant_enabled", False): + install_spinquant_rotations(eval_model, h, seed=h.spinquant_seed, log_fn=log) + CastedLinear._sq_active = True + log(f"spinquant:_sq_active=True (forward rotations armed)") + return eval_model + + +def _loss_bpb(loss_sum, token_count, byte_count): + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_count.item()) + return val_loss, val_bpb + + +def eval_val(h, device, val_data, model, forward_logits_fn=None): + seq_len = h.eval_seq_len + local_batch_tokens = h.val_batch_tokens // (h.world_size * h.grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + f"VAL_BATCH_SIZE must provide at least one sequence per rank; got VAL_BATCH_SIZE={h.val_batch_tokens}, WORLD_SIZE={h.world_size}, GRAD_ACCUM_STEPS={h.grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_data.val_tokens.numel() - 1) // seq_len + seq_start = total_seqs * h.rank // h.world_size + seq_end = total_seqs * (h.rank + 1) // h.world_size + + # TODO: Don't truncate this. + seq_end = seq_start + ((seq_end - seq_start) // local_batch_seqs) * local_batch_seqs + + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + run_forward_logits = ( + (model.module.forward_logits if hasattr(model, "module") else model.forward_logits) + if forward_logits_fn is None + else forward_logits_fn + ) + model.eval() + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + with torch.no_grad(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_data.val_tokens[raw_start:raw_end].to( + device=device, dtype=torch.int64, non_blocking=True + ) + x = local[:-1] + y = local[1:] + bos_pos = (x == BOS_ID).nonzero(as_tuple=True)[0].tolist() + cu_seqlens, max_seqlen = _build_cu_seqlens( + bos_pos, x.numel(), x.device, h.eval_seq_len, 64 + ) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + logits = run_forward_logits( + x[None], cu_seqlens=cu_seqlens, max_seqlen=max_seqlen + ).detach() + per_token_loss = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y.reshape(-1), + reduction="none", + ) + val_loss_sum += per_token_loss.to(torch.float64).sum() + val_token_count += float(y.numel()) + prev_ids = x + tgt_ids = y + if val_data.caseops_enabled and val_data.val_bytes is not None: + # CaseOps: read per-token byte budget from sidecar at the same + # global positions as the target tokens y. raw_start/raw_end + # span [raw_start, raw_end), x = local[:-1], y = local[1:], + # so y is at sidecar positions [raw_start + 1, raw_end). + sidecar_slice = val_data.val_bytes[raw_start + 1 : raw_end].to( + device=device, dtype=torch.int32, non_blocking=True + ) + val_byte_count += sidecar_slice.to(torch.float64).sum() + else: + token_bytes = val_data.base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += ( + val_data.has_leading_space_lut[tgt_ids] + & ~val_data.is_boundary_token_lut[prev_ids] + ).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + model.train() + return _loss_bpb(val_loss_sum, val_token_count, val_byte_count) + + +def _find_docs(all_tokens): + bos_positions = (all_tokens == BOS_ID).nonzero(as_tuple=True)[0].numpy() + docs = [] + for i in range(len(bos_positions)): + start = int(bos_positions[i]) + end = ( + int(bos_positions[i + 1]) + if i + 1 < len(bos_positions) + else all_tokens.numel() + ) + if i + 1 < len(bos_positions): + end += 1 + assert end - start >= 2 + docs.append((start, end - start)) + return docs + + +def _build_ttt_global_batches(doc_entries, h, ascending=False): + batch_size = h.ttt_batch_size + global_doc_entries = sorted(doc_entries, key=lambda x: x[1][1]) + global_batches = [ + global_doc_entries[i : i + batch_size] + for i in range(0, len(global_doc_entries), batch_size) + ] + indexed = list(enumerate(global_batches)) + if not ascending: + indexed.sort(key=lambda ib: -max(dl for _, (_, dl) in ib[1])) + return indexed + + +def _init_batch_counter(path): + with open(path, "wb") as f: + f.write((0).to_bytes(4, "little")) + + +def _claim_next_batch(counter_path, queue_len): + try: + with open(counter_path, "r+b") as f: + fcntl.flock(f, fcntl.LOCK_EX) + idx = int.from_bytes(f.read(4), "little") + f.seek(0) + f.write((idx + 1).to_bytes(4, "little")) + f.flush() + except FileNotFoundError: + return queue_len + return idx + + +def _compute_chunk_window(ci, pred_len, num_chunks, chunk_size, eval_seq_len): + chunk_end = pred_len if ci == num_chunks - 1 else (ci + 1) * chunk_size + win_start = max(0, chunk_end - eval_seq_len) + win_len = chunk_end - win_start + chunk_start = ci * chunk_size + chunk_offset = chunk_start - win_start + chunk_len = chunk_end - chunk_start + return win_start, win_len, chunk_offset, chunk_len + + +def _accumulate_bpb( + ptl, + x, + y, + chunk_offsets, + chunk_lens, + pos_idx, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + loss_sum, + byte_sum, + token_count, + y_bytes=None, +): + pos = pos_idx[: x.size(1)].unsqueeze(0) + mask = ( + (chunk_lens.unsqueeze(1) > 0) + & (pos >= chunk_offsets.unsqueeze(1)) + & (pos < (chunk_offsets + chunk_lens).unsqueeze(1)) + ) + mask_f64 = mask.to(torch.float64) + if y_bytes is not None: + tok_bytes = y_bytes.to(torch.float64) + else: + tok_bytes = base_bytes_lut[y].to(torch.float64) + tok_bytes += (has_leading_space_lut[y] & ~is_boundary_token_lut[x]).to( + torch.float64 + ) + loss_sum += (ptl.to(torch.float64) * mask_f64).sum() + byte_sum += (tok_bytes * mask_f64).sum() + token_count += chunk_lens.to(torch.float64).sum() + + +def _loss_bpb_from_sums(loss_sum, token_count, byte_sum): + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_sum.item()) + return val_loss, val_bpb + + +def _add_to_counter(path, delta): + try: + with open(path, "r+b") as f: + fcntl.flock(f, fcntl.LOCK_EX) + cur = int.from_bytes(f.read(8), "little", signed=True) + cur += int(delta) + f.seek(0) + f.write(int(cur).to_bytes(8, "little", signed=True)) + f.flush() + return cur + except FileNotFoundError: + return int(delta) + + +def _init_int64_counter(path): + with open(path, "wb") as f: + f.write((0).to_bytes(8, "little", signed=True)) + + +def _select_ttt_doc_entries(docs, h): + doc_entries = list(enumerate(docs)) + if h.val_doc_fraction < 1.0: + sample_n = max(1, int(round(len(docs) * h.val_doc_fraction))) + sampled_indices = sorted( + random.Random(h.seed).sample(range(len(docs)), sample_n) + ) + return [(i, docs[i]) for i in sampled_indices] + return doc_entries + + +def train_val_ttt_global_sgd_distributed(h, device, val_data, base_model, val_tokens, batch_seqs=None): + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + base_model.eval() + seq_len = h.eval_seq_len + total_tokens = val_tokens.numel() - 1 + ttt_chunk = h.global_ttt_chunk_tokens + batch_seqs = h.global_ttt_batch_seqs if batch_seqs is None else batch_seqs + num_chunks = (total_tokens + ttt_chunk - 1) // ttt_chunk + ttt_params = [p for p in base_model.parameters()] + for p in ttt_params: + p.requires_grad_(True) + optimizer = torch.optim.SGD( + ttt_params, lr=h.global_ttt_lr, momentum=h.global_ttt_momentum + ) + t_start = time.perf_counter() + for ci in range(num_chunks): + chunk_start = ci * ttt_chunk + chunk_end = min((ci + 1) * ttt_chunk, total_tokens) + is_last_chunk = ci == num_chunks - 1 + if is_last_chunk or h.global_ttt_epochs <= 0: + continue + base_model.train() + chunk_seqs = (chunk_end - chunk_start) // seq_len + if chunk_seqs <= 0: + continue + warmup_chunks = max(0, min(h.global_ttt_warmup_chunks, num_chunks - 1)) + if warmup_chunks > 0 and ci < warmup_chunks: + warmup_denom = max(warmup_chunks - 1, 1) + warmup_t = ci / warmup_denom + lr_now = ( + h.global_ttt_warmup_start_lr + + (h.global_ttt_lr - h.global_ttt_warmup_start_lr) * warmup_t + ) + else: + decay_steps = max(num_chunks - 1 - warmup_chunks, 1) + decay_ci = max(ci - warmup_chunks, 0) + lr_now = h.global_ttt_lr * 0.5 * ( + 1.0 + math.cos(math.pi * decay_ci / decay_steps) + ) + for pg in optimizer.param_groups: + pg["lr"] = lr_now + my_seq_s = chunk_seqs * h.rank // h.world_size + my_seq_e = chunk_seqs * (h.rank + 1) // h.world_size + my_chunk_seqs = my_seq_e - my_seq_s + for _ in range(h.global_ttt_epochs): + for bs in range(0, my_chunk_seqs, batch_seqs): + be = min(bs + batch_seqs, my_chunk_seqs) + actual_bs = my_seq_s + bs + start_tok = chunk_start + actual_bs * seq_len + end_tok = chunk_start + (my_seq_s + be) * seq_len + 1 + if end_tok > val_tokens.numel(): + continue + local = val_tokens[start_tok:end_tok].to(device=device, dtype=torch.int64) + x_flat = local[:-1] + y_flat = local[1:] + optimizer.zero_grad(set_to_none=True) + with torch.enable_grad(): + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + if h.global_ttt_respect_doc_boundaries: + bos_pos = (x_flat == BOS_ID).nonzero(as_tuple=True)[0].tolist() + cu_seqlens, max_seqlen = _build_cu_seqlens( + bos_pos, x_flat.numel(), x_flat.device, h.eval_seq_len, 64 + ) + loss = base_model( + x_flat[None], + y_flat[None], + cu_seqlens=cu_seqlens, + max_seqlen=max_seqlen, + ) + else: + x = x_flat.reshape(-1, seq_len) + y = y_flat.reshape(-1, seq_len) + loss = base_model(x, y) + loss.backward() + if dist.is_available() and dist.is_initialized(): + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.SUM) + p.grad.mul_(1.0 / h.world_size) + if h.global_ttt_grad_clip > 0: + torch.nn.utils.clip_grad_norm_(ttt_params, h.global_ttt_grad_clip) + optimizer.step() + base_model.eval() + if h.rank == 0: + elapsed = time.perf_counter() - t_start + log( + f"tttg: c{ci+1}/{num_chunks} lr:{lr_now:.6f} t:{elapsed:.1f}s" + ) + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + + +def eval_val_ttt_phased(h, base_model, device, val_data, forward_ttt_train): + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + base_model.eval() + for p in base_model.parameters(): + p.requires_grad_(False) + all_tokens = val_data.val_tokens + all_tokens_idx = all_tokens.to(torch.int32) + docs = _find_docs(all_tokens) + doc_entries = _select_ttt_doc_entries(docs, h) + prefix_doc_limit = max(0, min(len(doc_entries), int(h.phased_ttt_prefix_docs))) + num_phases = max(1, int(h.phased_ttt_num_phases)) + phase_boundaries = [] + for pi in range(num_phases): + boundary = prefix_doc_limit * (pi + 1) // num_phases + phase_boundaries.append(boundary) + current_phase = 0 + current_phase_boundary = phase_boundaries[0] + log( + "ttt_phased:" + f" total_docs:{len(doc_entries)} prefix_docs:{prefix_doc_limit} " + f"suffix_docs:{len(doc_entries) - prefix_doc_limit}" + f" num_phases:{num_phases} boundaries:{phase_boundaries}" + ) + chunk_size, eval_seq_len = h.ttt_chunk_size, h.ttt_eval_seq_len + eval_batch_set = None + if h.ttt_eval_batches: + eval_batch_set = set(int(x) for x in h.ttt_eval_batches.split(",") if x.strip()) + use_ascending = eval_batch_set is not None + global_batches_sorted = _build_ttt_global_batches( + doc_entries, h, ascending=use_ascending + ) + queue_len = len(global_batches_sorted) + counter_path = f"/tmp/ttt_counter_{h.run_id}" + prefix_counter_path = f"/tmp/ttt_prefix_counter_{h.run_id}" + pause_flag_path = f"/tmp/ttt_pause_flag_{h.run_id}" + if h.rank == 0: + _init_batch_counter(counter_path) + _init_int64_counter(prefix_counter_path) + try: + os.remove(pause_flag_path) + except FileNotFoundError: + pass + if dist.is_available() and dist.is_initialized(): + path_list = [counter_path, prefix_counter_path, pause_flag_path] + dist.broadcast_object_list(path_list, src=0) + counter_path, prefix_counter_path, pause_flag_path = path_list + dist.barrier() + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + byte_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + t_start = time.perf_counter() + reusable_lora = BatchedTTTLoRA( + h.ttt_batch_size, base_model, h.ttt_lora_rank, alpha=h.ttt_lora_alpha, + k_lora=h.ttt_k_lora, mlp_lora=h.ttt_mlp_lora, o_lora=h.ttt_o_lora, + ).to(device) + + def _build_opt(lora): + if h.ttt_optimizer == "sgd": + return torch.optim.SGD( + lora.parameters(), lr=h.ttt_lora_lr, + momentum=h.ttt_beta1, weight_decay=h.ttt_weight_decay, + ) + return torch.optim.AdamW( + lora.parameters(), lr=h.ttt_lora_lr, + betas=(h.ttt_beta1, h.ttt_beta2), + eps=1e-10, weight_decay=h.ttt_weight_decay, fused=True, + ) + + reusable_opt = _build_opt(reusable_lora) + local_scored_docs = [] + global_ttt_done = prefix_doc_limit == 0 + try: + while True: + queue_idx = _claim_next_batch(counter_path, queue_len) + if queue_idx >= queue_len: + break + orig_batch_idx, batch_entries = global_batches_sorted[queue_idx] + batch = [doc for _, doc in batch_entries] + bsz = len(batch) + prev_loss = loss_sum.item() + prev_bytes = byte_sum.item() + prev_tokens = token_count.item() + if bsz == reusable_lora.bsz: + reusable_lora.reset() + for s in reusable_opt.state.values(): + for k, v in s.items(): + if isinstance(v, torch.Tensor): + v.zero_() + elif k == "step": + s[k] = 0 + cur_lora = reusable_lora + cur_opt = reusable_opt + else: + cur_lora = BatchedTTTLoRA( + bsz, base_model, h.ttt_lora_rank, alpha=h.ttt_lora_alpha, + k_lora=h.ttt_k_lora, mlp_lora=h.ttt_mlp_lora, o_lora=h.ttt_o_lora, + ).to(device) + cur_opt = _build_opt(cur_lora) + pred_lens = [doc_len - 1 for _, doc_len in batch] + num_chunks = [(pl + chunk_size - 1) // chunk_size for pl in pred_lens] + max_nc = max(num_chunks) + num_chunks_t = torch.tensor(num_chunks, dtype=torch.int64, device=device) + for ci in range(max_nc): + active = [ci < nc for nc in num_chunks] + needs_train = any(ci < nc - 1 for nc in num_chunks) + tok_starts = torch.zeros(bsz, dtype=torch.int64) + tok_wls = torch.zeros(bsz, dtype=torch.int64) + chunk_offsets_cpu = torch.zeros(bsz, dtype=torch.int64) + chunk_lens_cpu = torch.zeros(bsz, dtype=torch.int64) + for b in range(bsz): + if not active[b]: + continue + doc_start, doc_len = batch[b] + win_start, win_len, chunk_offset, chunk_len = _compute_chunk_window( + ci, pred_lens[b], num_chunks[b], chunk_size, eval_seq_len + ) + tok_starts[b] = doc_start + win_start + tok_wls[b] = win_len + chunk_offsets_cpu[b] = chunk_offset + chunk_lens_cpu[b] = chunk_len + _, context_size, chunk_offset, _ = _compute_chunk_window( + ci, (ci + 1) * chunk_size, ci + 1, chunk_size, eval_seq_len + ) + col_idx = torch.arange(context_size + 1) + idx = tok_starts.unsqueeze(1) + col_idx.unsqueeze(0) + idx.clamp_(max=all_tokens.numel() - 1) + gathered_gpu = all_tokens_idx[idx].to( + device=device, dtype=torch.int64, non_blocking=True + ) + valid = (col_idx[:context_size].unsqueeze(0) < tok_wls.unsqueeze(1)).to( + device, non_blocking=True + ) + chunk_offsets = chunk_offsets_cpu.to(device, non_blocking=True) + chunk_lens = chunk_lens_cpu.to(device, non_blocking=True) + x = torch.where(valid, gathered_gpu[:, :context_size], 0) + y = torch.where(valid, gathered_gpu[:, 1 : context_size + 1], 0) + ctx_pos = torch.arange(context_size, device=device, dtype=torch.int64) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + per_tok_loss = forward_ttt_train(x, y, lora=cur_lora) + # CaseOps sidecar-driven byte budget. Mirror the index pattern + # used to build y from all_tokens: y[b, j] corresponds to the + # token at global position tok_starts[b] + 1 + j (when valid). + y_bytes_arg = None + if val_data.caseops_enabled and val_data.val_bytes is not None: + y_idx = ( + tok_starts.unsqueeze(1) + + 1 + + col_idx[:context_size].unsqueeze(0) + ) + y_idx = y_idx.clamp_(max=val_data.val_bytes.numel() - 1) + y_bytes_arg = val_data.val_bytes[y_idx].to( + device=device, dtype=torch.int32, non_blocking=True + ) + # Mirror the `valid` masking used for y so out-of-range tokens + # contribute zero bytes (matches y=0 substitution above). + y_bytes_arg = torch.where( + valid, y_bytes_arg, torch.zeros_like(y_bytes_arg) + ) + with torch.no_grad(): + _accumulate_bpb( + per_tok_loss, + x, + y, + chunk_offsets, + chunk_lens, + ctx_pos, + val_data.base_bytes_lut, + val_data.has_leading_space_lut, + val_data.is_boundary_token_lut, + loss_sum, + byte_sum, + token_count, + y_bytes=y_bytes_arg, + ) + if needs_train: + activate_chunk_mask = (num_chunks_t - 1 > ci).float() + for gi in range(h.ttt_grad_steps): + if gi > 0: + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + per_tok_loss = forward_ttt_train(x, y, lora=cur_lora) + per_doc = per_tok_loss[ + :, chunk_offset : chunk_offset + chunk_size + ].mean(dim=-1) + cur_opt.zero_grad(set_to_none=True) + (per_doc * activate_chunk_mask).sum().backward() + cur_opt.step() + else: + del per_tok_loss + batch_num = orig_batch_idx + 1 + doc_lens = [dl for _, dl in batch] + should_report = batch_num in eval_batch_set if eval_batch_set is not None else True + if should_report: + cur_tokens = token_count.item() + cur_loss_val = loss_sum.item() + cur_bytes_val = byte_sum.item() + dt = cur_tokens - prev_tokens + db = cur_bytes_val - prev_bytes + if dt > 0 and db > 0: + b_loss = (cur_loss_val - prev_loss) / dt + b_bpb = b_loss / math.log(2.0) * (dt / db) + else: + b_loss = b_bpb = 0.0 + r_loss = cur_loss_val / max(cur_tokens, 1) + r_bpb = r_loss / math.log(2.0) * (cur_tokens / max(cur_bytes_val, 1)) + elapsed = time.perf_counter() - t_start + log( + f"ttp: b{batch_num}/{queue_len} bl:{b_loss:.4f} bb:{b_bpb:.4f} " + f"rl:{r_loss:.4f} rb:{r_bpb:.4f} dl:{min(doc_lens)}-{max(doc_lens)} " + f"gd:{int(global_ttt_done)}" + ) + if not global_ttt_done: + local_scored_docs.extend( + (orig_batch_idx, pos, doc_start, doc_len) + for pos, (doc_start, doc_len) in enumerate(batch) + ) + prefix_done = _add_to_counter(prefix_counter_path, len(batch_entries)) + if prefix_done >= current_phase_boundary: + try: + with open(pause_flag_path, "x"): + pass + except FileExistsError: + pass + should_pause = os.path.exists(pause_flag_path) + if should_pause: + if dist.is_available() and dist.is_initialized(): + dist.barrier() + gathered_scored_docs = [None] * h.world_size + if dist.is_available() and dist.is_initialized(): + dist.all_gather_object(gathered_scored_docs, local_scored_docs) + else: + gathered_scored_docs = [local_scored_docs] + scored_docs_for_global = [] + for rank_docs in gathered_scored_docs: + if rank_docs: + scored_docs_for_global.extend(rank_docs) + scored_docs_for_global.sort(key=lambda x: (x[0], x[1])) + scored_docs_for_global = scored_docs_for_global[:current_phase_boundary] + scored_token_chunks = [ + val_data.val_tokens[doc_start : doc_start + doc_len] + for _, _, doc_start, doc_len in scored_docs_for_global + ] + if scored_token_chunks: + global_ttt_tokens = torch.cat(scored_token_chunks) + else: + global_ttt_tokens = val_data.val_tokens[:0] + if h.rank == 0: + prefix_done = 0 + try: + with open(prefix_counter_path, "rb") as f: + prefix_done = int.from_bytes( + f.read(8), "little", signed=True + ) + except FileNotFoundError: + pass + log( + f"ttpp: phase:{current_phase + 1}/{num_phases} pd:{prefix_done} " + f"gd:{len(scored_docs_for_global)} " + f"t:{time.perf_counter() - t_start:.1f}s" + ) + train_val_ttt_global_sgd_distributed( + h, device, val_data, base_model, global_ttt_tokens + ) + for p in base_model.parameters(): + p.requires_grad_(False) + reusable_lora = BatchedTTTLoRA( + h.ttt_batch_size, base_model, h.ttt_lora_rank, + k_lora=h.ttt_k_lora, mlp_lora=h.ttt_mlp_lora, o_lora=h.ttt_o_lora, + ).to(device) + reusable_opt = _build_opt(reusable_lora) + current_phase += 1 + if current_phase >= num_phases: + global_ttt_done = True + else: + current_phase_boundary = phase_boundaries[current_phase] + if h.rank == 0: + try: + os.remove(pause_flag_path) + except FileNotFoundError: + pass + if dist.is_available() and dist.is_initialized(): + dist.barrier() + if h.rank == 0: + log(f"ttpr: phase:{current_phase}/{num_phases} t:{time.perf_counter() - t_start:.1f}s") + del cur_lora, cur_opt + finally: + pass + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.train() + return _loss_bpb_from_sums(loss_sum, token_count, byte_sum) + + +def timed_eval(label, fn, *args, **kwargs): + torch.cuda.synchronize() + t0 = time.perf_counter() + val_loss, val_bpb = fn(*args, **kwargs) + torch.cuda.synchronize() + elapsed_ms = 1e3 * (time.perf_counter() - t0) + log( + f"{label} val_loss:{val_loss:.8f} val_bpb:{val_bpb:.8f} eval_time:{elapsed_ms:.0f}ms" + ) + return val_loss, val_bpb + + +def train_model(h, device, val_data): + base_model = GPT(h).to(device).bfloat16() + restore_fp32_params(base_model) + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + compiled_forward_logits = torch.compile( + base_model.forward_logits, dynamic=False, fullgraph=True + ) + model = compiled_model + log(f"model_params:{sum(p.numel()for p in base_model.parameters())}") + log( + f"recur_alpha: enabled={h.recur_alpha_enabled} " + f"num_loops={h.num_loops} loop_start={h.loop_start} loop_end={h.loop_end} " + f"diag_p2p_cos={h.recur_diag_p2p_cos}" + ) + optimizers = Optimizers(h, base_model) + train_loader = DocumentPackingLoader(h, device) + max_wallclock_ms = ( + 1e3 * h.max_wallclock_seconds if h.max_wallclock_seconds > 0 else None + ) + if max_wallclock_ms is not None: + max_wallclock_ms -= h.gptq_reserve_seconds * 1e3 + log( + f"gptq:reserving {h.gptq_reserve_seconds:.0f}s, effective={max_wallclock_ms:.0f}ms" + ) + + def training_frac(step, elapsed_ms): + if max_wallclock_ms is None: + return step / max(h.iterations, 1) + return elapsed_ms / max(max_wallclock_ms, 1e-09) + + def lr_mul(frac): + if h.warmdown_frac <= 0: + return 1.0 + if frac >= 1.0 - h.warmdown_frac: + return max((1.0 - frac) / h.warmdown_frac, h.min_lr) + return 1.0 + + def step_fn(step, lr_scale): + optimizers.zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(h.grad_accum_steps): + x, y, cu_seqlens, _max_seqlen = train_loader.next_batch( + h.train_batch_tokens, h.grad_accum_steps + ) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y, cu_seqlens=cu_seqlens, max_seqlen=h.train_seq_len) + train_loss += loss.detach() + (loss / h.grad_accum_steps).backward() + train_loss /= h.grad_accum_steps + frac = ( + min(step / h.muon_momentum_warmup_steps, 1.0) + if h.muon_momentum_warmup_steps > 0 + else 1.0 + ) + muon_momentum = ( + 1 - frac + ) * h.muon_momentum_warmup_start + frac * h.muon_momentum + for group in optimizers.optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * lr_scale + if h.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), h.grad_clip_norm) + # Snapshot α grad norm BEFORE optimizers.step() since step() ends with + # zero_grad_all(); reading recur_alpha.grad after step() always sees None/0. + # Spec 015 hit this bug — cosmetic (α values moved fine) but the logged + # grad_norm was unusable as a plumbing-check signal. + alpha_grad_norm = None # Spec 025b: recur_beta/recur_alpha are frozen buffers + optimizers.step(distributed=h.distributed) + return train_loss, alpha_grad_norm + + if h.warmup_steps > 0: + initial_model_state = { + name: tensor.detach().cpu().clone() + for (name, tensor) in base_model.state_dict().items() + } + initial_optimizer_states = [ + copy.deepcopy(opt.state_dict()) for opt in optimizers + ] + model.train() + num_tokens_local = h.train_batch_tokens // h.world_size + for blk in base_model.blocks: + blk.attn.rotary(num_tokens_local, device, torch.bfloat16) + cu_bucket_size = train_loader.cu_bucket_size + warmup_cu_buckets = tuple(cu_bucket_size * i for i in range(1, 5)) + warmup_cu_iters = 3 + x, y, cu_seqlens, _ = train_loader.next_batch( + h.train_batch_tokens, h.grad_accum_steps + ) + log(f"warmup_cu_buckets:{','.join(str(b) for b in warmup_cu_buckets)} iters_each:{warmup_cu_iters}") + def _run_cu_bucket_warmup(): + for bucket_len in warmup_cu_buckets: + boundaries = list(range(0, x.size(1), max(h.train_seq_len, 1))) + if boundaries[-1] != x.size(1): + boundaries.append(x.size(1)) + cu = torch.full((bucket_len,), x.size(1), dtype=torch.int32, device=device) + cu[: len(boundaries)] = torch.tensor(boundaries, dtype=torch.int32, device=device) + for _ in range(warmup_cu_iters): + optimizers.zero_grad_all() + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + wloss = model(x, y, cu_seqlens=cu, max_seqlen=h.train_seq_len) + (wloss / h.grad_accum_steps).backward() + optimizers.zero_grad_all() + _run_cu_bucket_warmup() + if h.num_loops > 0: + base_model.looping_active = True + _run_cu_bucket_warmup() + if h.loop_depth_upgrade_at > 0 and h.num_loops >= 2: + base_model.looping_depth = h.num_loops # pre-warm full-depth state + _run_cu_bucket_warmup() + base_model.looping_depth = h.num_loops - 1 # reset to curriculum start + base_model.looping_active = False + for warmup_step in range(h.warmup_steps): + step_fn(warmup_step, 1.0) + if ( + warmup_step <= 5 + or (warmup_step + 1) % 10 == 0 + or warmup_step + 1 == h.warmup_steps + ): + log(f"warmup_step: {warmup_step+1}/{h.warmup_steps}") + if h.num_loops > 0: + base_model.looping_active = True + log( + f"loop_warmup:enabled encoder:{base_model.encoder_indices} decoder:{base_model.decoder_indices}" + ) + for warmup_step in range(h.warmup_steps): + step_fn(warmup_step, 1.0) + if ( + warmup_step <= 5 + or (warmup_step + 1) % 10 == 0 + or warmup_step + 1 == h.warmup_steps + ): + log(f"loop_warmup_step: {warmup_step+1}/{h.warmup_steps}") + if h.loop_depth_upgrade_at > 0 and h.num_loops >= 2: + base_model.looping_depth = h.num_loops # pre-warm full-depth state + log(f"loop_warmup:depth_upgraded looping_depth:{h.num_loops + 1}") + for warmup_step in range(h.warmup_steps): + step_fn(warmup_step, 1.0) + if ( + warmup_step <= 5 + or (warmup_step + 1) % 10 == 0 + or warmup_step + 1 == h.warmup_steps + ): + log(f"loop_depth_warmup_step: {warmup_step+1}/{h.warmup_steps}") + base_model.looping_depth = h.num_loops - 1 # reset to curriculum start + base_model.looping_active = False + base_model.load_state_dict(initial_model_state, strict=True) + for (opt, state) in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + optimizers.zero_grad_all() + train_loader = DocumentPackingLoader(h, device) + ema_state = { + name: t.detach().float().clone() + for (name, t) in base_model.state_dict().items() + } + ema_decay = h.ema_decay + training_time_ms = 0.0 + stop_after_step = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = ( + step == h.iterations + or stop_after_step is not None + and step >= stop_after_step + ) + should_validate = ( + last_step or h.val_loss_every > 0 and step % h.val_loss_every == 0 + ) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1e3 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + h, device, val_data, model, compiled_forward_logits + ) + log( + f"{step}/{h.iterations} val_loss: {val_loss:.4f} val_bpb: {val_bpb:.4f}" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < h.iterations: + log( + f"stopping_early: wallclock_cap train_time: {training_time_ms:.0f}ms step: {step}/{h.iterations}" + ) + break + elapsed_ms = training_time_ms + 1e3 * (time.perf_counter() - t0) + frac = training_frac(step, elapsed_ms) + scale = lr_mul(frac) + if ( + h.num_loops > 0 + and not base_model.looping_active + and frac >= h.enable_looping_at + ): + base_model.looping_active = True + log( + f"layer_loop:enabled step:{step} frac:{frac:.3f} depth:{base_model.looping_depth + 1} encoder:{base_model.encoder_indices} decoder:{base_model.decoder_indices}" + ) + if ( + h.loop_depth_upgrade_at > 0 + and base_model.looping_active + and base_model.looping_depth < h.num_loops + and frac >= h.loop_depth_upgrade_at + ): + base_model.looping_depth = h.num_loops + log( + f"loop_depth:upgraded step:{step} frac:{frac:.3f} depth:{h.num_loops + 1} encoder:{base_model.encoder_indices} decoder:{base_model.decoder_indices}" + ) + train_loss, alpha_grad_norm = step_fn(step, scale) + with torch.no_grad(): + for (name, t) in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_( + t.detach().float(), alpha=1.0 - ema_decay + ) + step += 1 + approx_training_time_ms = training_time_ms + 1e3 * (time.perf_counter() - t0) + should_log_train = h.train_log_every > 0 and ( + step <= 5 or step % h.train_log_every == 0 or stop_after_step is not None + ) + if should_log_train: + tok_per_sec = step * h.train_batch_tokens / (approx_training_time_ms / 1e3) + log( + f"{step}/{h.iterations} train_loss: {train_loss.item():.4f} train_time: {approx_training_time_ms/60000:.1f}m tok/s: {tok_per_sec:.0f}" + ) + # Spec 015/016: Recur-Alpha diagnostics (alpha values, grad norms, p2p cosine). + # alpha_grad_norm comes from step_fn (snapshotted before optimizers.step() + # zeros grads); reading base_model.recur_alpha.grad here would always be None. + if getattr(base_model, "recur_alpha", None) is not None: + alpha_tensor = base_model.recur_alpha.detach().float().cpu().tolist() + beta_tensor = None + if getattr(base_model, "recur_beta", None) is not None: + beta_tensor = base_model.recur_beta.detach().float().cpu().tolist() + if alpha_grad_norm is None: + alpha_grad_norm = 0.0 + p2p_cos_str = "" + if getattr(base_model, "recur_diag_p2p_cos", False): + p2p = base_model._diag_p2p_cos.detach().float().cpu().tolist() + p2p_cos_str = f" p2p_cos: {p2p}" + log( + f"recur_alpha: beta={beta_tensor} alpha={alpha_tensor} grad_norm={alpha_grad_norm:.6f}{p2p_cos_str}" + ) + reached_cap = ( + max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + ) + if h.distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log( + f"peak memory allocated: {torch.cuda.max_memory_allocated()//1024//1024} MiB reserved: {torch.cuda.max_memory_reserved()//1024//1024} MiB" + ) + log("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = { + name: t.to(dtype=current_state[name].dtype) for (name, t) in ema_state.items() + } + base_model.load_state_dict(avg_state, strict=True) + return base_model, compiled_model, compiled_forward_logits + + +def train_and_eval(h, device): + random.seed(h.seed) + np.random.seed(h.seed) + torch.manual_seed(h.seed) + torch.cuda.manual_seed_all(h.seed) + if h.artifact_dir and h.is_main_process: + os.makedirs(h.artifact_dir, exist_ok=True) + val_data = ValidationData(h, device) + log( + f"train_shards: {len(list(Path(h.datasets_dir).resolve().glob('fineweb_train_*.bin')))}" + ) + log(f"val_tokens: {val_data.val_tokens.numel()-1}") + base_model, compiled_model, compiled_forward_logits = train_model( + h, device, val_data + ) + torch._dynamo.reset() + timed_eval( + "diagnostic pre-quantization post-ema", + eval_val, + h, + device, + val_data, + compiled_model, + compiled_forward_logits, + ) + serialize(h, base_model, Path(__file__).read_text(encoding="utf-8")) + if h.distributed: + dist.barrier() + eval_model = deserialize(h, device) + if h.num_loops > 0: + eval_model.looping_active = True + eval_model.looping_depth = h.num_loops # always full depth at eval/TTT + compiled_model = torch.compile(eval_model, dynamic=False, fullgraph=True) + compiled_forward_logits = torch.compile( + eval_model.forward_logits, dynamic=False, fullgraph=True + ) + timed_eval( + "diagnostic quantized", + eval_val, + h, + device, + val_data, + compiled_model, + compiled_forward_logits, + ) + if h.ttt_enabled: + del eval_model, compiled_model + torch._dynamo.reset() + torch.cuda.empty_cache() + ttt_model = deserialize(h, device) + if h.num_loops > 0: + ttt_model.looping_active = True + ttt_model.looping_depth = h.num_loops # always full depth at TTT + for p in ttt_model.parameters(): + p.requires_grad_(False) + + if h.rope_yarn: + _yarn_seqlen = h.train_batch_tokens // h.grad_accum_steps + for block in ttt_model.blocks: + block.attn.rotary(_yarn_seqlen, device, torch.bfloat16) + else: + for block in ttt_model.blocks: + block.attn.rotary._cos_cached = None + block.attn.rotary._sin_cached = None + block.attn.rotary._seq_len_cached = 0 + block.attn.rotary(h.ttt_eval_seq_len, device, torch.bfloat16) + + def _fwd_ttt_inner(input_ids, target_ids, lora): + return ttt_model.forward_ttt(input_ids, target_ids, lora=lora) + + _fwd_ttt_compiled_inner = None + + def _fwd_ttt(input_ids, target_ids, lora): + nonlocal _fwd_ttt_compiled_inner + if _fwd_ttt_compiled_inner is None: + _fwd_ttt_compiled_inner = torch.compile(_fwd_ttt_inner, dynamic=True) + return _fwd_ttt_compiled_inner(input_ids, target_ids, lora=lora) + + fwd_ttt_compiled = _fwd_ttt + log(f"ttt_lora:warming up compile (random tokens, no val data)") + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + t_warmup = time.perf_counter() + warmup_bszes = [h.ttt_batch_size] + for bsz in warmup_bszes: + wl = BatchedTTTLoRA( + bsz, ttt_model, h.ttt_lora_rank, alpha=h.ttt_lora_alpha, + k_lora=h.ttt_k_lora, mlp_lora=h.ttt_mlp_lora, o_lora=h.ttt_o_lora, + ).to(device) + wo = torch.optim.AdamW( + wl.parameters(), + lr=h.ttt_lora_lr, + betas=(h.ttt_beta1, h.ttt_beta2), + eps=1e-10, + weight_decay=h.ttt_weight_decay, + fused=True, + ) + for ctx_len in (h.ttt_chunk_size, h.ttt_eval_seq_len): + xw = torch.randint(0, h.vocab_size, (bsz, ctx_len), device=device, dtype=torch.int64) + yw = torch.randint(0, h.vocab_size, (bsz, ctx_len), device=device, dtype=torch.int64) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + ptl = fwd_ttt_compiled(xw, yw, lora=wl) + ptl[:, : min(h.ttt_chunk_size, ctx_len)].mean(dim=-1).sum().backward() + wo.step() + wo.zero_grad(set_to_none=True) + del wl, wo + torch.cuda.empty_cache() + compile_elapsed = time.perf_counter() - t_warmup + log(f"ttt_lora:compile warmup done ({compile_elapsed:.1f}s)") + log("\nbeginning TTT eval timer") + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_val_loss, ttt_val_bpb = eval_val_ttt_phased( + h, ttt_model, device, val_data, forward_ttt_train=fwd_ttt_compiled + ) + torch.cuda.synchronize() + ttt_eval_elapsed = time.perf_counter() - t_ttt + log( + "quantized_ttt_phased " + f"val_loss:{ttt_val_loss:.8f} val_bpb:{ttt_val_bpb:.8f} " + f"eval_time:{1e3*ttt_eval_elapsed:.0f}ms" + ) + log(f"total_eval_time:{ttt_eval_elapsed:.1f}s") + del ttt_model + + +def main(): + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError( + f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral" + ) + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + torch.set_float32_matmul_precision("high") + from torch.backends.cuda import ( + enable_cudnn_sdp, + enable_flash_sdp, + enable_math_sdp, + enable_mem_efficient_sdp, + ) + + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + torch._dynamo.config.optimize_ddp = False + torch._dynamo.config.cache_size_limit = 16 + h = Hyperparameters() + set_logging_hparams(h) + if h.is_main_process: + os.makedirs(h.artifact_dir if h.artifact_dir else "logs", exist_ok=True) + log(100 * "=", console=False) + log("Hyperparameters:", console=True) + for (k, v) in sorted(vars(type(h)).items()): + if not k.startswith("_"): + log(f" {k}: {v}", console=True) + log("=" * 100, console=False) + log("Source code:", console=False) + log("=" * 100, console=False) + with open(__file__, "r", encoding="utf-8") as _src: + log(_src.read(), console=False) + log("=" * 100, console=False) + log(f"Running Python {sys.version}", console=False) + log(f"Running PyTorch {torch.__version__}", console=False) + log("=" * 100, console=False) + train_and_eval(h, device) + if distributed: + dist.destroy_process_group() + + +if __name__ == "__main__": + main() diff --git a/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed0.log b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed0.log new file mode 100644 index 0000000000..e765670e18 --- /dev/null +++ b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed0.log @@ -0,0 +1,932 @@ +W0423 23:04:49.690000 183720 torch/distributed/run.py:803] +W0423 23:04:49.690000 183720 torch/distributed/run.py:803] ***************************************** +W0423 23:04:49.690000 183720 torch/distributed/run.py:803] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0423 23:04:49.690000 183720 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: /workspace/runs/036-035e-8h-promotion/seed_0 + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + beta1: 0.9 + beta2: 0.95 + caseops_enabled: True + compressor: brotli + data_dir: /workspace/parameter-golf/data + datasets_dir: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2048 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.005 + gated_attn_quant_gate: True + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 0.5 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: /workspace/runs/036-035e-8h-promotion/seed_0/b1427e96-6f17-4287-90c9-7028df16b516.txt + logit_softcap: 30.0 + loop_depth_upgrade_at: 0.0 + loop_end: 5 + loop_start: 3 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: /workspace/runs/036-035e-8h-promotion/seed_0/final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 3 + phased_ttt_prefix_docs: 2000 + qk_gain_init: 5.0 + quantized_model_path: /workspace/runs/036-035e-8h-promotion/seed_0/final_model.int6.ptz + rank: 0 + recur_alpha_enabled: True + recur_diag_p2p_cos: False + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: b1427e96-6f17-4287-90c9-7028df16b516 + scalar_lr: 0.02 + seed: 0 + skip_gates_enabled: True + smear_gate_enabled: False + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 1.0 + spinquant_enabled: False + spinquant_seed: 42 + spinquant_sites: attn_in,attn_proj_in,mlp_in,mlp_proj_in + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_train_*.bin + train_log_every: 100 + train_seq_len: 2048 + ttt_batch_size: 64 + ttt_beta1: 0.0 + ttt_beta2: 0.999 + ttt_chunk_size: 48 + ttt_enabled: True + ttt_eval_batches: + ttt_eval_seq_len: 2048 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_alpha: 144 + ttt_lora_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 1.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 0 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 47851520 +model_params:35945658 +recur_alpha: enabled=True num_loops=2 loop_start=3 loop_end=5 diag_p2p_cos=False +gptq:reserving 0s, effective=599500ms +warmup_cu_buckets:64,128,192,256 iters_each:3 +warmup_step: 1/20 +warmup_step: 2/20 +warmup_step: 3/20 +warmup_step: 4/20 +warmup_step: 5/20 +warmup_step: 6/20 +warmup_step: 10/20 +warmup_step: 20/20 +loop_warmup:enabled encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +loop_warmup_step: 1/20 +loop_warmup_step: 2/20 +loop_warmup_step: 3/20 +loop_warmup_step: 4/20 +loop_warmup_step: 5/20 +loop_warmup_step: 6/20 +loop_warmup_step: 10/20 +loop_warmup_step: 20/20 +1/20000 train_loss: 9.0086 train_time: 0.0m tok/s: 12585054 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2/20000 train_loss: 12.8192 train_time: 0.0m tok/s: 10765241 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3/20000 train_loss: 10.2218 train_time: 0.0m tok/s: 9413340 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4/20000 train_loss: 8.7237 train_time: 0.0m tok/s: 8896976 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +5/20000 train_loss: 7.9377 train_time: 0.0m tok/s: 8616992 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +100/20000 train_loss: 3.6015 train_time: 0.2m tok/s: 8456801 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +200/20000 train_loss: 3.1515 train_time: 0.3m tok/s: 8328127 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +300/20000 train_loss: 2.9221 train_time: 0.5m tok/s: 8302802 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +400/20000 train_loss: 2.5910 train_time: 0.6m tok/s: 8275445 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +500/20000 train_loss: 2.5756 train_time: 0.8m tok/s: 8303175 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +600/20000 train_loss: 2.6776 train_time: 1.0m tok/s: 8257307 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +700/20000 train_loss: 2.8786 train_time: 1.1m tok/s: 8244885 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +800/20000 train_loss: 2.7167 train_time: 1.3m tok/s: 8242585 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +900/20000 train_loss: 2.7614 train_time: 1.4m tok/s: 8245578 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1000/20000 train_loss: 2.8115 train_time: 1.6m tok/s: 8261910 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1100/20000 train_loss: 2.7714 train_time: 1.7m tok/s: 8263243 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1200/20000 train_loss: 2.7718 train_time: 1.9m tok/s: 8263986 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1300/20000 train_loss: 2.8340 train_time: 2.1m tok/s: 8267027 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1400/20000 train_loss: 2.5954 train_time: 2.2m tok/s: 8268520 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1500/20000 train_loss: 2.6376 train_time: 2.4m tok/s: 8278537 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1600/20000 train_loss: 2.7120 train_time: 2.5m tok/s: 8277956 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1700/20000 train_loss: 2.6823 train_time: 2.7m tok/s: 8277224 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1800/20000 train_loss: 2.6502 train_time: 2.9m tok/s: 8274668 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1900/20000 train_loss: 2.7465 train_time: 3.0m tok/s: 8282405 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2000/20000 train_loss: 2.6702 train_time: 3.2m tok/s: 8277917 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2100/20000 train_loss: 2.6933 train_time: 3.3m tok/s: 8276382 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2200/20000 train_loss: 2.5350 train_time: 3.5m tok/s: 8274428 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +layer_loop:enabled step:2208 frac:0.350 depth:3 encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +2300/20000 train_loss: 2.6093 train_time: 3.7m tok/s: 8120218 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2400/20000 train_loss: 2.6319 train_time: 3.9m tok/s: 7976413 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2500/20000 train_loss: 2.5590 train_time: 4.2m tok/s: 7840992 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2600/20000 train_loss: 2.5264 train_time: 4.4m tok/s: 7723683 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2700/20000 train_loss: 2.5126 train_time: 4.6m tok/s: 7619317 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2800/20000 train_loss: 2.5727 train_time: 4.9m tok/s: 7525248 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2900/20000 train_loss: 2.5492 train_time: 5.1m tok/s: 7440559 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3000/20000 train_loss: 2.5767 train_time: 5.3m tok/s: 7360576 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3100/20000 train_loss: 2.5035 train_time: 5.6m tok/s: 7285918 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3200/20000 train_loss: 2.4742 train_time: 5.8m tok/s: 7220028 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3300/20000 train_loss: 2.6636 train_time: 6.0m tok/s: 7160895 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3400/20000 train_loss: 2.5640 train_time: 6.3m tok/s: 7103390 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3500/20000 train_loss: 2.5757 train_time: 6.5m tok/s: 7050563 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3600/20000 train_loss: 2.4717 train_time: 6.7m tok/s: 7001071 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3700/20000 train_loss: 2.5584 train_time: 7.0m tok/s: 6936012 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3800/20000 train_loss: 2.5040 train_time: 7.2m tok/s: 6896057 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3900/20000 train_loss: 2.6271 train_time: 7.5m tok/s: 6856869 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4000/20000 train_loss: 2.4178 train_time: 7.7m tok/s: 6819657 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4100/20000 train_loss: 2.4215 train_time: 7.9m tok/s: 6784851 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4200/20000 train_loss: 2.3964 train_time: 8.2m tok/s: 6752110 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4300/20000 train_loss: 2.5155 train_time: 8.4m tok/s: 6691291 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4400/20000 train_loss: 2.4593 train_time: 8.7m tok/s: 6662989 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4500/20000 train_loss: 2.2924 train_time: 8.9m tok/s: 6635899 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4600/20000 train_loss: 2.3856 train_time: 9.1m tok/s: 6610132 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4700/20000 train_loss: 2.3330 train_time: 9.4m tok/s: 6585856 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4800/20000 train_loss: 2.3143 train_time: 9.6m tok/s: 6562623 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4900/20000 train_loss: 2.3084 train_time: 9.8m tok/s: 6540963 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4974/20000 val_loss: 2.3577 val_bpb: 1.0773 +stopping_early: wallclock_cap train_time: 599552ms step: 4974/20000 +peak memory allocated: 41610 MiB reserved: 46918 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.33482249 val_bpb:1.06685349 eval_time:7741ms +Serialized model: 135417469 bytes +pyminify unavailable (FileNotFoundError); skipping compressed-code-size measurement +Code size (uncompressed): 175063 bytes +Code size (compressed): 0 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 3.6s +Quantized weights: + gptq (int6): blocks.attn.c_k.weight, blocks.attn.c_q.weight, blocks.attn.c_v.weight, blocks.attn.proj.weight, blocks.mlp.fc.weight, blocks.mlp.proj.weight + gptq (int7): tok_emb.weight + passthrough (float16): blocks.attn.attn_gate_w, blocks.attn.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, recur_alpha, recur_beta, skip_gates, skip_weights +Serialized model quantized+brotli: 15904209 bytes +Total submission size quantized+brotli: 15904209 bytes +diagnostic quantized val_loss:2.35502122 val_bpb:1.07608292 eval_time:11888ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (95.9s) + +beginning TTT eval timer +ttt_phased: total_docs:50000 prefix_docs:2000 suffix_docs:48000 num_phases:3 boundaries:[666, 1333, 2000] +ttp: b780/782 bl:2.2436 bb:1.0809 rl:2.2436 rb:1.0809 dl:13091-17244 gd:0 +ttp: b767/782 bl:2.2734 bb:1.0757 rl:2.2509 rb:1.0796 dl:4681-4858 gd:0 +ttpp: phase:1/3 pd:1104 gd:666 t:206.3s +tttg: c1/111 lr:0.001000 t:0.4s +tttg: c2/111 lr:0.001000 t:0.5s +tttg: c3/111 lr:0.000999 t:0.5s +tttg: c4/111 lr:0.000998 t:0.6s +tttg: c5/111 lr:0.000997 t:0.7s +tttg: c6/111 lr:0.000995 t:0.8s +tttg: c7/111 lr:0.000993 t:0.8s +tttg: c8/111 lr:0.000990 t:0.9s +tttg: c9/111 lr:0.000987 t:1.0s +tttg: c10/111 lr:0.000984 t:1.0s +tttg: c11/111 lr:0.000980 t:1.1s +tttg: c12/111 lr:0.000976 t:1.2s +tttg: c13/111 lr:0.000971 t:1.2s +tttg: c14/111 lr:0.000966 t:1.3s +tttg: c15/111 lr:0.000961 t:1.4s +tttg: c16/111 lr:0.000955 t:1.4s +tttg: c17/111 lr:0.000949 t:1.5s +tttg: c18/111 lr:0.000942 t:1.6s +tttg: c19/111 lr:0.000935 t:1.7s +tttg: c20/111 lr:0.000928 t:1.7s +tttg: c21/111 lr:0.000921 t:1.8s +tttg: c22/111 lr:0.000913 t:1.9s +tttg: c23/111 lr:0.000905 t:1.9s +tttg: c24/111 lr:0.000896 t:2.0s +tttg: c25/111 lr:0.000887 t:2.1s +tttg: c26/111 lr:0.000878 t:2.1s +tttg: c27/111 lr:0.000868 t:2.2s +tttg: c28/111 lr:0.000859 t:2.3s +tttg: c29/111 lr:0.000848 t:2.4s +tttg: c30/111 lr:0.000838 t:2.4s +tttg: c31/111 lr:0.000827 t:2.5s +tttg: c32/111 lr:0.000817 t:2.6s +tttg: c33/111 lr:0.000805 t:2.6s +tttg: c34/111 lr:0.000794 t:2.7s +tttg: c35/111 lr:0.000782 t:2.8s +tttg: c36/111 lr:0.000770 t:2.8s +tttg: c37/111 lr:0.000758 t:2.9s +tttg: c38/111 lr:0.000746 t:3.0s +tttg: c39/111 lr:0.000733 t:3.0s +tttg: c40/111 lr:0.000721 t:3.1s +tttg: c41/111 lr:0.000708 t:3.2s +tttg: c42/111 lr:0.000695 t:3.3s +tttg: c43/111 lr:0.000681 t:3.3s +tttg: c44/111 lr:0.000668 t:3.4s +tttg: c45/111 lr:0.000655 t:3.5s +tttg: c46/111 lr:0.000641 t:3.5s +tttg: c47/111 lr:0.000627 t:3.6s +tttg: c48/111 lr:0.000613 t:3.7s +tttg: c49/111 lr:0.000599 t:3.8s +tttg: c50/111 lr:0.000585 t:3.8s +tttg: c51/111 lr:0.000571 t:3.9s +tttg: c52/111 lr:0.000557 t:4.0s +tttg: c53/111 lr:0.000543 t:4.0s +tttg: c54/111 lr:0.000529 t:4.1s +tttg: c55/111 lr:0.000514 t:4.2s +tttg: c56/111 lr:0.000500 t:4.2s +tttg: c57/111 lr:0.000486 t:4.3s +tttg: c58/111 lr:0.000471 t:4.4s +tttg: c59/111 lr:0.000457 t:4.4s +tttg: c60/111 lr:0.000443 t:4.5s +tttg: c61/111 lr:0.000429 t:4.6s +tttg: c62/111 lr:0.000415 t:4.7s +tttg: c63/111 lr:0.000401 t:4.7s +tttg: c64/111 lr:0.000387 t:4.8s +tttg: c65/111 lr:0.000373 t:4.9s +tttg: c66/111 lr:0.000359 t:4.9s +tttg: c67/111 lr:0.000345 t:5.0s +tttg: c68/111 lr:0.000332 t:5.1s +tttg: c69/111 lr:0.000319 t:5.1s +tttg: c70/111 lr:0.000305 t:5.2s +tttg: c71/111 lr:0.000292 t:5.3s +tttg: c72/111 lr:0.000279 t:5.3s +tttg: c73/111 lr:0.000267 t:5.4s +tttg: c74/111 lr:0.000254 t:5.5s +tttg: c75/111 lr:0.000242 t:5.5s +tttg: c76/111 lr:0.000230 t:5.6s +tttg: c77/111 lr:0.000218 t:5.7s +tttg: c78/111 lr:0.000206 t:5.8s +tttg: c79/111 lr:0.000195 t:5.8s +tttg: c80/111 lr:0.000183 t:5.9s +tttg: c81/111 lr:0.000173 t:6.0s +tttg: c82/111 lr:0.000162 t:6.0s +tttg: c83/111 lr:0.000152 t:6.1s +tttg: c84/111 lr:0.000141 t:6.2s +tttg: c85/111 lr:0.000132 t:6.2s +tttg: c86/111 lr:0.000122 t:6.3s +tttg: c87/111 lr:0.000113 t:6.4s +tttg: c88/111 lr:0.000104 t:6.4s +tttg: c89/111 lr:0.000095 t:6.5s +tttg: c90/111 lr:0.000087 t:6.6s +tttg: c91/111 lr:0.000079 t:6.7s +tttg: c92/111 lr:0.000072 t:6.7s +tttg: c93/111 lr:0.000065 t:6.8s +tttg: c94/111 lr:0.000058 t:6.9s +tttg: c95/111 lr:0.000051 t:6.9s +tttg: c96/111 lr:0.000045 t:7.0s +tttg: c97/111 lr:0.000039 t:7.1s +tttg: c98/111 lr:0.000034 t:7.1s +tttg: c99/111 lr:0.000029 t:7.2s +tttg: c100/111 lr:0.000024 t:7.3s +tttg: c101/111 lr:0.000020 t:7.3s +tttg: c102/111 lr:0.000016 t:7.4s +tttg: c103/111 lr:0.000013 t:7.5s +tttg: c104/111 lr:0.000010 t:7.5s +tttg: c105/111 lr:0.000007 t:7.6s +tttg: c106/111 lr:0.000005 t:7.7s +tttg: c107/111 lr:0.000003 t:7.8s +tttg: c108/111 lr:0.000002 t:7.8s +tttg: c109/111 lr:0.000001 t:7.9s +tttg: c110/111 lr:0.000000 t:8.0s +ttpr: phase:1/3 t:216.1s +ttp: b757/782 bl:2.2857 bb:1.0640 rl:2.2563 rb:1.0771 dl:3550-3633 gd:0 +ttp: b754/782 bl:2.2908 bb:1.0596 rl:2.2607 rb:1.0748 dl:3345-3397 gd:0 +ttpp: phase:2/3 pd:1808 gd:1333 t:281.4s +tttg: c1/185 lr:0.001000 t:0.1s +tttg: c2/185 lr:0.001000 t:0.1s +tttg: c3/185 lr:0.001000 t:0.2s +tttg: c4/185 lr:0.000999 t:0.3s +tttg: c5/185 lr:0.000999 t:0.4s +tttg: c6/185 lr:0.000998 t:0.4s +tttg: c7/185 lr:0.000997 t:0.5s +tttg: c8/185 lr:0.000996 t:0.6s +tttg: c9/185 lr:0.000995 t:0.6s +tttg: c10/185 lr:0.000994 t:0.7s +tttg: c11/185 lr:0.000993 t:0.8s +tttg: c12/185 lr:0.000991 t:0.9s +tttg: c13/185 lr:0.000990 t:0.9s +tttg: c14/185 lr:0.000988 t:1.0s +tttg: c15/185 lr:0.000986 t:1.1s +tttg: c16/185 lr:0.000984 t:1.1s +tttg: c17/185 lr:0.000981 t:1.2s +tttg: c18/185 lr:0.000979 t:1.3s +tttg: c19/185 lr:0.000977 t:1.3s +tttg: c20/185 lr:0.000974 t:1.4s +tttg: c21/185 lr:0.000971 t:1.5s +tttg: c22/185 lr:0.000968 t:1.6s +tttg: c23/185 lr:0.000965 t:1.6s +tttg: c24/185 lr:0.000962 t:1.7s +tttg: c25/185 lr:0.000959 t:1.8s +tttg: c26/185 lr:0.000955 t:1.8s +tttg: c27/185 lr:0.000952 t:1.9s +tttg: c28/185 lr:0.000948 t:2.0s +tttg: c29/185 lr:0.000944 t:2.1s +tttg: c30/185 lr:0.000940 t:2.1s +tttg: c31/185 lr:0.000936 t:2.2s +tttg: c32/185 lr:0.000932 t:2.3s +tttg: c33/185 lr:0.000927 t:2.3s +tttg: c34/185 lr:0.000923 t:2.4s +tttg: c35/185 lr:0.000918 t:2.5s +tttg: c36/185 lr:0.000913 t:2.6s +tttg: c37/185 lr:0.000908 t:2.6s +tttg: c38/185 lr:0.000904 t:2.7s +tttg: c39/185 lr:0.000898 t:2.8s +tttg: c40/185 lr:0.000893 t:2.8s +tttg: c41/185 lr:0.000888 t:2.9s +tttg: c42/185 lr:0.000882 t:3.0s +tttg: c43/185 lr:0.000877 t:3.0s +tttg: c44/185 lr:0.000871 t:3.1s +tttg: c45/185 lr:0.000865 t:3.2s +tttg: c46/185 lr:0.000860 t:3.2s +tttg: c47/185 lr:0.000854 t:3.3s +tttg: c48/185 lr:0.000847 t:3.4s +tttg: c49/185 lr:0.000841 t:3.5s +tttg: c50/185 lr:0.000835 t:3.5s +tttg: c51/185 lr:0.000829 t:3.6s +tttg: c52/185 lr:0.000822 t:3.7s +tttg: c53/185 lr:0.000816 t:3.7s +tttg: c54/185 lr:0.000809 t:3.8s +tttg: c55/185 lr:0.000802 t:3.9s +tttg: c56/185 lr:0.000795 t:4.0s +tttg: c57/185 lr:0.000788 t:4.0s +tttg: c58/185 lr:0.000781 t:4.1s +tttg: c59/185 lr:0.000774 t:4.2s +tttg: c60/185 lr:0.000767 t:4.2s +tttg: c61/185 lr:0.000760 t:4.3s +tttg: c62/185 lr:0.000752 t:4.4s +tttg: c63/185 lr:0.000745 t:4.4s +tttg: c64/185 lr:0.000738 t:4.5s +tttg: c65/185 lr:0.000730 t:4.6s +tttg: c66/185 lr:0.000722 t:4.7s +tttg: c67/185 lr:0.000715 t:4.7s +tttg: c68/185 lr:0.000707 t:4.8s +tttg: c69/185 lr:0.000699 t:4.9s +tttg: c70/185 lr:0.000691 t:4.9s +tttg: c71/185 lr:0.000683 t:5.0s +tttg: c72/185 lr:0.000675 t:5.1s +tttg: c73/185 lr:0.000667 t:5.1s +tttg: c74/185 lr:0.000659 t:5.2s +tttg: c75/185 lr:0.000651 t:5.3s +tttg: c76/185 lr:0.000643 t:5.3s +tttg: c77/185 lr:0.000635 t:5.4s +tttg: c78/185 lr:0.000627 t:5.5s +tttg: c79/185 lr:0.000618 t:5.6s +tttg: c80/185 lr:0.000610 t:5.6s +tttg: c81/185 lr:0.000602 t:5.7s +tttg: c82/185 lr:0.000593 t:5.8s +tttg: c83/185 lr:0.000585 t:5.9s +tttg: c84/185 lr:0.000577 t:5.9s +tttg: c85/185 lr:0.000568 t:6.0s +tttg: c86/185 lr:0.000560 t:6.1s +tttg: c87/185 lr:0.000551 t:6.1s +tttg: c88/185 lr:0.000543 t:6.2s +tttg: c89/185 lr:0.000534 t:6.3s +tttg: c90/185 lr:0.000526 t:6.3s +tttg: c91/185 lr:0.000517 t:6.4s +tttg: c92/185 lr:0.000509 t:6.5s +tttg: c93/185 lr:0.000500 t:6.6s +tttg: c94/185 lr:0.000491 t:6.6s +tttg: c95/185 lr:0.000483 t:6.7s +tttg: c96/185 lr:0.000474 t:6.8s +tttg: c97/185 lr:0.000466 t:6.8s +tttg: c98/185 lr:0.000457 t:6.9s +tttg: c99/185 lr:0.000449 t:7.0s +tttg: c100/185 lr:0.000440 t:7.1s +tttg: c101/185 lr:0.000432 t:7.1s +tttg: c102/185 lr:0.000423 t:7.2s +tttg: c103/185 lr:0.000415 t:7.3s +tttg: c104/185 lr:0.000407 t:7.3s +tttg: c105/185 lr:0.000398 t:7.4s +tttg: c106/185 lr:0.000390 t:7.5s +tttg: c107/185 lr:0.000382 t:7.5s +tttg: c108/185 lr:0.000373 t:7.6s +tttg: c109/185 lr:0.000365 t:7.7s +tttg: c110/185 lr:0.000357 t:7.8s +tttg: c111/185 lr:0.000349 t:7.8s +tttg: c112/185 lr:0.000341 t:7.9s +tttg: c113/185 lr:0.000333 t:8.0s +tttg: c114/185 lr:0.000325 t:8.1s +tttg: c115/185 lr:0.000317 t:8.1s +tttg: c116/185 lr:0.000309 t:8.2s +tttg: c117/185 lr:0.000301 t:8.3s +tttg: c118/185 lr:0.000293 t:8.3s +tttg: c119/185 lr:0.000285 t:8.5s +tttg: c120/185 lr:0.000278 t:8.5s +tttg: c121/185 lr:0.000270 t:8.6s +tttg: c122/185 lr:0.000262 t:8.7s +tttg: c123/185 lr:0.000255 t:8.8s +tttg: c124/185 lr:0.000248 t:8.8s +tttg: c125/185 lr:0.000240 t:8.9s +tttg: c126/185 lr:0.000233 t:9.0s +tttg: c127/185 lr:0.000226 t:9.0s +tttg: c128/185 lr:0.000219 t:9.1s +tttg: c129/185 lr:0.000212 t:9.2s +tttg: c130/185 lr:0.000205 t:9.2s +tttg: c131/185 lr:0.000198 t:9.3s +tttg: c132/185 lr:0.000191 t:9.4s +tttg: c133/185 lr:0.000184 t:9.5s +tttg: c134/185 lr:0.000178 t:9.5s +tttg: c135/185 lr:0.000171 t:9.6s +tttg: c136/185 lr:0.000165 t:9.7s +tttg: c137/185 lr:0.000159 t:9.8s +tttg: c138/185 lr:0.000153 t:9.8s +tttg: c139/185 lr:0.000146 t:9.9s +tttg: c140/185 lr:0.000140 t:10.0s +tttg: c141/185 lr:0.000135 t:10.0s +tttg: c142/185 lr:0.000129 t:10.1s +tttg: c143/185 lr:0.000123 t:10.2s +tttg: c144/185 lr:0.000118 t:10.2s +tttg: c145/185 lr:0.000112 t:10.3s +tttg: c146/185 lr:0.000107 t:10.4s +tttg: c147/185 lr:0.000102 t:10.5s +tttg: c148/185 lr:0.000096 t:10.5s +tttg: c149/185 lr:0.000092 t:10.6s +tttg: c150/185 lr:0.000087 t:10.7s +tttg: c151/185 lr:0.000082 t:10.7s +tttg: c152/185 lr:0.000077 t:10.8s +tttg: c153/185 lr:0.000073 t:12.5s +tttg: c154/185 lr:0.000068 t:12.6s +tttg: c155/185 lr:0.000064 t:12.7s +tttg: c156/185 lr:0.000060 t:12.7s +tttg: c157/185 lr:0.000056 t:12.8s +tttg: c158/185 lr:0.000052 t:12.9s +tttg: c159/185 lr:0.000048 t:12.9s +tttg: c160/185 lr:0.000045 t:13.0s +tttg: c161/185 lr:0.000041 t:13.1s +tttg: c162/185 lr:0.000038 t:13.2s +tttg: c163/185 lr:0.000035 t:13.2s +tttg: c164/185 lr:0.000032 t:13.3s +tttg: c165/185 lr:0.000029 t:13.4s +tttg: c166/185 lr:0.000026 t:13.4s +tttg: c167/185 lr:0.000023 t:13.5s +tttg: c168/185 lr:0.000021 t:13.6s +tttg: c169/185 lr:0.000019 t:13.7s +tttg: c170/185 lr:0.000016 t:13.7s +tttg: c171/185 lr:0.000014 t:13.8s +tttg: c172/185 lr:0.000012 t:13.9s +tttg: c173/185 lr:0.000010 t:13.9s +tttg: c174/185 lr:0.000009 t:14.0s +tttg: c175/185 lr:0.000007 t:14.1s +tttg: c176/185 lr:0.000006 t:14.1s +tttg: c177/185 lr:0.000005 t:14.2s +tttg: c178/185 lr:0.000004 t:14.3s +tttg: c179/185 lr:0.000003 t:14.4s +tttg: c180/185 lr:0.000002 t:14.4s +tttg: c181/185 lr:0.000001 t:14.5s +tttg: c182/185 lr:0.000001 t:14.6s +tttg: c183/185 lr:0.000000 t:14.7s +tttg: c184/185 lr:0.000000 t:14.8s +ttpr: phase:2/3 t:298.0s +ttp: b746/782 bl:2.4182 bb:1.0655 rl:2.2764 rb:1.0738 dl:2884-2943 gd:0 +ttp: b745/782 bl:2.2406 bb:1.0258 rl:2.2732 rb:1.0694 dl:2842-2883 gd:0 +ttpp: phase:3/3 pd:2448 gd:2000 t:315.1s +tttg: c1/250 lr:0.001000 t:0.1s +tttg: c2/250 lr:0.001000 t:0.2s +tttg: c3/250 lr:0.001000 t:0.2s +tttg: c4/250 lr:0.001000 t:0.3s +tttg: c5/250 lr:0.000999 t:0.4s +tttg: c6/250 lr:0.000999 t:0.4s +tttg: c7/250 lr:0.000999 t:0.5s +tttg: c8/250 lr:0.000998 t:0.6s +tttg: c9/250 lr:0.000997 t:0.7s +tttg: c10/250 lr:0.000997 t:0.7s +tttg: c11/250 lr:0.000996 t:0.8s +tttg: c12/250 lr:0.000995 t:0.9s +tttg: c13/250 lr:0.000994 t:0.9s +tttg: c14/250 lr:0.000993 t:1.0s +tttg: c15/250 lr:0.000992 t:1.1s +tttg: c16/250 lr:0.000991 t:1.2s +tttg: c17/250 lr:0.000990 t:1.2s +tttg: c18/250 lr:0.000989 t:1.3s +tttg: c19/250 lr:0.000987 t:1.4s +tttg: c20/250 lr:0.000986 t:1.4s +tttg: c21/250 lr:0.000984 t:1.5s +tttg: c22/250 lr:0.000983 t:1.6s +tttg: c23/250 lr:0.000981 t:1.6s +tttg: c24/250 lr:0.000979 t:1.7s +tttg: c25/250 lr:0.000977 t:1.8s +tttg: c26/250 lr:0.000975 t:1.9s +tttg: c27/250 lr:0.000973 t:1.9s +tttg: c28/250 lr:0.000971 t:2.0s +tttg: c29/250 lr:0.000969 t:2.1s +tttg: c30/250 lr:0.000967 t:2.2s +tttg: c31/250 lr:0.000965 t:2.2s +tttg: c32/250 lr:0.000962 t:2.3s +tttg: c33/250 lr:0.000960 t:2.4s +tttg: c34/250 lr:0.000957 t:2.4s +tttg: c35/250 lr:0.000955 t:2.5s +tttg: c36/250 lr:0.000952 t:2.6s +tttg: c37/250 lr:0.000949 t:2.7s +tttg: c38/250 lr:0.000947 t:2.7s +tttg: c39/250 lr:0.000944 t:2.8s +tttg: c40/250 lr:0.000941 t:2.9s +tttg: c41/250 lr:0.000938 t:2.9s +tttg: c42/250 lr:0.000935 t:3.0s +tttg: c43/250 lr:0.000931 t:3.1s +tttg: c44/250 lr:0.000928 t:3.2s +tttg: c45/250 lr:0.000925 t:3.2s +tttg: c46/250 lr:0.000922 t:3.3s +tttg: c47/250 lr:0.000918 t:3.4s +tttg: c48/250 lr:0.000915 t:3.4s +tttg: c49/250 lr:0.000911 t:3.5s +tttg: c50/250 lr:0.000907 t:3.6s +tttg: c51/250 lr:0.000904 t:3.6s +tttg: c52/250 lr:0.000900 t:3.7s +tttg: c53/250 lr:0.000896 t:3.8s +tttg: c54/250 lr:0.000892 t:3.9s +tttg: c55/250 lr:0.000888 t:3.9s +tttg: c56/250 lr:0.000884 t:4.0s +tttg: c57/250 lr:0.000880 t:4.1s +tttg: c58/250 lr:0.000876 t:4.2s +tttg: c59/250 lr:0.000872 t:4.2s +tttg: c60/250 lr:0.000868 t:4.3s +tttg: c61/250 lr:0.000863 t:4.4s +tttg: c62/250 lr:0.000859 t:4.4s +tttg: c63/250 lr:0.000855 t:4.5s +tttg: c64/250 lr:0.000850 t:4.6s +tttg: c65/250 lr:0.000846 t:4.7s +tttg: c66/250 lr:0.000841 t:4.7s +tttg: c67/250 lr:0.000836 t:4.8s +tttg: c68/250 lr:0.000832 t:4.9s +tttg: c69/250 lr:0.000827 t:4.9s +tttg: c70/250 lr:0.000822 t:5.0s +tttg: c71/250 lr:0.000817 t:5.1s +tttg: c72/250 lr:0.000812 t:5.2s +tttg: c73/250 lr:0.000807 t:5.2s +tttg: c74/250 lr:0.000803 t:5.3s +tttg: c75/250 lr:0.000797 t:5.4s +tttg: c76/250 lr:0.000792 t:5.5s +tttg: c77/250 lr:0.000787 t:5.5s +tttg: c78/250 lr:0.000782 t:5.6s +tttg: c79/250 lr:0.000777 t:5.7s +tttg: c80/250 lr:0.000772 t:5.7s +tttg: c81/250 lr:0.000766 t:5.8s +tttg: c82/250 lr:0.000761 t:5.9s +tttg: c83/250 lr:0.000755 t:6.0s +tttg: c84/250 lr:0.000750 t:6.0s +tttg: c85/250 lr:0.000745 t:6.1s +tttg: c86/250 lr:0.000739 t:6.2s +tttg: c87/250 lr:0.000733 t:6.2s +tttg: c88/250 lr:0.000728 t:6.3s +tttg: c89/250 lr:0.000722 t:6.4s +tttg: c90/250 lr:0.000717 t:6.4s +tttg: c91/250 lr:0.000711 t:6.5s +tttg: c92/250 lr:0.000705 t:6.6s +tttg: c93/250 lr:0.000699 t:6.7s +tttg: c94/250 lr:0.000694 t:6.7s +tttg: c95/250 lr:0.000688 t:6.8s +tttg: c96/250 lr:0.000682 t:6.9s +tttg: c97/250 lr:0.000676 t:6.9s +tttg: c98/250 lr:0.000670 t:7.0s +tttg: c99/250 lr:0.000664 t:7.1s +tttg: c100/250 lr:0.000658 t:7.2s +tttg: c101/250 lr:0.000652 t:7.2s +tttg: c102/250 lr:0.000646 t:7.3s +tttg: c103/250 lr:0.000640 t:7.4s +tttg: c104/250 lr:0.000634 t:7.4s +tttg: c105/250 lr:0.000628 t:7.5s +tttg: c106/250 lr:0.000622 t:7.6s +tttg: c107/250 lr:0.000616 t:7.6s +tttg: c108/250 lr:0.000610 t:7.7s +tttg: c109/250 lr:0.000603 t:7.8s +tttg: c110/250 lr:0.000597 t:7.9s +tttg: c111/250 lr:0.000591 t:7.9s +tttg: c112/250 lr:0.000585 t:8.0s +tttg: c113/250 lr:0.000579 t:8.1s +tttg: c114/250 lr:0.000572 t:8.1s +tttg: c115/250 lr:0.000566 t:8.2s +tttg: c116/250 lr:0.000560 t:8.3s +tttg: c117/250 lr:0.000554 t:8.3s +tttg: c118/250 lr:0.000547 t:8.4s +tttg: c119/250 lr:0.000541 t:8.5s +tttg: c120/250 lr:0.000535 t:8.6s +tttg: c121/250 lr:0.000528 t:8.6s +tttg: c122/250 lr:0.000522 t:8.7s +tttg: c123/250 lr:0.000516 t:8.8s +tttg: c124/250 lr:0.000509 t:8.8s +tttg: c125/250 lr:0.000503 t:8.9s +tttg: c126/250 lr:0.000497 t:9.0s +tttg: c127/250 lr:0.000491 t:9.1s +tttg: c128/250 lr:0.000484 t:9.2s +tttg: c129/250 lr:0.000478 t:9.2s +tttg: c130/250 lr:0.000472 t:9.3s +tttg: c131/250 lr:0.000465 t:9.4s +tttg: c132/250 lr:0.000459 t:9.4s +tttg: c133/250 lr:0.000453 t:9.5s +tttg: c134/250 lr:0.000446 t:11.6s +tttg: c135/250 lr:0.000440 t:11.6s +tttg: c136/250 lr:0.000434 t:11.7s +tttg: c137/250 lr:0.000428 t:11.8s +tttg: c138/250 lr:0.000421 t:11.9s +tttg: c139/250 lr:0.000415 t:11.9s +tttg: c140/250 lr:0.000409 t:12.0s +tttg: c141/250 lr:0.000403 t:12.1s +tttg: c142/250 lr:0.000397 t:12.2s +tttg: c143/250 lr:0.000390 t:12.2s +tttg: c144/250 lr:0.000384 t:12.3s +tttg: c145/250 lr:0.000378 t:12.4s +tttg: c146/250 lr:0.000372 t:12.4s +tttg: c147/250 lr:0.000366 t:12.5s +tttg: c148/250 lr:0.000360 t:12.6s +tttg: c149/250 lr:0.000354 t:12.7s +tttg: c150/250 lr:0.000348 t:12.7s +tttg: c151/250 lr:0.000342 t:12.8s +tttg: c152/250 lr:0.000336 t:12.9s +tttg: c153/250 lr:0.000330 t:13.0s +tttg: c154/250 lr:0.000324 t:13.0s +tttg: c155/250 lr:0.000318 t:13.1s +tttg: c156/250 lr:0.000312 t:13.2s +tttg: c157/250 lr:0.000306 t:13.2s +tttg: c158/250 lr:0.000301 t:13.3s +tttg: c159/250 lr:0.000295 t:13.4s +tttg: c160/250 lr:0.000289 t:13.4s +tttg: c161/250 lr:0.000283 t:13.5s +tttg: c162/250 lr:0.000278 t:13.6s +tttg: c163/250 lr:0.000272 t:13.7s +tttg: c164/250 lr:0.000267 t:13.7s +tttg: c165/250 lr:0.000261 t:13.8s +tttg: c166/250 lr:0.000255 t:13.9s +tttg: c167/250 lr:0.000250 t:13.9s +tttg: c168/250 lr:0.000245 t:14.0s +tttg: c169/250 lr:0.000239 t:14.1s +tttg: c170/250 lr:0.000234 t:14.2s +tttg: c171/250 lr:0.000228 t:14.2s +tttg: c172/250 lr:0.000223 t:14.3s +tttg: c173/250 lr:0.000218 t:14.4s +tttg: c174/250 lr:0.000213 t:14.4s +tttg: c175/250 lr:0.000208 t:14.5s +tttg: c176/250 lr:0.000203 t:14.6s +tttg: c177/250 lr:0.000197 t:14.7s +tttg: c178/250 lr:0.000193 t:14.7s +tttg: c179/250 lr:0.000188 t:14.8s +tttg: c180/250 lr:0.000183 t:14.9s +tttg: c181/250 lr:0.000178 t:14.9s +tttg: c182/250 lr:0.000173 t:15.0s +tttg: c183/250 lr:0.000168 t:15.1s +tttg: c184/250 lr:0.000164 t:15.1s +tttg: c185/250 lr:0.000159 t:15.2s +tttg: c186/250 lr:0.000154 t:15.3s +tttg: c187/250 lr:0.000150 t:15.4s +tttg: c188/250 lr:0.000145 t:15.4s +tttg: c189/250 lr:0.000141 t:15.5s +tttg: c190/250 lr:0.000137 t:15.6s +tttg: c191/250 lr:0.000132 t:15.6s +tttg: c192/250 lr:0.000128 t:15.7s +tttg: c193/250 lr:0.000124 t:15.8s +tttg: c194/250 lr:0.000120 t:15.8s +tttg: c195/250 lr:0.000116 t:15.9s +tttg: c196/250 lr:0.000112 t:16.0s +tttg: c197/250 lr:0.000108 t:16.1s +tttg: c198/250 lr:0.000104 t:16.1s +tttg: c199/250 lr:0.000100 t:16.2s +tttg: c200/250 lr:0.000096 t:16.3s +tttg: c201/250 lr:0.000093 t:16.4s +tttg: c202/250 lr:0.000089 t:16.4s +tttg: c203/250 lr:0.000085 t:16.5s +tttg: c204/250 lr:0.000082 t:16.6s +tttg: c205/250 lr:0.000078 t:16.6s +tttg: c206/250 lr:0.000075 t:16.7s +tttg: c207/250 lr:0.000072 t:16.8s +tttg: c208/250 lr:0.000069 t:16.8s +tttg: c209/250 lr:0.000065 t:16.9s +tttg: c210/250 lr:0.000062 t:17.0s +tttg: c211/250 lr:0.000059 t:17.1s +tttg: c212/250 lr:0.000056 t:17.1s +tttg: c213/250 lr:0.000053 t:17.2s +tttg: c214/250 lr:0.000051 t:17.3s +tttg: c215/250 lr:0.000048 t:17.3s +tttg: c216/250 lr:0.000045 t:17.4s +tttg: c217/250 lr:0.000043 t:17.5s +tttg: c218/250 lr:0.000040 t:17.5s +tttg: c219/250 lr:0.000038 t:17.6s +tttg: c220/250 lr:0.000035 t:17.7s +tttg: c221/250 lr:0.000033 t:17.8s +tttg: c222/250 lr:0.000031 t:17.8s +tttg: c223/250 lr:0.000029 t:17.9s +tttg: c224/250 lr:0.000027 t:18.0s +tttg: c225/250 lr:0.000025 t:18.0s +tttg: c226/250 lr:0.000023 t:18.1s +tttg: c227/250 lr:0.000021 t:18.2s +tttg: c228/250 lr:0.000019 t:18.2s +tttg: c229/250 lr:0.000017 t:18.3s +tttg: c230/250 lr:0.000016 t:18.4s +tttg: c231/250 lr:0.000014 t:18.5s +tttg: c232/250 lr:0.000013 t:18.5s +tttg: c233/250 lr:0.000011 t:18.6s +tttg: c234/250 lr:0.000010 t:18.7s +tttg: c235/250 lr:0.000009 t:18.7s +tttg: c236/250 lr:0.000008 t:18.8s +tttg: c237/250 lr:0.000007 t:18.9s +tttg: c238/250 lr:0.000006 t:18.9s +tttg: c239/250 lr:0.000005 t:19.0s +tttg: c240/250 lr:0.000004 t:19.1s +tttg: c241/250 lr:0.000003 t:19.2s +tttg: c242/250 lr:0.000003 t:19.2s +tttg: c243/250 lr:0.000002 t:19.3s +tttg: c244/250 lr:0.000001 t:19.4s +tttg: c245/250 lr:0.000001 t:19.4s +tttg: c246/250 lr:0.000001 t:19.5s +tttg: c247/250 lr:0.000000 t:19.6s +tttg: c248/250 lr:0.000000 t:19.6s +tttg: c249/250 lr:0.000000 t:19.7s +ttpr: phase:3/3 t:336.7s +ttp: b736/782 bl:2.2505 bb:1.0603 rl:2.2715 rb:1.0688 dl:2526-2550 gd:1 +ttp: b735/782 bl:2.3932 bb:1.1010 rl:2.2797 rb:1.0710 dl:2495-2526 gd:1 +ttp: b724/782 bl:2.3217 bb:1.0601 rl:2.2821 rb:1.0703 dl:2203-2231 gd:1 +ttp: b714/782 bl:2.3135 bb:1.0247 rl:2.2836 rb:1.0680 dl:2018-2035 gd:1 +ttp: b709/782 bl:2.4496 bb:1.0957 rl:2.2911 rb:1.0693 dl:1937-1952 gd:1 +ttp: b701/782 bl:2.3140 bb:1.0375 rl:2.2920 rb:1.0679 dl:1835-1847 gd:1 +ttp: b689/782 bl:2.3983 bb:1.0798 rl:2.2959 rb:1.0684 dl:1706-1715 gd:1 +ttp: b684/782 bl:2.3772 bb:1.0473 rl:2.2986 rb:1.0676 dl:1658-1665 gd:1 +ttp: b677/782 bl:2.3129 bb:1.0362 rl:2.2991 rb:1.0666 dl:1595-1601 gd:1 +ttp: b667/782 bl:2.3717 bb:1.0721 rl:2.3012 rb:1.0668 dl:1514-1521 gd:1 +ttp: b660/782 bl:2.3768 bb:1.0507 rl:2.3033 rb:1.0663 dl:1466-1474 gd:1 +ttp: b653/782 bl:2.2946 bb:1.0403 rl:2.3031 rb:1.0656 dl:1419-1425 gd:1 +ttp: b645/782 bl:2.3032 bb:1.0305 rl:2.3031 rb:1.0647 dl:1367-1375 gd:1 +ttp: b637/782 bl:2.3691 bb:1.0804 rl:2.3046 rb:1.0651 dl:1320-1325 gd:1 +ttp: b629/782 bl:2.3570 bb:1.0143 rl:2.3058 rb:1.0639 dl:1276-1280 gd:1 +ttp: b621/782 bl:2.3037 bb:1.0520 rl:2.3057 rb:1.0637 dl:1231-1237 gd:1 +ttp: b613/782 bl:2.3392 bb:1.0415 rl:2.3064 rb:1.0632 dl:1190-1195 gd:1 +ttp: b605/782 bl:2.2518 bb:1.0269 rl:2.3053 rb:1.0625 dl:1154-1159 gd:1 +ttp: b593/782 bl:2.2957 bb:1.0134 rl:2.3052 rb:1.0616 dl:1103-1107 gd:1 +ttp: b584/782 bl:2.3022 bb:1.0409 rl:2.3051 rb:1.0613 dl:1064-1069 gd:1 +ttp: b576/782 bl:2.3821 bb:1.0957 rl:2.3063 rb:1.0618 dl:1033-1037 gd:1 +ttp: b568/782 bl:2.3608 bb:1.0838 rl:2.3072 rb:1.0622 dl:1004-1007 gd:1 +ttp: b565/782 bl:2.3914 bb:1.0360 rl:2.3084 rb:1.0617 dl:993-997 gd:1 +ttp: b556/782 bl:2.3813 bb:1.0706 rl:2.3094 rb:1.0619 dl:961-965 gd:1 +ttp: b545/782 bl:2.3352 bb:1.0326 rl:2.3098 rb:1.0615 dl:927-930 gd:1 +ttp: b537/782 bl:2.3818 bb:1.0743 rl:2.3107 rb:1.0616 dl:902-905 gd:1 +ttp: b531/782 bl:2.3007 bb:1.0444 rl:2.3106 rb:1.0614 dl:884-887 gd:1 +ttp: b522/782 bl:2.3093 bb:1.0357 rl:2.3106 rb:1.0611 dl:858-860 gd:1 +ttp: b518/782 bl:2.2469 bb:1.0113 rl:2.3098 rb:1.0605 dl:846-850 gd:1 +ttp: b509/782 bl:2.3646 bb:1.0381 rl:2.3104 rb:1.0603 dl:820-823 gd:1 +ttp: b497/782 bl:2.3415 bb:1.0442 rl:2.3108 rb:1.0601 dl:788-791 gd:1 +ttp: b489/782 bl:2.3942 bb:1.0772 rl:2.3116 rb:1.0603 dl:769-771 gd:1 +ttp: b481/782 bl:2.3049 bb:1.0477 rl:2.3116 rb:1.0601 dl:749-752 gd:1 +ttp: b473/782 bl:2.2691 bb:1.0325 rl:2.3112 rb:1.0599 dl:730-733 gd:1 +ttp: b467/782 bl:2.3552 bb:1.0557 rl:2.3116 rb:1.0598 dl:717-719 gd:1 +ttp: b461/782 bl:2.3802 bb:1.0413 rl:2.3122 rb:1.0597 dl:703-706 gd:1 +ttp: b453/782 bl:2.3445 bb:1.0593 rl:2.3125 rb:1.0596 dl:687-689 gd:1 +ttp: b445/782 bl:2.3621 bb:1.0498 rl:2.3129 rb:1.0596 dl:670-672 gd:1 +ttp: b441/782 bl:2.3487 bb:1.0473 rl:2.3132 rb:1.0595 dl:662-664 gd:1 +ttp: b423/782 bl:2.3104 bb:1.0542 rl:2.3132 rb:1.0594 dl:626-629 gd:1 +ttp: b415/782 bl:2.2901 bb:1.0608 rl:2.3130 rb:1.0594 dl:611-613 gd:1 +ttp: b407/782 bl:2.2741 bb:1.0411 rl:2.3127 rb:1.0593 dl:595-597 gd:1 +ttp: b399/782 bl:2.2867 bb:1.0320 rl:2.3125 rb:1.0591 dl:581-582 gd:1 +ttp: b391/782 bl:2.3160 bb:1.0668 rl:2.3125 rb:1.0592 dl:566-568 gd:1 +ttp: b383/782 bl:2.2844 bb:1.0474 rl:2.3124 rb:1.0591 dl:552-554 gd:1 +ttp: b375/782 bl:2.4123 bb:1.0759 rl:2.3130 rb:1.0592 dl:538-540 gd:1 +ttp: b367/782 bl:2.3056 bb:1.0880 rl:2.3130 rb:1.0594 dl:525-527 gd:1 +ttp: b359/782 bl:2.2592 bb:1.0374 rl:2.3126 rb:1.0592 dl:512-513 gd:1 +ttp: b351/782 bl:2.3667 bb:1.0835 rl:2.3129 rb:1.0594 dl:498-499 gd:1 +ttp: b343/782 bl:2.2273 bb:1.0482 rl:2.3125 rb:1.0593 dl:486-488 gd:1 +ttp: b335/782 bl:2.3751 bb:1.0760 rl:2.3128 rb:1.0594 dl:474-476 gd:1 +ttp: b327/782 bl:2.3333 bb:1.0849 rl:2.3129 rb:1.0595 dl:462-463 gd:1 +ttp: b319/782 bl:2.3983 bb:1.0815 rl:2.3134 rb:1.0596 dl:450-451 gd:1 +ttp: b311/782 bl:2.3468 bb:1.0817 rl:2.3135 rb:1.0598 dl:438-439 gd:1 +ttp: b304/782 bl:2.3481 bb:1.0770 rl:2.3137 rb:1.0598 dl:427-429 gd:1 +ttp: b297/782 bl:2.4047 bb:1.0865 rl:2.3141 rb:1.0600 dl:417-418 gd:1 +ttp: b289/782 bl:2.3276 bb:1.0825 rl:2.3142 rb:1.0601 dl:405-406 gd:1 +ttp: b281/782 bl:2.2940 bb:1.0875 rl:2.3141 rb:1.0602 dl:394-395 gd:1 +ttp: b273/782 bl:2.3394 bb:1.0784 rl:2.3142 rb:1.0603 dl:383-384 gd:1 +ttp: b265/782 bl:2.3682 bb:1.1019 rl:2.3144 rb:1.0604 dl:372-374 gd:1 +ttp: b257/782 bl:2.4489 bb:1.1140 rl:2.3149 rb:1.0606 dl:362-364 gd:1 +ttp: b248/782 bl:2.4711 bb:1.1926 rl:2.3156 rb:1.0611 dl:351-352 gd:1 +ttp: b239/782 bl:2.3808 bb:1.1055 rl:2.3158 rb:1.0613 dl:340-341 gd:1 +ttp: b229/782 bl:2.3679 bb:1.0672 rl:2.3160 rb:1.0613 dl:328-329 gd:1 +ttp: b220/782 bl:2.4162 bb:1.1432 rl:2.3163 rb:1.0616 dl:317-318 gd:1 +ttp: b211/782 bl:2.4082 bb:1.0970 rl:2.3166 rb:1.0617 dl:307-308 gd:1 +ttp: b203/782 bl:2.4376 bb:1.1120 rl:2.3170 rb:1.0619 dl:299-300 gd:1 +ttp: b195/782 bl:2.4303 bb:1.1335 rl:2.3174 rb:1.0621 dl:290-291 gd:1 +ttp: b187/782 bl:2.4558 bb:1.1348 rl:2.3178 rb:1.0623 dl:281-282 gd:1 +ttp: b179/782 bl:2.3774 bb:1.1334 rl:2.3180 rb:1.0625 dl:273-274 gd:1 +ttp: b171/782 bl:2.4731 bb:1.1404 rl:2.3184 rb:1.0627 dl:266-266 gd:1 +ttp: b163/782 bl:2.3812 bb:1.1218 rl:2.3186 rb:1.0629 dl:257-259 gd:1 +ttp: b155/782 bl:2.4093 bb:1.1139 rl:2.3188 rb:1.0630 dl:250-251 gd:1 +ttp: b147/782 bl:2.4708 bb:1.1238 rl:2.3192 rb:1.0632 dl:242-243 gd:1 +ttp: b138/782 bl:2.3912 bb:1.1125 rl:2.3194 rb:1.0633 dl:233-234 gd:1 +ttp: b130/782 bl:2.5817 bb:1.1831 rl:2.3200 rb:1.0636 dl:226-227 gd:1 +ttp: b120/782 bl:2.3940 bb:1.1124 rl:2.3202 rb:1.0637 dl:217-218 gd:1 +ttp: b112/782 bl:2.4757 bb:1.1817 rl:2.3205 rb:1.0639 dl:210-210 gd:1 +ttp: b102/782 bl:2.5835 bb:1.1975 rl:2.3211 rb:1.0642 dl:201-202 gd:1 +ttp: b94/782 bl:2.5635 bb:1.2113 rl:2.3215 rb:1.0645 dl:193-194 gd:1 +ttp: b84/782 bl:2.5272 bb:1.2017 rl:2.3219 rb:1.0647 dl:184-185 gd:1 +ttp: b76/782 bl:2.4965 bb:1.1725 rl:2.3223 rb:1.0649 dl:177-178 gd:1 +ttp: b68/782 bl:2.5205 bb:1.1763 rl:2.3226 rb:1.0651 dl:170-171 gd:1 +ttp: b60/782 bl:2.4753 bb:1.1897 rl:2.3229 rb:1.0653 dl:163-164 gd:1 +ttp: b51/782 bl:2.4837 bb:1.1883 rl:2.3231 rb:1.0655 dl:154-155 gd:1 +ttp: b42/782 bl:2.4840 bb:1.2095 rl:2.3234 rb:1.0657 dl:145-146 gd:1 +ttp: b30/782 bl:2.6031 bb:1.2692 rl:2.3237 rb:1.0660 dl:133-134 gd:1 +ttp: b22/782 bl:2.5640 bb:1.2002 rl:2.3241 rb:1.0662 dl:124-126 gd:1 +ttp: b14/782 bl:2.6006 bb:1.1872 rl:2.3244 rb:1.0663 dl:114-115 gd:1 +ttp: b6/782 bl:2.7099 bb:1.2082 rl:2.3248 rb:1.0664 dl:99-101 gd:1 +quantized_ttt_phased val_loss:2.32648401 val_bpb:1.06311240 eval_time:436134ms +total_eval_time:436.1s diff --git a/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed1234.log b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed1234.log new file mode 100644 index 0000000000..114d2aaf3c --- /dev/null +++ b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed1234.log @@ -0,0 +1,933 @@ +W0423 23:28:06.360000 203907 torch/distributed/run.py:803] +W0423 23:28:06.360000 203907 torch/distributed/run.py:803] ***************************************** +W0423 23:28:06.360000 203907 torch/distributed/run.py:803] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0423 23:28:06.360000 203907 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: /workspace/runs/036-035e-8h-promotion/seed_1234 + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + beta1: 0.9 + beta2: 0.95 + caseops_enabled: True + compressor: brotli + data_dir: /workspace/parameter-golf/data + datasets_dir: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2048 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.005 + gated_attn_quant_gate: True + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 0.5 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: /workspace/runs/036-035e-8h-promotion/seed_1234/af3a1db6-4929-4c62-9c52-a3634d2bff54.txt + logit_softcap: 30.0 + loop_depth_upgrade_at: 0.0 + loop_end: 5 + loop_start: 3 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: /workspace/runs/036-035e-8h-promotion/seed_1234/final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 3 + phased_ttt_prefix_docs: 2000 + qk_gain_init: 5.0 + quantized_model_path: /workspace/runs/036-035e-8h-promotion/seed_1234/final_model.int6.ptz + rank: 0 + recur_alpha_enabled: True + recur_diag_p2p_cos: False + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: af3a1db6-4929-4c62-9c52-a3634d2bff54 + scalar_lr: 0.02 + seed: 1234 + skip_gates_enabled: True + smear_gate_enabled: False + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 1.0 + spinquant_enabled: False + spinquant_seed: 42 + spinquant_sites: attn_in,attn_proj_in,mlp_in,mlp_proj_in + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_train_*.bin + train_log_every: 100 + train_seq_len: 2048 + ttt_batch_size: 64 + ttt_beta1: 0.0 + ttt_beta2: 0.999 + ttt_chunk_size: 48 + ttt_enabled: True + ttt_eval_batches: + ttt_eval_seq_len: 2048 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_alpha: 144 + ttt_lora_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 1.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 0 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 47851520 +model_params:35945658 +recur_alpha: enabled=True num_loops=2 loop_start=3 loop_end=5 diag_p2p_cos=False +gptq:reserving 0s, effective=599500ms +warmup_cu_buckets:64,128,192,256 iters_each:3 +warmup_step: 1/20 +warmup_step: 2/20 +warmup_step: 3/20 +warmup_step: 4/20 +warmup_step: 5/20 +warmup_step: 6/20 +warmup_step: 10/20 +warmup_step: 20/20 +loop_warmup:enabled encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +loop_warmup_step: 1/20 +loop_warmup_step: 2/20 +loop_warmup_step: 3/20 +loop_warmup_step: 4/20 +loop_warmup_step: 5/20 +loop_warmup_step: 6/20 +loop_warmup_step: 10/20 +loop_warmup_step: 20/20 +1/20000 train_loss: 9.0077 train_time: 0.0m tok/s: 12744235 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2/20000 train_loss: 12.8821 train_time: 0.0m tok/s: 7617729 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3/20000 train_loss: 10.2472 train_time: 0.0m tok/s: 7641064 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4/20000 train_loss: 8.7833 train_time: 0.0m tok/s: 7640919 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +5/20000 train_loss: 8.0392 train_time: 0.0m tok/s: 7615960 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +100/20000 train_loss: 3.6121 train_time: 0.2m tok/s: 8413313 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +200/20000 train_loss: 3.1502 train_time: 0.3m tok/s: 8328220 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +300/20000 train_loss: 2.9134 train_time: 0.5m tok/s: 8302230 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +400/20000 train_loss: 2.5797 train_time: 0.6m tok/s: 8297802 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +500/20000 train_loss: 2.5709 train_time: 0.8m tok/s: 8323272 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +600/20000 train_loss: 2.6752 train_time: 0.9m tok/s: 8313899 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +700/20000 train_loss: 2.8714 train_time: 1.1m tok/s: 8310031 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +800/20000 train_loss: 2.7168 train_time: 1.3m tok/s: 8306262 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +900/20000 train_loss: 2.7582 train_time: 1.4m tok/s: 8304979 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1000/20000 train_loss: 2.8074 train_time: 1.6m tok/s: 8316385 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1100/20000 train_loss: 2.7670 train_time: 1.7m tok/s: 8311841 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1200/20000 train_loss: 2.7662 train_time: 1.9m tok/s: 8306361 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1300/20000 train_loss: 2.8331 train_time: 2.1m tok/s: 8304923 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1400/20000 train_loss: 2.5870 train_time: 2.2m tok/s: 8301161 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1500/20000 train_loss: 2.6375 train_time: 2.4m tok/s: 8309357 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1600/20000 train_loss: 2.7075 train_time: 2.5m tok/s: 8306190 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1700/20000 train_loss: 2.6801 train_time: 2.7m tok/s: 8304059 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1800/20000 train_loss: 2.6482 train_time: 2.8m tok/s: 8302200 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1900/20000 train_loss: 2.7408 train_time: 3.0m tok/s: 8308664 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2000/20000 train_loss: 2.6654 train_time: 3.2m tok/s: 8306974 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2100/20000 train_loss: 2.6840 train_time: 3.3m tok/s: 8304083 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2200/20000 train_loss: 2.5310 train_time: 3.5m tok/s: 8302323 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +layer_loop:enabled step:2215 frac:0.350 depth:3 encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +2300/20000 train_loss: 2.6126 train_time: 3.7m tok/s: 8158764 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2400/20000 train_loss: 2.6285 train_time: 3.9m tok/s: 8011988 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2500/20000 train_loss: 2.5528 train_time: 4.2m tok/s: 7877034 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2600/20000 train_loss: 2.5205 train_time: 4.4m tok/s: 7757885 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2700/20000 train_loss: 2.5068 train_time: 4.6m tok/s: 7650732 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2800/20000 train_loss: 2.5707 train_time: 4.9m tok/s: 7551597 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2900/20000 train_loss: 2.5478 train_time: 5.1m tok/s: 7465922 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3000/20000 train_loss: 2.5696 train_time: 5.3m tok/s: 7385194 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3100/20000 train_loss: 2.4997 train_time: 5.6m tok/s: 7311845 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3200/20000 train_loss: 2.4714 train_time: 5.8m tok/s: 7244250 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3300/20000 train_loss: 2.6616 train_time: 6.0m tok/s: 7184309 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3400/20000 train_loss: 2.5665 train_time: 6.3m tok/s: 7124465 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3500/20000 train_loss: 2.5709 train_time: 6.5m tok/s: 7069879 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3600/20000 train_loss: 2.4660 train_time: 6.7m tok/s: 7019391 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3700/20000 train_loss: 2.5541 train_time: 7.0m tok/s: 6938495 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3800/20000 train_loss: 2.5010 train_time: 7.2m tok/s: 6898388 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3900/20000 train_loss: 2.6288 train_time: 7.5m tok/s: 6858864 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4000/20000 train_loss: 2.4178 train_time: 7.7m tok/s: 6821417 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4100/20000 train_loss: 2.4178 train_time: 7.9m tok/s: 6786592 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4200/20000 train_loss: 2.4096 train_time: 8.2m tok/s: 6752411 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4300/20000 train_loss: 2.5105 train_time: 8.4m tok/s: 6689408 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4400/20000 train_loss: 2.4573 train_time: 8.7m tok/s: 6660712 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4500/20000 train_loss: 2.2876 train_time: 8.9m tok/s: 6633429 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4600/20000 train_loss: 2.3834 train_time: 9.1m tok/s: 6607625 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4700/20000 train_loss: 2.3298 train_time: 9.4m tok/s: 6584006 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4800/20000 train_loss: 2.3104 train_time: 9.6m tok/s: 6560918 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4900/20000 train_loss: 2.3039 train_time: 9.8m tok/s: 6538983 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4973/20000 val_loss: 2.3553 val_bpb: 1.0762 +stopping_early: wallclock_cap train_time: 599598ms step: 4973/20000 +peak memory allocated: 41610 MiB reserved: 46918 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.33247035 val_bpb:1.06577872 eval_time:7636ms +Serialized model: 135417469 bytes +pyminify unavailable (FileNotFoundError); skipping compressed-code-size measurement +Code size (uncompressed): 175063 bytes +Code size (compressed): 0 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 3.6s +Quantized weights: + gptq (int6): blocks.attn.c_k.weight, blocks.attn.c_q.weight, blocks.attn.c_v.weight, blocks.attn.proj.weight, blocks.mlp.fc.weight, blocks.mlp.proj.weight + gptq (int7): tok_emb.weight + passthrough (float16): blocks.attn.attn_gate_w, blocks.attn.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, recur_alpha, recur_beta, skip_gates, skip_weights +Serialized model quantized+brotli: 15909401 bytes +Total submission size quantized+brotli: 15909401 bytes +diagnostic quantized val_loss:2.35283859 val_bpb:1.07508561 eval_time:11996ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (86.1s) + +beginning TTT eval timer +ttt_phased: total_docs:50000 prefix_docs:2000 suffix_docs:48000 num_phases:3 boundaries:[666, 1333, 2000] +ttp: b776/782 bl:2.2582 bb:1.0707 rl:2.2582 rb:1.0707 dl:7534-8350 gd:0 +ttp: b773/782 bl:2.2019 bb:1.0369 rl:2.2333 rb:1.0557 dl:6104-6447 gd:0 +ttp: b768/782 bl:2.2432 bb:1.0447 rl:2.2358 rb:1.0528 dl:4859-5083 gd:0 +ttpp: phase:1/3 pd:1104 gd:666 t:169.5s +tttg: c1/111 lr:0.001000 t:0.4s +tttg: c2/111 lr:0.001000 t:0.5s +tttg: c3/111 lr:0.000999 t:0.6s +tttg: c4/111 lr:0.000998 t:0.6s +tttg: c5/111 lr:0.000997 t:0.7s +tttg: c6/111 lr:0.000995 t:0.8s +tttg: c7/111 lr:0.000993 t:0.8s +tttg: c8/111 lr:0.000990 t:0.9s +tttg: c9/111 lr:0.000987 t:1.0s +tttg: c10/111 lr:0.000984 t:1.1s +tttg: c11/111 lr:0.000980 t:1.1s +tttg: c12/111 lr:0.000976 t:1.2s +tttg: c13/111 lr:0.000971 t:1.3s +tttg: c14/111 lr:0.000966 t:1.3s +tttg: c15/111 lr:0.000961 t:1.4s +tttg: c16/111 lr:0.000955 t:1.5s +tttg: c17/111 lr:0.000949 t:1.5s +tttg: c18/111 lr:0.000942 t:1.6s +tttg: c19/111 lr:0.000935 t:1.7s +tttg: c20/111 lr:0.000928 t:1.7s +tttg: c21/111 lr:0.000921 t:1.8s +tttg: c22/111 lr:0.000913 t:1.9s +tttg: c23/111 lr:0.000905 t:1.9s +tttg: c24/111 lr:0.000896 t:2.0s +tttg: c25/111 lr:0.000887 t:2.1s +tttg: c26/111 lr:0.000878 t:2.2s +tttg: c27/111 lr:0.000868 t:2.2s +tttg: c28/111 lr:0.000859 t:2.3s +tttg: c29/111 lr:0.000848 t:2.4s +tttg: c30/111 lr:0.000838 t:2.4s +tttg: c31/111 lr:0.000827 t:2.5s +tttg: c32/111 lr:0.000817 t:2.6s +tttg: c33/111 lr:0.000805 t:2.7s +tttg: c34/111 lr:0.000794 t:2.7s +tttg: c35/111 lr:0.000782 t:2.8s +tttg: c36/111 lr:0.000770 t:2.9s +tttg: c37/111 lr:0.000758 t:2.9s +tttg: c38/111 lr:0.000746 t:3.0s +tttg: c39/111 lr:0.000733 t:3.1s +tttg: c40/111 lr:0.000721 t:3.1s +tttg: c41/111 lr:0.000708 t:3.2s +tttg: c42/111 lr:0.000695 t:3.3s +tttg: c43/111 lr:0.000681 t:3.3s +tttg: c44/111 lr:0.000668 t:3.4s +tttg: c45/111 lr:0.000655 t:3.5s +tttg: c46/111 lr:0.000641 t:3.6s +tttg: c47/111 lr:0.000627 t:3.6s +tttg: c48/111 lr:0.000613 t:3.7s +tttg: c49/111 lr:0.000599 t:3.8s +tttg: c50/111 lr:0.000585 t:3.8s +tttg: c51/111 lr:0.000571 t:3.9s +tttg: c52/111 lr:0.000557 t:4.0s +tttg: c53/111 lr:0.000543 t:4.0s +tttg: c54/111 lr:0.000529 t:4.1s +tttg: c55/111 lr:0.000514 t:4.2s +tttg: c56/111 lr:0.000500 t:4.2s +tttg: c57/111 lr:0.000486 t:4.3s +tttg: c58/111 lr:0.000471 t:4.4s +tttg: c59/111 lr:0.000457 t:4.5s +tttg: c60/111 lr:0.000443 t:4.5s +tttg: c61/111 lr:0.000429 t:4.6s +tttg: c62/111 lr:0.000415 t:4.7s +tttg: c63/111 lr:0.000401 t:4.7s +tttg: c64/111 lr:0.000387 t:4.8s +tttg: c65/111 lr:0.000373 t:4.9s +tttg: c66/111 lr:0.000359 t:4.9s +tttg: c67/111 lr:0.000345 t:5.0s +tttg: c68/111 lr:0.000332 t:5.1s +tttg: c69/111 lr:0.000319 t:5.1s +tttg: c70/111 lr:0.000305 t:5.2s +tttg: c71/111 lr:0.000292 t:5.3s +tttg: c72/111 lr:0.000279 t:5.4s +tttg: c73/111 lr:0.000267 t:5.4s +tttg: c74/111 lr:0.000254 t:5.5s +tttg: c75/111 lr:0.000242 t:5.6s +tttg: c76/111 lr:0.000230 t:5.6s +tttg: c77/111 lr:0.000218 t:5.7s +tttg: c78/111 lr:0.000206 t:5.8s +tttg: c79/111 lr:0.000195 t:5.8s +tttg: c80/111 lr:0.000183 t:5.9s +tttg: c81/111 lr:0.000173 t:6.0s +tttg: c82/111 lr:0.000162 t:6.0s +tttg: c83/111 lr:0.000152 t:6.1s +tttg: c84/111 lr:0.000141 t:6.2s +tttg: c85/111 lr:0.000132 t:6.2s +tttg: c86/111 lr:0.000122 t:6.3s +tttg: c87/111 lr:0.000113 t:6.4s +tttg: c88/111 lr:0.000104 t:6.5s +tttg: c89/111 lr:0.000095 t:6.5s +tttg: c90/111 lr:0.000087 t:6.6s +tttg: c91/111 lr:0.000079 t:6.7s +tttg: c92/111 lr:0.000072 t:6.7s +tttg: c93/111 lr:0.000065 t:6.8s +tttg: c94/111 lr:0.000058 t:6.9s +tttg: c95/111 lr:0.000051 t:6.9s +tttg: c96/111 lr:0.000045 t:7.0s +tttg: c97/111 lr:0.000039 t:7.1s +tttg: c98/111 lr:0.000034 t:7.2s +tttg: c99/111 lr:0.000029 t:7.2s +tttg: c100/111 lr:0.000024 t:7.3s +tttg: c101/111 lr:0.000020 t:7.4s +tttg: c102/111 lr:0.000016 t:7.4s +tttg: c103/111 lr:0.000013 t:7.5s +tttg: c104/111 lr:0.000010 t:7.6s +tttg: c105/111 lr:0.000007 t:7.6s +tttg: c106/111 lr:0.000005 t:7.7s +tttg: c107/111 lr:0.000003 t:7.8s +tttg: c108/111 lr:0.000002 t:7.8s +tttg: c109/111 lr:0.000001 t:7.9s +tttg: c110/111 lr:0.000000 t:8.0s +ttpr: phase:1/3 t:179.3s +ttp: b757/782 bl:2.2828 bb:1.0626 rl:2.2432 rb:1.0543 dl:3550-3633 gd:0 +ttp: b756/782 bl:2.3320 bb:1.0379 rl:2.2551 rb:1.0520 dl:3466-3549 gd:0 +ttpp: phase:2/3 pd:1808 gd:1333 t:244.6s +tttg: c1/185 lr:0.001000 t:0.1s +tttg: c2/185 lr:0.001000 t:0.1s +tttg: c3/185 lr:0.001000 t:0.2s +tttg: c4/185 lr:0.000999 t:0.3s +tttg: c5/185 lr:0.000999 t:0.3s +tttg: c6/185 lr:0.000998 t:0.4s +tttg: c7/185 lr:0.000997 t:0.5s +tttg: c8/185 lr:0.000996 t:0.6s +tttg: c9/185 lr:0.000995 t:0.6s +tttg: c10/185 lr:0.000994 t:0.7s +tttg: c11/185 lr:0.000993 t:0.8s +tttg: c12/185 lr:0.000991 t:0.8s +tttg: c13/185 lr:0.000990 t:0.9s +tttg: c14/185 lr:0.000988 t:1.0s +tttg: c15/185 lr:0.000986 t:1.0s +tttg: c16/185 lr:0.000984 t:1.1s +tttg: c17/185 lr:0.000981 t:1.2s +tttg: c18/185 lr:0.000979 t:1.3s +tttg: c19/185 lr:0.000977 t:1.3s +tttg: c20/185 lr:0.000974 t:1.4s +tttg: c21/185 lr:0.000971 t:1.5s +tttg: c22/185 lr:0.000968 t:1.5s +tttg: c23/185 lr:0.000965 t:1.6s +tttg: c24/185 lr:0.000962 t:1.7s +tttg: c25/185 lr:0.000959 t:1.7s +tttg: c26/185 lr:0.000955 t:1.8s +tttg: c27/185 lr:0.000952 t:1.9s +tttg: c28/185 lr:0.000948 t:1.9s +tttg: c29/185 lr:0.000944 t:2.0s +tttg: c30/185 lr:0.000940 t:2.1s +tttg: c31/185 lr:0.000936 t:2.2s +tttg: c32/185 lr:0.000932 t:2.2s +tttg: c33/185 lr:0.000927 t:2.3s +tttg: c34/185 lr:0.000923 t:2.4s +tttg: c35/185 lr:0.000918 t:2.4s +tttg: c36/185 lr:0.000913 t:2.5s +tttg: c37/185 lr:0.000908 t:2.6s +tttg: c38/185 lr:0.000904 t:2.6s +tttg: c39/185 lr:0.000898 t:2.7s +tttg: c40/185 lr:0.000893 t:2.8s +tttg: c41/185 lr:0.000888 t:2.9s +tttg: c42/185 lr:0.000882 t:2.9s +tttg: c43/185 lr:0.000877 t:3.0s +tttg: c44/185 lr:0.000871 t:3.1s +tttg: c45/185 lr:0.000865 t:3.1s +tttg: c46/185 lr:0.000860 t:3.2s +tttg: c47/185 lr:0.000854 t:3.3s +tttg: c48/185 lr:0.000847 t:3.3s +tttg: c49/185 lr:0.000841 t:3.4s +tttg: c50/185 lr:0.000835 t:3.5s +tttg: c51/185 lr:0.000829 t:3.6s +tttg: c52/185 lr:0.000822 t:3.6s +tttg: c53/185 lr:0.000816 t:3.7s +tttg: c54/185 lr:0.000809 t:3.8s +tttg: c55/185 lr:0.000802 t:3.8s +tttg: c56/185 lr:0.000795 t:3.9s +tttg: c57/185 lr:0.000788 t:4.0s +tttg: c58/185 lr:0.000781 t:4.0s +tttg: c59/185 lr:0.000774 t:4.1s +tttg: c60/185 lr:0.000767 t:4.2s +tttg: c61/185 lr:0.000760 t:4.3s +tttg: c62/185 lr:0.000752 t:4.3s +tttg: c63/185 lr:0.000745 t:4.4s +tttg: c64/185 lr:0.000738 t:4.5s +tttg: c65/185 lr:0.000730 t:4.5s +tttg: c66/185 lr:0.000722 t:4.6s +tttg: c67/185 lr:0.000715 t:4.7s +tttg: c68/185 lr:0.000707 t:4.7s +tttg: c69/185 lr:0.000699 t:4.8s +tttg: c70/185 lr:0.000691 t:4.9s +tttg: c71/185 lr:0.000683 t:5.0s +tttg: c72/185 lr:0.000675 t:5.0s +tttg: c73/185 lr:0.000667 t:5.1s +tttg: c74/185 lr:0.000659 t:5.2s +tttg: c75/185 lr:0.000651 t:5.2s +tttg: c76/185 lr:0.000643 t:5.3s +tttg: c77/185 lr:0.000635 t:5.4s +tttg: c78/185 lr:0.000627 t:5.4s +tttg: c79/185 lr:0.000618 t:5.5s +tttg: c80/185 lr:0.000610 t:5.6s +tttg: c81/185 lr:0.000602 t:5.7s +tttg: c82/185 lr:0.000593 t:5.7s +tttg: c83/185 lr:0.000585 t:5.8s +tttg: c84/185 lr:0.000577 t:5.9s +tttg: c85/185 lr:0.000568 t:5.9s +tttg: c86/185 lr:0.000560 t:6.0s +tttg: c87/185 lr:0.000551 t:6.1s +tttg: c88/185 lr:0.000543 t:6.1s +tttg: c89/185 lr:0.000534 t:6.2s +tttg: c90/185 lr:0.000526 t:6.3s +tttg: c91/185 lr:0.000517 t:6.3s +tttg: c92/185 lr:0.000509 t:6.4s +tttg: c93/185 lr:0.000500 t:6.5s +tttg: c94/185 lr:0.000491 t:6.6s +tttg: c95/185 lr:0.000483 t:6.6s +tttg: c96/185 lr:0.000474 t:6.7s +tttg: c97/185 lr:0.000466 t:6.8s +tttg: c98/185 lr:0.000457 t:6.8s +tttg: c99/185 lr:0.000449 t:6.9s +tttg: c100/185 lr:0.000440 t:7.0s +tttg: c101/185 lr:0.000432 t:7.1s +tttg: c102/185 lr:0.000423 t:7.1s +tttg: c103/185 lr:0.000415 t:7.2s +tttg: c104/185 lr:0.000407 t:7.3s +tttg: c105/185 lr:0.000398 t:7.3s +tttg: c106/185 lr:0.000390 t:7.4s +tttg: c107/185 lr:0.000382 t:7.5s +tttg: c108/185 lr:0.000373 t:7.5s +tttg: c109/185 lr:0.000365 t:7.6s +tttg: c110/185 lr:0.000357 t:7.7s +tttg: c111/185 lr:0.000349 t:7.8s +tttg: c112/185 lr:0.000341 t:7.8s +tttg: c113/185 lr:0.000333 t:7.9s +tttg: c114/185 lr:0.000325 t:8.0s +tttg: c115/185 lr:0.000317 t:8.0s +tttg: c116/185 lr:0.000309 t:8.1s +tttg: c117/185 lr:0.000301 t:8.2s +tttg: c118/185 lr:0.000293 t:8.2s +tttg: c119/185 lr:0.000285 t:8.3s +tttg: c120/185 lr:0.000278 t:8.4s +tttg: c121/185 lr:0.000270 t:8.4s +tttg: c122/185 lr:0.000262 t:8.5s +tttg: c123/185 lr:0.000255 t:8.6s +tttg: c124/185 lr:0.000248 t:8.7s +tttg: c125/185 lr:0.000240 t:8.7s +tttg: c126/185 lr:0.000233 t:8.8s +tttg: c127/185 lr:0.000226 t:8.9s +tttg: c128/185 lr:0.000219 t:8.9s +tttg: c129/185 lr:0.000212 t:9.0s +tttg: c130/185 lr:0.000205 t:9.1s +tttg: c131/185 lr:0.000198 t:9.1s +tttg: c132/185 lr:0.000191 t:9.2s +tttg: c133/185 lr:0.000184 t:9.3s +tttg: c134/185 lr:0.000178 t:9.3s +tttg: c135/185 lr:0.000171 t:9.4s +tttg: c136/185 lr:0.000165 t:9.5s +tttg: c137/185 lr:0.000159 t:9.6s +tttg: c138/185 lr:0.000153 t:9.6s +tttg: c139/185 lr:0.000146 t:9.7s +tttg: c140/185 lr:0.000140 t:9.8s +tttg: c141/185 lr:0.000135 t:9.9s +tttg: c142/185 lr:0.000129 t:9.9s +tttg: c143/185 lr:0.000123 t:10.0s +tttg: c144/185 lr:0.000118 t:10.1s +tttg: c145/185 lr:0.000112 t:10.1s +tttg: c146/185 lr:0.000107 t:10.2s +tttg: c147/185 lr:0.000102 t:10.3s +tttg: c148/185 lr:0.000096 t:10.4s +tttg: c149/185 lr:0.000092 t:10.4s +tttg: c150/185 lr:0.000087 t:10.5s +tttg: c151/185 lr:0.000082 t:10.6s +tttg: c152/185 lr:0.000077 t:10.6s +tttg: c153/185 lr:0.000073 t:10.7s +tttg: c154/185 lr:0.000068 t:10.8s +tttg: c155/185 lr:0.000064 t:10.9s +tttg: c156/185 lr:0.000060 t:10.9s +tttg: c157/185 lr:0.000056 t:11.0s +tttg: c158/185 lr:0.000052 t:11.1s +tttg: c159/185 lr:0.000048 t:11.1s +tttg: c160/185 lr:0.000045 t:11.2s +tttg: c161/185 lr:0.000041 t:11.3s +tttg: c162/185 lr:0.000038 t:11.4s +tttg: c163/185 lr:0.000035 t:11.4s +tttg: c164/185 lr:0.000032 t:11.5s +tttg: c165/185 lr:0.000029 t:11.6s +tttg: c166/185 lr:0.000026 t:11.6s +tttg: c167/185 lr:0.000023 t:11.7s +tttg: c168/185 lr:0.000021 t:11.8s +tttg: c169/185 lr:0.000019 t:11.8s +tttg: c170/185 lr:0.000016 t:11.9s +tttg: c171/185 lr:0.000014 t:12.0s +tttg: c172/185 lr:0.000012 t:12.1s +tttg: c173/185 lr:0.000010 t:12.1s +tttg: c174/185 lr:0.000009 t:12.2s +tttg: c175/185 lr:0.000007 t:12.3s +tttg: c176/185 lr:0.000006 t:12.3s +tttg: c177/185 lr:0.000005 t:12.4s +tttg: c178/185 lr:0.000004 t:12.5s +tttg: c179/185 lr:0.000003 t:12.6s +tttg: c180/185 lr:0.000002 t:12.6s +tttg: c181/185 lr:0.000001 t:12.7s +tttg: c182/185 lr:0.000001 t:12.8s +tttg: c183/185 lr:0.000000 t:12.9s +tttg: c184/185 lr:0.000000 t:12.9s +ttpr: phase:2/3 t:259.4s +ttp: b746/782 bl:2.4147 bb:1.0640 rl:2.2710 rb:1.0533 dl:2884-2943 gd:0 +ttp: b745/782 bl:2.2377 bb:1.0244 rl:2.2680 rb:1.0507 dl:2842-2883 gd:0 +ttpp: phase:3/3 pd:2448 gd:2000 t:276.4s +tttg: c1/250 lr:0.001000 t:0.1s +tttg: c2/250 lr:0.001000 t:0.1s +tttg: c3/250 lr:0.001000 t:0.2s +tttg: c4/250 lr:0.001000 t:0.3s +tttg: c5/250 lr:0.000999 t:0.3s +tttg: c6/250 lr:0.000999 t:0.4s +tttg: c7/250 lr:0.000999 t:0.5s +tttg: c8/250 lr:0.000998 t:0.6s +tttg: c9/250 lr:0.000997 t:0.6s +tttg: c10/250 lr:0.000997 t:0.7s +tttg: c11/250 lr:0.000996 t:0.8s +tttg: c12/250 lr:0.000995 t:0.8s +tttg: c13/250 lr:0.000994 t:0.9s +tttg: c14/250 lr:0.000993 t:1.0s +tttg: c15/250 lr:0.000992 t:1.0s +tttg: c16/250 lr:0.000991 t:1.1s +tttg: c17/250 lr:0.000990 t:1.2s +tttg: c18/250 lr:0.000989 t:1.2s +tttg: c19/250 lr:0.000987 t:1.3s +tttg: c20/250 lr:0.000986 t:1.4s +tttg: c21/250 lr:0.000984 t:1.5s +tttg: c22/250 lr:0.000983 t:1.5s +tttg: c23/250 lr:0.000981 t:1.6s +tttg: c24/250 lr:0.000979 t:1.7s +tttg: c25/250 lr:0.000977 t:1.7s +tttg: c26/250 lr:0.000975 t:1.8s +tttg: c27/250 lr:0.000973 t:1.9s +tttg: c28/250 lr:0.000971 t:1.9s +tttg: c29/250 lr:0.000969 t:2.0s +tttg: c30/250 lr:0.000967 t:2.1s +tttg: c31/250 lr:0.000965 t:2.2s +tttg: c32/250 lr:0.000962 t:2.2s +tttg: c33/250 lr:0.000960 t:2.3s +tttg: c34/250 lr:0.000957 t:2.4s +tttg: c35/250 lr:0.000955 t:2.4s +tttg: c36/250 lr:0.000952 t:2.5s +tttg: c37/250 lr:0.000949 t:2.6s +tttg: c38/250 lr:0.000947 t:2.6s +tttg: c39/250 lr:0.000944 t:2.7s +tttg: c40/250 lr:0.000941 t:2.8s +tttg: c41/250 lr:0.000938 t:2.8s +tttg: c42/250 lr:0.000935 t:2.9s +tttg: c43/250 lr:0.000931 t:3.0s +tttg: c44/250 lr:0.000928 t:3.0s +tttg: c45/250 lr:0.000925 t:3.1s +tttg: c46/250 lr:0.000922 t:3.2s +tttg: c47/250 lr:0.000918 t:3.3s +tttg: c48/250 lr:0.000915 t:3.3s +tttg: c49/250 lr:0.000911 t:3.4s +tttg: c50/250 lr:0.000907 t:3.5s +tttg: c51/250 lr:0.000904 t:3.5s +tttg: c52/250 lr:0.000900 t:3.6s +tttg: c53/250 lr:0.000896 t:3.7s +tttg: c54/250 lr:0.000892 t:3.7s +tttg: c55/250 lr:0.000888 t:3.8s +tttg: c56/250 lr:0.000884 t:3.9s +tttg: c57/250 lr:0.000880 t:3.9s +tttg: c58/250 lr:0.000876 t:4.0s +tttg: c59/250 lr:0.000872 t:4.1s +tttg: c60/250 lr:0.000868 t:4.2s +tttg: c61/250 lr:0.000863 t:4.2s +tttg: c62/250 lr:0.000859 t:4.3s +tttg: c63/250 lr:0.000855 t:4.4s +tttg: c64/250 lr:0.000850 t:4.4s +tttg: c65/250 lr:0.000846 t:4.5s +tttg: c66/250 lr:0.000841 t:4.6s +tttg: c67/250 lr:0.000836 t:4.6s +tttg: c68/250 lr:0.000832 t:4.7s +tttg: c69/250 lr:0.000827 t:4.8s +tttg: c70/250 lr:0.000822 t:4.8s +tttg: c71/250 lr:0.000817 t:4.9s +tttg: c72/250 lr:0.000812 t:5.0s +tttg: c73/250 lr:0.000807 t:5.1s +tttg: c74/250 lr:0.000803 t:5.1s +tttg: c75/250 lr:0.000797 t:5.2s +tttg: c76/250 lr:0.000792 t:5.3s +tttg: c77/250 lr:0.000787 t:5.3s +tttg: c78/250 lr:0.000782 t:5.4s +tttg: c79/250 lr:0.000777 t:5.5s +tttg: c80/250 lr:0.000772 t:5.5s +tttg: c81/250 lr:0.000766 t:5.6s +tttg: c82/250 lr:0.000761 t:5.7s +tttg: c83/250 lr:0.000755 t:5.8s +tttg: c84/250 lr:0.000750 t:5.8s +tttg: c85/250 lr:0.000745 t:5.9s +tttg: c86/250 lr:0.000739 t:6.0s +tttg: c87/250 lr:0.000733 t:6.0s +tttg: c88/250 lr:0.000728 t:6.1s +tttg: c89/250 lr:0.000722 t:6.2s +tttg: c90/250 lr:0.000717 t:6.3s +tttg: c91/250 lr:0.000711 t:6.3s +tttg: c92/250 lr:0.000705 t:6.4s +tttg: c93/250 lr:0.000699 t:6.5s +tttg: c94/250 lr:0.000694 t:6.5s +tttg: c95/250 lr:0.000688 t:6.6s +tttg: c96/250 lr:0.000682 t:6.7s +tttg: c97/250 lr:0.000676 t:6.7s +tttg: c98/250 lr:0.000670 t:6.8s +tttg: c99/250 lr:0.000664 t:6.9s +tttg: c100/250 lr:0.000658 t:7.0s +tttg: c101/250 lr:0.000652 t:7.0s +tttg: c102/250 lr:0.000646 t:7.1s +tttg: c103/250 lr:0.000640 t:7.2s +tttg: c104/250 lr:0.000634 t:7.2s +tttg: c105/250 lr:0.000628 t:7.3s +tttg: c106/250 lr:0.000622 t:7.4s +tttg: c107/250 lr:0.000616 t:7.4s +tttg: c108/250 lr:0.000610 t:7.5s +tttg: c109/250 lr:0.000603 t:7.6s +tttg: c110/250 lr:0.000597 t:7.7s +tttg: c111/250 lr:0.000591 t:7.7s +tttg: c112/250 lr:0.000585 t:7.8s +tttg: c113/250 lr:0.000579 t:7.9s +tttg: c114/250 lr:0.000572 t:7.9s +tttg: c115/250 lr:0.000566 t:8.0s +tttg: c116/250 lr:0.000560 t:8.1s +tttg: c117/250 lr:0.000554 t:8.1s +tttg: c118/250 lr:0.000547 t:8.2s +tttg: c119/250 lr:0.000541 t:8.3s +tttg: c120/250 lr:0.000535 t:8.4s +tttg: c121/250 lr:0.000528 t:8.4s +tttg: c122/250 lr:0.000522 t:8.5s +tttg: c123/250 lr:0.000516 t:8.6s +tttg: c124/250 lr:0.000509 t:8.6s +tttg: c125/250 lr:0.000503 t:8.7s +tttg: c126/250 lr:0.000497 t:8.8s +tttg: c127/250 lr:0.000491 t:8.8s +tttg: c128/250 lr:0.000484 t:8.9s +tttg: c129/250 lr:0.000478 t:9.0s +tttg: c130/250 lr:0.000472 t:9.1s +tttg: c131/250 lr:0.000465 t:9.1s +tttg: c132/250 lr:0.000459 t:9.2s +tttg: c133/250 lr:0.000453 t:9.3s +tttg: c134/250 lr:0.000446 t:9.3s +tttg: c135/250 lr:0.000440 t:9.4s +tttg: c136/250 lr:0.000434 t:9.5s +tttg: c137/250 lr:0.000428 t:9.5s +tttg: c138/250 lr:0.000421 t:9.6s +tttg: c139/250 lr:0.000415 t:9.7s +tttg: c140/250 lr:0.000409 t:9.7s +tttg: c141/250 lr:0.000403 t:9.8s +tttg: c142/250 lr:0.000397 t:9.9s +tttg: c143/250 lr:0.000390 t:10.0s +tttg: c144/250 lr:0.000384 t:10.0s +tttg: c145/250 lr:0.000378 t:10.1s +tttg: c146/250 lr:0.000372 t:10.2s +tttg: c147/250 lr:0.000366 t:10.2s +tttg: c148/250 lr:0.000360 t:10.3s +tttg: c149/250 lr:0.000354 t:10.4s +tttg: c150/250 lr:0.000348 t:10.4s +tttg: c151/250 lr:0.000342 t:10.5s +tttg: c152/250 lr:0.000336 t:10.6s +tttg: c153/250 lr:0.000330 t:10.7s +tttg: c154/250 lr:0.000324 t:10.7s +tttg: c155/250 lr:0.000318 t:10.8s +tttg: c156/250 lr:0.000312 t:10.9s +tttg: c157/250 lr:0.000306 t:10.9s +tttg: c158/250 lr:0.000301 t:11.0s +tttg: c159/250 lr:0.000295 t:11.1s +tttg: c160/250 lr:0.000289 t:11.1s +tttg: c161/250 lr:0.000283 t:11.2s +tttg: c162/250 lr:0.000278 t:11.3s +tttg: c163/250 lr:0.000272 t:11.4s +tttg: c164/250 lr:0.000267 t:11.4s +tttg: c165/250 lr:0.000261 t:11.5s +tttg: c166/250 lr:0.000255 t:11.6s +tttg: c167/250 lr:0.000250 t:11.6s +tttg: c168/250 lr:0.000245 t:11.7s +tttg: c169/250 lr:0.000239 t:11.8s +tttg: c170/250 lr:0.000234 t:11.8s +tttg: c171/250 lr:0.000228 t:11.9s +tttg: c172/250 lr:0.000223 t:12.0s +tttg: c173/250 lr:0.000218 t:12.0s +tttg: c174/250 lr:0.000213 t:12.1s +tttg: c175/250 lr:0.000208 t:12.2s +tttg: c176/250 lr:0.000203 t:12.3s +tttg: c177/250 lr:0.000197 t:12.3s +tttg: c178/250 lr:0.000193 t:12.4s +tttg: c179/250 lr:0.000188 t:12.5s +tttg: c180/250 lr:0.000183 t:12.5s +tttg: c181/250 lr:0.000178 t:12.6s +tttg: c182/250 lr:0.000173 t:12.7s +tttg: c183/250 lr:0.000168 t:12.7s +tttg: c184/250 lr:0.000164 t:12.8s +tttg: c185/250 lr:0.000159 t:12.9s +tttg: c186/250 lr:0.000154 t:13.0s +tttg: c187/250 lr:0.000150 t:13.0s +tttg: c188/250 lr:0.000145 t:13.1s +tttg: c189/250 lr:0.000141 t:13.2s +tttg: c190/250 lr:0.000137 t:13.2s +tttg: c191/250 lr:0.000132 t:13.3s +tttg: c192/250 lr:0.000128 t:13.4s +tttg: c193/250 lr:0.000124 t:13.4s +tttg: c194/250 lr:0.000120 t:13.5s +tttg: c195/250 lr:0.000116 t:13.6s +tttg: c196/250 lr:0.000112 t:13.7s +tttg: c197/250 lr:0.000108 t:13.7s +tttg: c198/250 lr:0.000104 t:13.8s +tttg: c199/250 lr:0.000100 t:13.9s +tttg: c200/250 lr:0.000096 t:15.5s +tttg: c201/250 lr:0.000093 t:15.6s +tttg: c202/250 lr:0.000089 t:15.7s +tttg: c203/250 lr:0.000085 t:15.8s +tttg: c204/250 lr:0.000082 t:15.8s +tttg: c205/250 lr:0.000078 t:15.9s +tttg: c206/250 lr:0.000075 t:16.0s +tttg: c207/250 lr:0.000072 t:16.0s +tttg: c208/250 lr:0.000069 t:16.1s +tttg: c209/250 lr:0.000065 t:16.2s +tttg: c210/250 lr:0.000062 t:16.2s +tttg: c211/250 lr:0.000059 t:16.3s +tttg: c212/250 lr:0.000056 t:16.4s +tttg: c213/250 lr:0.000053 t:16.5s +tttg: c214/250 lr:0.000051 t:16.5s +tttg: c215/250 lr:0.000048 t:16.6s +tttg: c216/250 lr:0.000045 t:16.7s +tttg: c217/250 lr:0.000043 t:16.7s +tttg: c218/250 lr:0.000040 t:16.8s +tttg: c219/250 lr:0.000038 t:16.9s +tttg: c220/250 lr:0.000035 t:16.9s +tttg: c221/250 lr:0.000033 t:17.0s +tttg: c222/250 lr:0.000031 t:17.1s +tttg: c223/250 lr:0.000029 t:17.1s +tttg: c224/250 lr:0.000027 t:17.2s +tttg: c225/250 lr:0.000025 t:17.3s +tttg: c226/250 lr:0.000023 t:17.4s +tttg: c227/250 lr:0.000021 t:17.4s +tttg: c228/250 lr:0.000019 t:17.5s +tttg: c229/250 lr:0.000017 t:17.6s +tttg: c230/250 lr:0.000016 t:17.6s +tttg: c231/250 lr:0.000014 t:17.7s +tttg: c232/250 lr:0.000013 t:17.8s +tttg: c233/250 lr:0.000011 t:17.8s +tttg: c234/250 lr:0.000010 t:17.9s +tttg: c235/250 lr:0.000009 t:18.0s +tttg: c236/250 lr:0.000008 t:18.1s +tttg: c237/250 lr:0.000007 t:18.1s +tttg: c238/250 lr:0.000006 t:18.2s +tttg: c239/250 lr:0.000005 t:18.3s +tttg: c240/250 lr:0.000004 t:18.3s +tttg: c241/250 lr:0.000003 t:18.4s +tttg: c242/250 lr:0.000003 t:18.5s +tttg: c243/250 lr:0.000002 t:18.6s +tttg: c244/250 lr:0.000001 t:18.6s +tttg: c245/250 lr:0.000001 t:18.7s +tttg: c246/250 lr:0.000001 t:18.8s +tttg: c247/250 lr:0.000000 t:18.8s +tttg: c248/250 lr:0.000000 t:18.9s +tttg: c249/250 lr:0.000000 t:19.0s +ttpr: phase:3/3 t:297.2s +ttp: b736/782 bl:2.2463 bb:1.0583 rl:2.2664 rb:1.0512 dl:2526-2550 gd:1 +ttp: b735/782 bl:2.3914 bb:1.1001 rl:2.2749 rb:1.0546 dl:2495-2526 gd:1 +ttp: b725/782 bl:2.3227 bb:1.0448 rl:2.2776 rb:1.0540 dl:2232-2254 gd:1 +ttp: b718/782 bl:2.2950 bb:1.0301 rl:2.2785 rb:1.0528 dl:2089-2106 gd:1 +ttp: b708/782 bl:2.3118 bb:1.0341 rl:2.2800 rb:1.0519 dl:1924-1937 gd:1 +ttp: b698/782 bl:2.2498 bb:1.0298 rl:2.2788 rb:1.0510 dl:1803-1814 gd:1 +ttp: b693/782 bl:2.3372 bb:1.0499 rl:2.2810 rb:1.0510 dl:1746-1757 gd:1 +ttp: b684/782 bl:2.3758 bb:1.0467 rl:2.2842 rb:1.0508 dl:1658-1665 gd:1 +ttp: b677/782 bl:2.3070 bb:1.0336 rl:2.2849 rb:1.0502 dl:1595-1601 gd:1 +ttp: b665/782 bl:2.3346 bb:1.0488 rl:2.2864 rb:1.0502 dl:1500-1507 gd:1 +ttp: b659/782 bl:2.3087 bb:1.0419 rl:2.2870 rb:1.0500 dl:1459-1466 gd:1 +ttp: b652/782 bl:2.2516 bb:1.0235 rl:2.2861 rb:1.0493 dl:1411-1419 gd:1 +ttp: b643/782 bl:2.3583 bb:1.0270 rl:2.2878 rb:1.0487 dl:1356-1362 gd:1 +ttp: b635/782 bl:2.3449 bb:1.0582 rl:2.2891 rb:1.0489 dl:1308-1314 gd:1 +ttp: b626/782 bl:2.3144 bb:1.0284 rl:2.2897 rb:1.0485 dl:1260-1265 gd:1 +ttp: b618/782 bl:2.4159 bb:1.0753 rl:2.2922 rb:1.0490 dl:1216-1221 gd:1 +ttp: b610/782 bl:2.2558 bb:1.0087 rl:2.2915 rb:1.0482 dl:1177-1182 gd:1 +ttp: b602/782 bl:2.3797 bb:1.0497 rl:2.2932 rb:1.0483 dl:1141-1146 gd:1 +ttp: b597/782 bl:2.3704 bb:1.0541 rl:2.2945 rb:1.0484 dl:1119-1124 gd:1 +ttp: b588/782 bl:2.3230 bb:1.0455 rl:2.2950 rb:1.0483 dl:1081-1086 gd:1 +ttp: b580/782 bl:2.3138 bb:1.0151 rl:2.2953 rb:1.0478 dl:1048-1052 gd:1 +ttp: b573/782 bl:2.3662 bb:1.0666 rl:2.2964 rb:1.0481 dl:1021-1025 gd:1 +ttp: b565/782 bl:2.3858 bb:1.0336 rl:2.2977 rb:1.0478 dl:993-997 gd:1 +ttp: b557/782 bl:2.3422 bb:1.0521 rl:2.2984 rb:1.0479 dl:965-968 gd:1 +ttp: b547/782 bl:2.3355 bb:1.0497 rl:2.2989 rb:1.0479 dl:934-937 gd:1 +ttp: b538/782 bl:2.3395 bb:1.0474 rl:2.2994 rb:1.0479 dl:905-909 gd:1 +ttp: b535/782 bl:2.3811 bb:1.0327 rl:2.3004 rb:1.0477 dl:896-899 gd:1 +ttp: b527/782 bl:2.3429 bb:1.0283 rl:2.3009 rb:1.0475 dl:872-875 gd:1 +ttp: b516/782 bl:2.3570 bb:1.0459 rl:2.3016 rb:1.0474 dl:841-843 gd:1 +ttp: b508/782 bl:2.3969 bb:1.0538 rl:2.3026 rb:1.0475 dl:817-820 gd:1 +ttp: b498/782 bl:2.3561 bb:1.0529 rl:2.3032 rb:1.0476 dl:791-794 gd:1 +ttp: b490/782 bl:2.3910 bb:1.0559 rl:2.3041 rb:1.0477 dl:771-773 gd:1 +ttp: b482/782 bl:2.3297 bb:1.0474 rl:2.3044 rb:1.0477 dl:752-754 gd:1 +ttp: b474/782 bl:2.3473 bb:1.0748 rl:2.3048 rb:1.0479 dl:733-735 gd:1 +ttp: b466/782 bl:2.3896 bb:1.0302 rl:2.3056 rb:1.0478 dl:714-717 gd:1 +ttp: b459/782 bl:2.2859 bb:1.0466 rl:2.3054 rb:1.0477 dl:700-701 gd:1 +ttp: b451/782 bl:2.4042 bb:1.0879 rl:2.3062 rb:1.0481 dl:682-685 gd:1 +ttp: b443/782 bl:2.2312 bb:1.0498 rl:2.3056 rb:1.0481 dl:666-668 gd:1 +ttp: b436/782 bl:2.2731 bb:1.0500 rl:2.3054 rb:1.0481 dl:651-653 gd:1 +ttp: b431/782 bl:2.3762 bb:1.0541 rl:2.3059 rb:1.0482 dl:642-643 gd:1 +ttp: b424/782 bl:2.3483 bb:1.0648 rl:2.3062 rb:1.0483 dl:629-630 gd:1 +ttp: b418/782 bl:2.2851 bb:1.0376 rl:2.3061 rb:1.0482 dl:617-618 gd:1 +ttp: b410/782 bl:2.3200 bb:1.0186 rl:2.3062 rb:1.0480 dl:601-603 gd:1 +ttp: b402/782 bl:2.2423 bb:0.9979 rl:2.3057 rb:1.0476 dl:586-588 gd:1 +ttp: b394/782 bl:2.2544 bb:0.9925 rl:2.3054 rb:1.0472 dl:571-573 gd:1 +ttp: b386/782 bl:2.3424 bb:1.1000 rl:2.3056 rb:1.0476 dl:557-559 gd:1 +ttp: b378/782 bl:2.4302 bb:1.0545 rl:2.3064 rb:1.0476 dl:544-545 gd:1 +ttp: b370/782 bl:2.3723 bb:1.0860 rl:2.3068 rb:1.0479 dl:530-532 gd:1 +ttp: b363/782 bl:2.3830 bb:1.0666 rl:2.3073 rb:1.0480 dl:518-521 gd:1 +ttp: b356/782 bl:2.3445 bb:1.0557 rl:2.3075 rb:1.0480 dl:506-508 gd:1 +ttp: b348/782 bl:2.3603 bb:1.0586 rl:2.3078 rb:1.0481 dl:494-495 gd:1 +ttp: b340/782 bl:2.4576 bb:1.0804 rl:2.3086 rb:1.0483 dl:482-483 gd:1 +ttp: b332/782 bl:2.3101 bb:1.0457 rl:2.3087 rb:1.0483 dl:469-471 gd:1 +ttp: b323/782 bl:2.3866 bb:1.0775 rl:2.3091 rb:1.0484 dl:457-458 gd:1 +ttp: b315/782 bl:2.4012 bb:1.1031 rl:2.3095 rb:1.0487 dl:444-445 gd:1 +ttp: b307/782 bl:2.3346 bb:1.1286 rl:2.3096 rb:1.0490 dl:432-433 gd:1 +ttp: b301/782 bl:2.3573 bb:1.0943 rl:2.3099 rb:1.0493 dl:422-424 gd:1 +ttp: b293/782 bl:2.4392 bb:1.0998 rl:2.3105 rb:1.0495 dl:410-412 gd:1 +ttp: b285/782 bl:2.3703 bb:1.0798 rl:2.3107 rb:1.0496 dl:399-400 gd:1 +ttp: b277/782 bl:2.2669 bb:1.0675 rl:2.3105 rb:1.0497 dl:388-389 gd:1 +ttp: b268/782 bl:2.3598 bb:1.0781 rl:2.3107 rb:1.0498 dl:376-378 gd:1 +ttp: b259/782 bl:2.3385 bb:1.0967 rl:2.3108 rb:1.0500 dl:365-366 gd:1 +ttp: b251/782 bl:2.3737 bb:1.0974 rl:2.3111 rb:1.0502 dl:355-356 gd:1 +ttp: b242/782 bl:2.3807 bb:1.1020 rl:2.3113 rb:1.0504 dl:344-345 gd:1 +ttp: b234/782 bl:2.4143 bb:1.1440 rl:2.3117 rb:1.0507 dl:334-335 gd:1 +ttp: b201/782 bl:2.2932 bb:1.0939 rl:2.3117 rb:1.0508 dl:297-298 gd:1 +ttp: b192/782 bl:2.3792 bb:1.1555 rl:2.3119 rb:1.0511 dl:286-288 gd:1 +ttp: b184/782 bl:2.3965 bb:1.1298 rl:2.3121 rb:1.0513 dl:278-279 gd:1 +ttp: b175/782 bl:2.3992 bb:1.1593 rl:2.3124 rb:1.0516 dl:269-270 gd:1 +ttp: b166/782 bl:2.4746 bb:1.1061 rl:2.3128 rb:1.0518 dl:260-262 gd:1 +ttp: b157/782 bl:2.3694 bb:1.1348 rl:2.3130 rb:1.0520 dl:252-253 gd:1 +ttp: b149/782 bl:2.3663 bb:1.1526 rl:2.3131 rb:1.0522 dl:244-245 gd:1 +ttp: b141/782 bl:2.4713 bb:1.1278 rl:2.3135 rb:1.0524 dl:236-237 gd:1 +ttp: b133/782 bl:2.3664 bb:1.1350 rl:2.3136 rb:1.0526 dl:229-230 gd:1 +ttp: b124/782 bl:2.3901 bb:1.1675 rl:2.3138 rb:1.0529 dl:220-222 gd:1 +ttp: b118/782 bl:2.4712 bb:1.1280 rl:2.3142 rb:1.0530 dl:215-216 gd:1 +ttp: b112/782 bl:2.4764 bb:1.1819 rl:2.3145 rb:1.0533 dl:210-210 gd:1 +ttp: b104/782 bl:2.4991 bb:1.1798 rl:2.3149 rb:1.0536 dl:202-203 gd:1 +ttp: b97/782 bl:2.4556 bb:1.1622 rl:2.3152 rb:1.0538 dl:196-197 gd:1 +ttp: b89/782 bl:2.5002 bb:1.1553 rl:2.3156 rb:1.0540 dl:189-190 gd:1 +ttp: b81/782 bl:2.4694 bb:1.1207 rl:2.3158 rb:1.0541 dl:182-183 gd:1 +ttp: b73/782 bl:2.5522 bb:1.2527 rl:2.3163 rb:1.0544 dl:174-175 gd:1 +ttp: b66/782 bl:2.6471 bb:1.2388 rl:2.3168 rb:1.0547 dl:169-169 gd:1 +ttp: b56/782 bl:2.5427 bb:1.2188 rl:2.3172 rb:1.0550 dl:159-160 gd:1 +ttp: b48/782 bl:2.5187 bb:1.2144 rl:2.3175 rb:1.0552 dl:151-152 gd:1 +ttp: b40/782 bl:2.4833 bb:1.1504 rl:2.3178 rb:1.0554 dl:143-144 gd:1 +ttp: b32/782 bl:2.6153 bb:1.2194 rl:2.3182 rb:1.0556 dl:135-136 gd:1 +ttp: b23/782 bl:2.6064 bb:1.2241 rl:2.3186 rb:1.0558 dl:126-127 gd:1 +ttp: b15/782 bl:2.6523 bb:1.2317 rl:2.3189 rb:1.0560 dl:115-117 gd:1 +ttp: b7/782 bl:2.7490 bb:1.2372 rl:2.3194 rb:1.0562 dl:101-103 gd:1 +quantized_ttt_phased val_loss:2.32368487 val_bpb:1.06183331 eval_time:406521ms +total_eval_time:406.5s diff --git a/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed42.log b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed42.log new file mode 100644 index 0000000000..acaf822964 --- /dev/null +++ b/records/track_10min_16mb/2026-04-24_SP8192_CaseOps_SparseGate_QuantGate_Loop45_PhasedTTT_PolarNS_MinLR_FusedCE_UpdatedCarry/train_seed42.log @@ -0,0 +1,934 @@ +W0423 22:39:32.672000 145471 torch/distributed/run.py:803] +W0423 22:39:32.672000 145471 torch/distributed/run.py:803] ***************************************** +W0423 22:39:32.672000 145471 torch/distributed/run.py:803] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0423 22:39:32.672000 145471 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: /workspace/runs/036-035e-8h-promotion/seed_42 + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + beta1: 0.9 + beta2: 0.95 + caseops_enabled: True + compressor: brotli + data_dir: /workspace/parameter-golf/data + datasets_dir: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2048 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.005 + gated_attn_quant_gate: True + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 0.5 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: /workspace/runs/036-035e-8h-promotion/seed_42/e75febaa-ac16-46e3-acbc-087c22dd08e3.txt + logit_softcap: 30.0 + loop_depth_upgrade_at: 0.0 + loop_end: 5 + loop_start: 3 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: /workspace/runs/036-035e-8h-promotion/seed_42/final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 3 + phased_ttt_prefix_docs: 2000 + qk_gain_init: 5.0 + quantized_model_path: /workspace/runs/036-035e-8h-promotion/seed_42/final_model.int6.ptz + rank: 0 + recur_alpha_enabled: True + recur_diag_p2p_cos: False + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: e75febaa-ac16-46e3-acbc-087c22dd08e3 + scalar_lr: 0.02 + seed: 42 + skip_gates_enabled: True + smear_gate_enabled: False + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 1.0 + spinquant_enabled: False + spinquant_seed: 42 + spinquant_sites: attn_in,attn_proj_in,mlp_in,mlp_proj_in + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_train_*.bin + train_log_every: 100 + train_seq_len: 2048 + ttt_batch_size: 64 + ttt_beta1: 0.0 + ttt_beta2: 0.999 + ttt_chunk_size: 48 + ttt_enabled: True + ttt_eval_batches: + ttt_eval_seq_len: 2048 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_alpha: 144 + ttt_lora_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 1.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 0 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 47851520 +model_params:35945658 +recur_alpha: enabled=True num_loops=2 loop_start=3 loop_end=5 diag_p2p_cos=False +gptq:reserving 0s, effective=599500ms +warmup_cu_buckets:64,128,192,256 iters_each:3 +warmup_step: 1/20 +warmup_step: 2/20 +warmup_step: 3/20 +warmup_step: 4/20 +warmup_step: 5/20 +warmup_step: 6/20 +warmup_step: 10/20 +warmup_step: 20/20 +loop_warmup:enabled encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +loop_warmup_step: 1/20 +loop_warmup_step: 2/20 +loop_warmup_step: 3/20 +loop_warmup_step: 4/20 +loop_warmup_step: 5/20 +loop_warmup_step: 6/20 +loop_warmup_step: 10/20 +loop_warmup_step: 20/20 +1/20000 train_loss: 9.0168 train_time: 0.0m tok/s: 12785698 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2/20000 train_loss: 12.8636 train_time: 0.0m tok/s: 7648299 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3/20000 train_loss: 10.2423 train_time: 0.0m tok/s: 7629634 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4/20000 train_loss: 8.7625 train_time: 0.0m tok/s: 7656935 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +5/20000 train_loss: 8.0144 train_time: 0.0m tok/s: 7657453 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +100/20000 train_loss: 3.6336 train_time: 0.2m tok/s: 8427902 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +200/20000 train_loss: 3.1446 train_time: 0.3m tok/s: 8348721 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +300/20000 train_loss: 2.9201 train_time: 0.5m tok/s: 8333068 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +400/20000 train_loss: 2.5840 train_time: 0.6m tok/s: 8319550 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +500/20000 train_loss: 2.5802 train_time: 0.8m tok/s: 8342231 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +600/20000 train_loss: 2.6791 train_time: 0.9m tok/s: 8332819 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +700/20000 train_loss: 2.8796 train_time: 1.1m tok/s: 8326420 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +800/20000 train_loss: 2.7150 train_time: 1.3m tok/s: 8318389 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +900/20000 train_loss: 2.7603 train_time: 1.4m tok/s: 8317397 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1000/20000 train_loss: 2.8126 train_time: 1.6m tok/s: 8327956 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1100/20000 train_loss: 2.7757 train_time: 1.7m tok/s: 8321685 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1200/20000 train_loss: 2.7716 train_time: 1.9m tok/s: 8318258 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1300/20000 train_loss: 2.8392 train_time: 2.0m tok/s: 8315040 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1400/20000 train_loss: 2.5971 train_time: 2.2m tok/s: 8311573 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1500/20000 train_loss: 2.6449 train_time: 2.4m tok/s: 8319236 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1600/20000 train_loss: 2.7156 train_time: 2.5m tok/s: 8316749 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1700/20000 train_loss: 2.6935 train_time: 2.7m tok/s: 8314151 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1800/20000 train_loss: 2.6543 train_time: 2.8m tok/s: 8312328 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +1900/20000 train_loss: 2.7530 train_time: 3.0m tok/s: 8318143 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2000/20000 train_loss: 2.6717 train_time: 3.2m tok/s: 8315645 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2100/20000 train_loss: 2.6977 train_time: 3.3m tok/s: 8314212 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2200/20000 train_loss: 2.5385 train_time: 3.5m tok/s: 8312192 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +layer_loop:enabled step:2218 frac:0.350 depth:3 encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +2300/20000 train_loss: 2.6160 train_time: 3.7m tok/s: 8171686 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2400/20000 train_loss: 2.6368 train_time: 3.9m tok/s: 8024439 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2500/20000 train_loss: 2.5592 train_time: 4.2m tok/s: 7889734 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2600/20000 train_loss: 2.5258 train_time: 4.4m tok/s: 7769062 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2700/20000 train_loss: 2.5149 train_time: 4.6m tok/s: 7661531 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2800/20000 train_loss: 2.5748 train_time: 4.9m tok/s: 7565097 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +2900/20000 train_loss: 2.5499 train_time: 5.1m tok/s: 7478504 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3000/20000 train_loss: 2.5740 train_time: 5.3m tok/s: 7397182 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3100/20000 train_loss: 2.5060 train_time: 5.5m tok/s: 7322692 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3200/20000 train_loss: 2.4783 train_time: 5.8m tok/s: 7254698 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3300/20000 train_loss: 2.6665 train_time: 6.0m tok/s: 7194160 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3400/20000 train_loss: 2.5669 train_time: 6.2m tok/s: 7135135 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3500/20000 train_loss: 2.5776 train_time: 6.5m tok/s: 7081014 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3600/20000 train_loss: 2.4741 train_time: 6.7m tok/s: 7030519 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3700/20000 train_loss: 2.5608 train_time: 7.0m tok/s: 6949972 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3800/20000 train_loss: 2.5064 train_time: 7.2m tok/s: 6909441 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +3900/20000 train_loss: 2.6304 train_time: 7.4m tok/s: 6869734 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4000/20000 train_loss: 2.4203 train_time: 7.7m tok/s: 6832479 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4100/20000 train_loss: 2.4229 train_time: 7.9m tok/s: 6797076 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4200/20000 train_loss: 2.4368 train_time: 8.1m tok/s: 6763111 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4300/20000 train_loss: 2.5136 train_time: 8.4m tok/s: 6715365 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4400/20000 train_loss: 2.4612 train_time: 8.6m tok/s: 6686101 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4500/20000 train_loss: 2.2934 train_time: 8.9m tok/s: 6658473 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4600/20000 train_loss: 2.3888 train_time: 9.1m tok/s: 6632392 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4700/20000 train_loss: 2.3347 train_time: 9.3m tok/s: 6608101 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4800/20000 train_loss: 2.3132 train_time: 9.6m tok/s: 6585338 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4900/20000 train_loss: 2.3087 train_time: 9.8m tok/s: 6562880 +recur_alpha: beta=[1.5625, 1.8515625, 2.125] alpha=[[0.23046875, 0.0400390625, 0.030029296875], [0.1298828125, -0.33984375, 0.010009765625], [0.06005859375, 0.1904296875, -0.02001953125]] grad_norm=0.000000 +4989/20000 val_loss: 2.3587 val_bpb: 1.0778 +stopping_early: wallclock_cap train_time: 599631ms step: 4989/20000 +peak memory allocated: 41610 MiB reserved: 46918 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.33621683 val_bpb:1.06749061 eval_time:7600ms +Serialized model: 135417469 bytes +pyminify unavailable (FileNotFoundError); skipping compressed-code-size measurement +Code size (uncompressed): 175063 bytes +Code size (compressed): 0 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 3.6s +Quantized weights: + gptq (int6): blocks.attn.c_k.weight, blocks.attn.c_q.weight, blocks.attn.c_v.weight, blocks.attn.proj.weight, blocks.mlp.fc.weight, blocks.mlp.proj.weight + gptq (int7): tok_emb.weight + passthrough (float16): blocks.attn.attn_gate_w, blocks.attn.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, recur_alpha, recur_beta, skip_gates, skip_weights +Serialized model quantized+brotli: 15909254 bytes +Total submission size quantized+brotli: 15909254 bytes +diagnostic quantized val_loss:2.35655108 val_bpb:1.07678196 eval_time:11855ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (148.7s) + +beginning TTT eval timer +ttt_phased: total_docs:50000 prefix_docs:2000 suffix_docs:48000 num_phases:3 boundaries:[666, 1333, 2000] +ttp: b775/782 bl:2.2839 bb:1.0618 rl:2.2839 rb:1.0618 dl:6892-7524 gd:0 +ttp: b774/782 bl:2.2945 bb:1.0683 rl:2.2890 rb:1.0649 dl:6447-6872 gd:0 +ttp: b768/782 bl:2.2491 bb:1.0474 rl:2.2785 rb:1.0603 dl:4859-5083 gd:0 +ttpp: phase:1/3 pd:1104 gd:666 t:222.2s +tttg: c1/111 lr:0.001000 t:0.4s +tttg: c2/111 lr:0.001000 t:0.5s +tttg: c3/111 lr:0.000999 t:0.5s +tttg: c4/111 lr:0.000998 t:0.6s +tttg: c5/111 lr:0.000997 t:0.7s +tttg: c6/111 lr:0.000995 t:0.8s +tttg: c7/111 lr:0.000993 t:0.8s +tttg: c8/111 lr:0.000990 t:0.9s +tttg: c9/111 lr:0.000987 t:1.0s +tttg: c10/111 lr:0.000984 t:1.0s +tttg: c11/111 lr:0.000980 t:1.1s +tttg: c12/111 lr:0.000976 t:1.2s +tttg: c13/111 lr:0.000971 t:1.2s +tttg: c14/111 lr:0.000966 t:1.3s +tttg: c15/111 lr:0.000961 t:1.4s +tttg: c16/111 lr:0.000955 t:1.4s +tttg: c17/111 lr:0.000949 t:1.5s +tttg: c18/111 lr:0.000942 t:1.6s +tttg: c19/111 lr:0.000935 t:1.6s +tttg: c20/111 lr:0.000928 t:1.7s +tttg: c21/111 lr:0.000921 t:1.8s +tttg: c22/111 lr:0.000913 t:1.8s +tttg: c23/111 lr:0.000905 t:1.9s +tttg: c24/111 lr:0.000896 t:2.0s +tttg: c25/111 lr:0.000887 t:2.0s +tttg: c26/111 lr:0.000878 t:2.1s +tttg: c27/111 lr:0.000868 t:2.2s +tttg: c28/111 lr:0.000859 t:2.2s +tttg: c29/111 lr:0.000848 t:2.3s +tttg: c30/111 lr:0.000838 t:2.4s +tttg: c31/111 lr:0.000827 t:2.4s +tttg: c32/111 lr:0.000817 t:2.5s +tttg: c33/111 lr:0.000805 t:2.6s +tttg: c34/111 lr:0.000794 t:2.6s +tttg: c35/111 lr:0.000782 t:2.7s +tttg: c36/111 lr:0.000770 t:2.8s +tttg: c37/111 lr:0.000758 t:2.8s +tttg: c38/111 lr:0.000746 t:4.7s +tttg: c39/111 lr:0.000733 t:4.8s +tttg: c40/111 lr:0.000721 t:4.8s +tttg: c41/111 lr:0.000708 t:4.9s +tttg: c42/111 lr:0.000695 t:5.0s +tttg: c43/111 lr:0.000681 t:5.0s +tttg: c44/111 lr:0.000668 t:5.1s +tttg: c45/111 lr:0.000655 t:5.2s +tttg: c46/111 lr:0.000641 t:5.2s +tttg: c47/111 lr:0.000627 t:5.3s +tttg: c48/111 lr:0.000613 t:5.4s +tttg: c49/111 lr:0.000599 t:5.4s +tttg: c50/111 lr:0.000585 t:5.5s +tttg: c51/111 lr:0.000571 t:5.6s +tttg: c52/111 lr:0.000557 t:5.6s +tttg: c53/111 lr:0.000543 t:5.7s +tttg: c54/111 lr:0.000529 t:5.8s +tttg: c55/111 lr:0.000514 t:5.8s +tttg: c56/111 lr:0.000500 t:5.9s +tttg: c57/111 lr:0.000486 t:6.0s +tttg: c58/111 lr:0.000471 t:6.0s +tttg: c59/111 lr:0.000457 t:6.1s +tttg: c60/111 lr:0.000443 t:6.2s +tttg: c61/111 lr:0.000429 t:6.2s +tttg: c62/111 lr:0.000415 t:6.3s +tttg: c63/111 lr:0.000401 t:6.4s +tttg: c64/111 lr:0.000387 t:6.4s +tttg: c65/111 lr:0.000373 t:6.5s +tttg: c66/111 lr:0.000359 t:6.6s +tttg: c67/111 lr:0.000345 t:6.6s +tttg: c68/111 lr:0.000332 t:6.7s +tttg: c69/111 lr:0.000319 t:6.8s +tttg: c70/111 lr:0.000305 t:6.8s +tttg: c71/111 lr:0.000292 t:6.9s +tttg: c72/111 lr:0.000279 t:7.0s +tttg: c73/111 lr:0.000267 t:7.0s +tttg: c74/111 lr:0.000254 t:7.1s +tttg: c75/111 lr:0.000242 t:7.2s +tttg: c76/111 lr:0.000230 t:7.2s +tttg: c77/111 lr:0.000218 t:7.3s +tttg: c78/111 lr:0.000206 t:7.4s +tttg: c79/111 lr:0.000195 t:7.4s +tttg: c80/111 lr:0.000183 t:7.5s +tttg: c81/111 lr:0.000173 t:7.6s +tttg: c82/111 lr:0.000162 t:7.6s +tttg: c83/111 lr:0.000152 t:7.7s +tttg: c84/111 lr:0.000141 t:7.8s +tttg: c85/111 lr:0.000132 t:7.9s +tttg: c86/111 lr:0.000122 t:7.9s +tttg: c87/111 lr:0.000113 t:8.0s +tttg: c88/111 lr:0.000104 t:8.1s +tttg: c89/111 lr:0.000095 t:8.1s +tttg: c90/111 lr:0.000087 t:8.2s +tttg: c91/111 lr:0.000079 t:8.3s +tttg: c92/111 lr:0.000072 t:8.3s +tttg: c93/111 lr:0.000065 t:8.4s +tttg: c94/111 lr:0.000058 t:8.5s +tttg: c95/111 lr:0.000051 t:8.5s +tttg: c96/111 lr:0.000045 t:8.6s +tttg: c97/111 lr:0.000039 t:8.7s +tttg: c98/111 lr:0.000034 t:8.7s +tttg: c99/111 lr:0.000029 t:8.8s +tttg: c100/111 lr:0.000024 t:8.9s +tttg: c101/111 lr:0.000020 t:8.9s +tttg: c102/111 lr:0.000016 t:9.0s +tttg: c103/111 lr:0.000013 t:9.1s +tttg: c104/111 lr:0.000010 t:9.1s +tttg: c105/111 lr:0.000007 t:9.2s +tttg: c106/111 lr:0.000005 t:9.3s +tttg: c107/111 lr:0.000003 t:9.3s +tttg: c108/111 lr:0.000002 t:9.4s +tttg: c109/111 lr:0.000001 t:9.5s +tttg: c110/111 lr:0.000000 t:9.5s +ttpr: phase:1/3 t:233.6s +ttp: b757/782 bl:2.2864 bb:1.0643 rl:2.2797 rb:1.0609 dl:3550-3633 gd:0 +ttpp: phase:2/3 pd:1808 gd:1333 t:349.5s +tttg: c1/185 lr:0.001000 t:0.1s +tttg: c2/185 lr:0.001000 t:0.1s +tttg: c3/185 lr:0.001000 t:0.2s +tttg: c4/185 lr:0.000999 t:0.3s +tttg: c5/185 lr:0.000999 t:0.3s +tttg: c6/185 lr:0.000998 t:0.4s +tttg: c7/185 lr:0.000997 t:0.5s +tttg: c8/185 lr:0.000996 t:0.6s +tttg: c9/185 lr:0.000995 t:0.6s +tttg: c10/185 lr:0.000994 t:0.7s +tttg: c11/185 lr:0.000993 t:0.8s +tttg: c12/185 lr:0.000991 t:0.8s +tttg: c13/185 lr:0.000990 t:0.9s +tttg: c14/185 lr:0.000988 t:1.0s +tttg: c15/185 lr:0.000986 t:1.0s +tttg: c16/185 lr:0.000984 t:1.1s +tttg: c17/185 lr:0.000981 t:1.2s +tttg: c18/185 lr:0.000979 t:1.2s +tttg: c19/185 lr:0.000977 t:1.3s +tttg: c20/185 lr:0.000974 t:1.4s +tttg: c21/185 lr:0.000971 t:1.4s +tttg: c22/185 lr:0.000968 t:1.5s +tttg: c23/185 lr:0.000965 t:1.6s +tttg: c24/185 lr:0.000962 t:1.6s +tttg: c25/185 lr:0.000959 t:1.7s +tttg: c26/185 lr:0.000955 t:1.8s +tttg: c27/185 lr:0.000952 t:1.8s +tttg: c28/185 lr:0.000948 t:1.9s +tttg: c29/185 lr:0.000944 t:2.0s +tttg: c30/185 lr:0.000940 t:2.0s +tttg: c31/185 lr:0.000936 t:2.1s +tttg: c32/185 lr:0.000932 t:2.2s +tttg: c33/185 lr:0.000927 t:2.2s +tttg: c34/185 lr:0.000923 t:2.3s +tttg: c35/185 lr:0.000918 t:2.4s +tttg: c36/185 lr:0.000913 t:2.4s +tttg: c37/185 lr:0.000908 t:2.5s +tttg: c38/185 lr:0.000904 t:2.6s +tttg: c39/185 lr:0.000898 t:2.6s +tttg: c40/185 lr:0.000893 t:2.7s +tttg: c41/185 lr:0.000888 t:2.8s +tttg: c42/185 lr:0.000882 t:2.8s +tttg: c43/185 lr:0.000877 t:2.9s +tttg: c44/185 lr:0.000871 t:3.0s +tttg: c45/185 lr:0.000865 t:3.0s +tttg: c46/185 lr:0.000860 t:3.1s +tttg: c47/185 lr:0.000854 t:3.2s +tttg: c48/185 lr:0.000847 t:3.2s +tttg: c49/185 lr:0.000841 t:3.3s +tttg: c50/185 lr:0.000835 t:3.4s +tttg: c51/185 lr:0.000829 t:3.4s +tttg: c52/185 lr:0.000822 t:3.5s +tttg: c53/185 lr:0.000816 t:3.6s +tttg: c54/185 lr:0.000809 t:3.6s +tttg: c55/185 lr:0.000802 t:3.7s +tttg: c56/185 lr:0.000795 t:3.8s +tttg: c57/185 lr:0.000788 t:3.9s +tttg: c58/185 lr:0.000781 t:3.9s +tttg: c59/185 lr:0.000774 t:4.0s +tttg: c60/185 lr:0.000767 t:4.1s +tttg: c61/185 lr:0.000760 t:4.1s +tttg: c62/185 lr:0.000752 t:4.2s +tttg: c63/185 lr:0.000745 t:4.3s +tttg: c64/185 lr:0.000738 t:4.3s +tttg: c65/185 lr:0.000730 t:4.4s +tttg: c66/185 lr:0.000722 t:4.5s +tttg: c67/185 lr:0.000715 t:4.5s +tttg: c68/185 lr:0.000707 t:4.6s +tttg: c69/185 lr:0.000699 t:4.7s +tttg: c70/185 lr:0.000691 t:4.7s +tttg: c71/185 lr:0.000683 t:4.8s +tttg: c72/185 lr:0.000675 t:4.9s +tttg: c73/185 lr:0.000667 t:4.9s +tttg: c74/185 lr:0.000659 t:5.0s +tttg: c75/185 lr:0.000651 t:5.1s +tttg: c76/185 lr:0.000643 t:5.1s +tttg: c77/185 lr:0.000635 t:5.2s +tttg: c78/185 lr:0.000627 t:5.3s +tttg: c79/185 lr:0.000618 t:5.3s +tttg: c80/185 lr:0.000610 t:5.4s +tttg: c81/185 lr:0.000602 t:5.5s +tttg: c82/185 lr:0.000593 t:5.5s +tttg: c83/185 lr:0.000585 t:5.6s +tttg: c84/185 lr:0.000577 t:5.7s +tttg: c85/185 lr:0.000568 t:5.8s +tttg: c86/185 lr:0.000560 t:5.8s +tttg: c87/185 lr:0.000551 t:5.9s +tttg: c88/185 lr:0.000543 t:6.0s +tttg: c89/185 lr:0.000534 t:6.0s +tttg: c90/185 lr:0.000526 t:6.1s +tttg: c91/185 lr:0.000517 t:6.2s +tttg: c92/185 lr:0.000509 t:6.2s +tttg: c93/185 lr:0.000500 t:6.3s +tttg: c94/185 lr:0.000491 t:6.4s +tttg: c95/185 lr:0.000483 t:6.4s +tttg: c96/185 lr:0.000474 t:6.5s +tttg: c97/185 lr:0.000466 t:6.6s +tttg: c98/185 lr:0.000457 t:6.6s +tttg: c99/185 lr:0.000449 t:6.7s +tttg: c100/185 lr:0.000440 t:6.8s +tttg: c101/185 lr:0.000432 t:6.8s +tttg: c102/185 lr:0.000423 t:6.9s +tttg: c103/185 lr:0.000415 t:7.0s +tttg: c104/185 lr:0.000407 t:7.0s +tttg: c105/185 lr:0.000398 t:7.1s +tttg: c106/185 lr:0.000390 t:7.2s +tttg: c107/185 lr:0.000382 t:7.2s +tttg: c108/185 lr:0.000373 t:7.3s +tttg: c109/185 lr:0.000365 t:7.4s +tttg: c110/185 lr:0.000357 t:7.4s +tttg: c111/185 lr:0.000349 t:7.5s +tttg: c112/185 lr:0.000341 t:7.6s +tttg: c113/185 lr:0.000333 t:7.6s +tttg: c114/185 lr:0.000325 t:7.7s +tttg: c115/185 lr:0.000317 t:7.8s +tttg: c116/185 lr:0.000309 t:7.8s +tttg: c117/185 lr:0.000301 t:7.9s +tttg: c118/185 lr:0.000293 t:8.0s +tttg: c119/185 lr:0.000285 t:8.0s +tttg: c120/185 lr:0.000278 t:8.1s +tttg: c121/185 lr:0.000270 t:8.2s +tttg: c122/185 lr:0.000262 t:8.2s +tttg: c123/185 lr:0.000255 t:8.3s +tttg: c124/185 lr:0.000248 t:8.4s +tttg: c125/185 lr:0.000240 t:8.5s +tttg: c126/185 lr:0.000233 t:8.5s +tttg: c127/185 lr:0.000226 t:8.6s +tttg: c128/185 lr:0.000219 t:8.7s +tttg: c129/185 lr:0.000212 t:8.7s +tttg: c130/185 lr:0.000205 t:8.8s +tttg: c131/185 lr:0.000198 t:8.9s +tttg: c132/185 lr:0.000191 t:8.9s +tttg: c133/185 lr:0.000184 t:9.0s +tttg: c134/185 lr:0.000178 t:9.1s +tttg: c135/185 lr:0.000171 t:9.1s +tttg: c136/185 lr:0.000165 t:9.2s +tttg: c137/185 lr:0.000159 t:9.3s +tttg: c138/185 lr:0.000153 t:9.3s +tttg: c139/185 lr:0.000146 t:9.4s +tttg: c140/185 lr:0.000140 t:9.5s +tttg: c141/185 lr:0.000135 t:9.5s +tttg: c142/185 lr:0.000129 t:9.6s +tttg: c143/185 lr:0.000123 t:9.7s +tttg: c144/185 lr:0.000118 t:9.7s +tttg: c145/185 lr:0.000112 t:9.8s +tttg: c146/185 lr:0.000107 t:9.9s +tttg: c147/185 lr:0.000102 t:9.9s +tttg: c148/185 lr:0.000096 t:10.0s +tttg: c149/185 lr:0.000092 t:10.1s +tttg: c150/185 lr:0.000087 t:10.2s +tttg: c151/185 lr:0.000082 t:10.2s +tttg: c152/185 lr:0.000077 t:10.3s +tttg: c153/185 lr:0.000073 t:10.4s +tttg: c154/185 lr:0.000068 t:10.4s +tttg: c155/185 lr:0.000064 t:10.5s +tttg: c156/185 lr:0.000060 t:10.6s +tttg: c157/185 lr:0.000056 t:10.6s +tttg: c158/185 lr:0.000052 t:10.7s +tttg: c159/185 lr:0.000048 t:10.8s +tttg: c160/185 lr:0.000045 t:10.8s +tttg: c161/185 lr:0.000041 t:10.9s +tttg: c162/185 lr:0.000038 t:11.0s +tttg: c163/185 lr:0.000035 t:11.0s +tttg: c164/185 lr:0.000032 t:11.1s +tttg: c165/185 lr:0.000029 t:11.2s +tttg: c166/185 lr:0.000026 t:11.2s +tttg: c167/185 lr:0.000023 t:11.3s +tttg: c168/185 lr:0.000021 t:11.4s +tttg: c169/185 lr:0.000019 t:11.4s +tttg: c170/185 lr:0.000016 t:11.5s +tttg: c171/185 lr:0.000014 t:11.6s +tttg: c172/185 lr:0.000012 t:11.6s +tttg: c173/185 lr:0.000010 t:11.7s +tttg: c174/185 lr:0.000009 t:11.8s +tttg: c175/185 lr:0.000007 t:11.8s +tttg: c176/185 lr:0.000006 t:11.9s +tttg: c177/185 lr:0.000005 t:12.0s +tttg: c178/185 lr:0.000004 t:12.1s +tttg: c179/185 lr:0.000003 t:12.1s +tttg: c180/185 lr:0.000002 t:12.2s +tttg: c181/185 lr:0.000001 t:12.3s +tttg: c182/185 lr:0.000001 t:12.3s +tttg: c183/185 lr:0.000000 t:12.4s +tttg: c184/185 lr:0.000000 t:12.5s +ttpr: phase:2/3 t:363.9s +ttp: b746/782 bl:2.4164 bb:1.0647 rl:2.2955 rb:1.0614 dl:2884-2943 gd:0 +ttp: b745/782 bl:2.2430 bb:1.0269 rl:2.2901 rb:1.0579 dl:2842-2883 gd:0 +ttpp: phase:3/3 pd:2448 gd:2000 t:380.9s +tttg: c1/250 lr:0.001000 t:0.1s +tttg: c2/250 lr:0.001000 t:0.1s +tttg: c3/250 lr:0.001000 t:0.2s +tttg: c4/250 lr:0.001000 t:0.3s +tttg: c5/250 lr:0.000999 t:0.3s +tttg: c6/250 lr:0.000999 t:0.4s +tttg: c7/250 lr:0.000999 t:0.5s +tttg: c8/250 lr:0.000998 t:0.5s +tttg: c9/250 lr:0.000997 t:0.6s +tttg: c10/250 lr:0.000997 t:0.7s +tttg: c11/250 lr:0.000996 t:0.7s +tttg: c12/250 lr:0.000995 t:0.8s +tttg: c13/250 lr:0.000994 t:0.9s +tttg: c14/250 lr:0.000993 t:0.9s +tttg: c15/250 lr:0.000992 t:1.0s +tttg: c16/250 lr:0.000991 t:1.1s +tttg: c17/250 lr:0.000990 t:1.2s +tttg: c18/250 lr:0.000989 t:1.2s +tttg: c19/250 lr:0.000987 t:1.3s +tttg: c20/250 lr:0.000986 t:1.4s +tttg: c21/250 lr:0.000984 t:1.4s +tttg: c22/250 lr:0.000983 t:1.5s +tttg: c23/250 lr:0.000981 t:1.6s +tttg: c24/250 lr:0.000979 t:1.6s +tttg: c25/250 lr:0.000977 t:1.7s +tttg: c26/250 lr:0.000975 t:1.8s +tttg: c27/250 lr:0.000973 t:1.8s +tttg: c28/250 lr:0.000971 t:1.9s +tttg: c29/250 lr:0.000969 t:2.0s +tttg: c30/250 lr:0.000967 t:2.0s +tttg: c31/250 lr:0.000965 t:2.1s +tttg: c32/250 lr:0.000962 t:2.2s +tttg: c33/250 lr:0.000960 t:2.2s +tttg: c34/250 lr:0.000957 t:2.3s +tttg: c35/250 lr:0.000955 t:2.4s +tttg: c36/250 lr:0.000952 t:2.4s +tttg: c37/250 lr:0.000949 t:2.5s +tttg: c38/250 lr:0.000947 t:2.6s +tttg: c39/250 lr:0.000944 t:2.6s +tttg: c40/250 lr:0.000941 t:2.7s +tttg: c41/250 lr:0.000938 t:2.8s +tttg: c42/250 lr:0.000935 t:2.8s +tttg: c43/250 lr:0.000931 t:2.9s +tttg: c44/250 lr:0.000928 t:3.0s +tttg: c45/250 lr:0.000925 t:3.0s +tttg: c46/250 lr:0.000922 t:3.1s +tttg: c47/250 lr:0.000918 t:3.2s +tttg: c48/250 lr:0.000915 t:3.2s +tttg: c49/250 lr:0.000911 t:3.3s +tttg: c50/250 lr:0.000907 t:3.4s +tttg: c51/250 lr:0.000904 t:3.4s +tttg: c52/250 lr:0.000900 t:3.5s +tttg: c53/250 lr:0.000896 t:3.6s +tttg: c54/250 lr:0.000892 t:3.6s +tttg: c55/250 lr:0.000888 t:3.7s +tttg: c56/250 lr:0.000884 t:3.8s +tttg: c57/250 lr:0.000880 t:3.8s +tttg: c58/250 lr:0.000876 t:3.9s +tttg: c59/250 lr:0.000872 t:4.0s +tttg: c60/250 lr:0.000868 t:4.0s +tttg: c61/250 lr:0.000863 t:4.1s +tttg: c62/250 lr:0.000859 t:4.2s +tttg: c63/250 lr:0.000855 t:4.3s +tttg: c64/250 lr:0.000850 t:4.3s +tttg: c65/250 lr:0.000846 t:4.4s +tttg: c66/250 lr:0.000841 t:4.5s +tttg: c67/250 lr:0.000836 t:4.5s +tttg: c68/250 lr:0.000832 t:4.6s +tttg: c69/250 lr:0.000827 t:4.7s +tttg: c70/250 lr:0.000822 t:4.7s +tttg: c71/250 lr:0.000817 t:4.8s +tttg: c72/250 lr:0.000812 t:4.9s +tttg: c73/250 lr:0.000807 t:4.9s +tttg: c74/250 lr:0.000803 t:5.0s +tttg: c75/250 lr:0.000797 t:5.1s +tttg: c76/250 lr:0.000792 t:5.1s +tttg: c77/250 lr:0.000787 t:5.2s +tttg: c78/250 lr:0.000782 t:5.3s +tttg: c79/250 lr:0.000777 t:5.3s +tttg: c80/250 lr:0.000772 t:5.4s +tttg: c81/250 lr:0.000766 t:5.5s +tttg: c82/250 lr:0.000761 t:5.5s +tttg: c83/250 lr:0.000755 t:5.6s +tttg: c84/250 lr:0.000750 t:5.7s +tttg: c85/250 lr:0.000745 t:5.7s +tttg: c86/250 lr:0.000739 t:5.8s +tttg: c87/250 lr:0.000733 t:5.9s +tttg: c88/250 lr:0.000728 t:5.9s +tttg: c89/250 lr:0.000722 t:6.0s +tttg: c90/250 lr:0.000717 t:6.1s +tttg: c91/250 lr:0.000711 t:6.1s +tttg: c92/250 lr:0.000705 t:6.2s +tttg: c93/250 lr:0.000699 t:6.3s +tttg: c94/250 lr:0.000694 t:6.4s +tttg: c95/250 lr:0.000688 t:6.4s +tttg: c96/250 lr:0.000682 t:6.5s +tttg: c97/250 lr:0.000676 t:6.6s +tttg: c98/250 lr:0.000670 t:6.6s +tttg: c99/250 lr:0.000664 t:6.7s +tttg: c100/250 lr:0.000658 t:6.8s +tttg: c101/250 lr:0.000652 t:6.8s +tttg: c102/250 lr:0.000646 t:6.9s +tttg: c103/250 lr:0.000640 t:7.0s +tttg: c104/250 lr:0.000634 t:7.0s +tttg: c105/250 lr:0.000628 t:7.1s +tttg: c106/250 lr:0.000622 t:7.2s +tttg: c107/250 lr:0.000616 t:7.2s +tttg: c108/250 lr:0.000610 t:7.3s +tttg: c109/250 lr:0.000603 t:7.4s +tttg: c110/250 lr:0.000597 t:7.4s +tttg: c111/250 lr:0.000591 t:7.5s +tttg: c112/250 lr:0.000585 t:7.6s +tttg: c113/250 lr:0.000579 t:7.6s +tttg: c114/250 lr:0.000572 t:7.7s +tttg: c115/250 lr:0.000566 t:7.8s +tttg: c116/250 lr:0.000560 t:7.8s +tttg: c117/250 lr:0.000554 t:7.9s +tttg: c118/250 lr:0.000547 t:8.0s +tttg: c119/250 lr:0.000541 t:8.0s +tttg: c120/250 lr:0.000535 t:8.1s +tttg: c121/250 lr:0.000528 t:8.2s +tttg: c122/250 lr:0.000522 t:8.2s +tttg: c123/250 lr:0.000516 t:8.3s +tttg: c124/250 lr:0.000509 t:8.4s +tttg: c125/250 lr:0.000503 t:8.4s +tttg: c126/250 lr:0.000497 t:8.5s +tttg: c127/250 lr:0.000491 t:8.6s +tttg: c128/250 lr:0.000484 t:8.7s +tttg: c129/250 lr:0.000478 t:8.7s +tttg: c130/250 lr:0.000472 t:8.8s +tttg: c131/250 lr:0.000465 t:8.9s +tttg: c132/250 lr:0.000459 t:8.9s +tttg: c133/250 lr:0.000453 t:9.0s +tttg: c134/250 lr:0.000446 t:9.1s +tttg: c135/250 lr:0.000440 t:9.1s +tttg: c136/250 lr:0.000434 t:9.2s +tttg: c137/250 lr:0.000428 t:9.3s +tttg: c138/250 lr:0.000421 t:9.3s +tttg: c139/250 lr:0.000415 t:9.4s +tttg: c140/250 lr:0.000409 t:9.5s +tttg: c141/250 lr:0.000403 t:9.5s +tttg: c142/250 lr:0.000397 t:9.6s +tttg: c143/250 lr:0.000390 t:9.7s +tttg: c144/250 lr:0.000384 t:9.7s +tttg: c145/250 lr:0.000378 t:9.8s +tttg: c146/250 lr:0.000372 t:9.9s +tttg: c147/250 lr:0.000366 t:9.9s +tttg: c148/250 lr:0.000360 t:10.0s +tttg: c149/250 lr:0.000354 t:10.1s +tttg: c150/250 lr:0.000348 t:10.1s +tttg: c151/250 lr:0.000342 t:10.2s +tttg: c152/250 lr:0.000336 t:10.3s +tttg: c153/250 lr:0.000330 t:10.3s +tttg: c154/250 lr:0.000324 t:10.4s +tttg: c155/250 lr:0.000318 t:10.5s +tttg: c156/250 lr:0.000312 t:10.5s +tttg: c157/250 lr:0.000306 t:10.6s +tttg: c158/250 lr:0.000301 t:10.7s +tttg: c159/250 lr:0.000295 t:10.7s +tttg: c160/250 lr:0.000289 t:10.8s +tttg: c161/250 lr:0.000283 t:10.9s +tttg: c162/250 lr:0.000278 t:10.9s +tttg: c163/250 lr:0.000272 t:11.0s +tttg: c164/250 lr:0.000267 t:11.1s +tttg: c165/250 lr:0.000261 t:11.2s +tttg: c166/250 lr:0.000255 t:11.2s +tttg: c167/250 lr:0.000250 t:11.3s +tttg: c168/250 lr:0.000245 t:11.4s +tttg: c169/250 lr:0.000239 t:11.4s +tttg: c170/250 lr:0.000234 t:11.5s +tttg: c171/250 lr:0.000228 t:11.6s +tttg: c172/250 lr:0.000223 t:11.6s +tttg: c173/250 lr:0.000218 t:11.7s +tttg: c174/250 lr:0.000213 t:11.8s +tttg: c175/250 lr:0.000208 t:11.8s +tttg: c176/250 lr:0.000203 t:11.9s +tttg: c177/250 lr:0.000197 t:12.0s +tttg: c178/250 lr:0.000193 t:12.0s +tttg: c179/250 lr:0.000188 t:12.1s +tttg: c180/250 lr:0.000183 t:12.2s +tttg: c181/250 lr:0.000178 t:12.2s +tttg: c182/250 lr:0.000173 t:12.3s +tttg: c183/250 lr:0.000168 t:12.4s +tttg: c184/250 lr:0.000164 t:12.4s +tttg: c185/250 lr:0.000159 t:12.5s +tttg: c186/250 lr:0.000154 t:12.6s +tttg: c187/250 lr:0.000150 t:12.6s +tttg: c188/250 lr:0.000145 t:12.7s +tttg: c189/250 lr:0.000141 t:12.8s +tttg: c190/250 lr:0.000137 t:12.8s +tttg: c191/250 lr:0.000132 t:12.9s +tttg: c192/250 lr:0.000128 t:13.0s +tttg: c193/250 lr:0.000124 t:13.0s +tttg: c194/250 lr:0.000120 t:13.1s +tttg: c195/250 lr:0.000116 t:13.2s +tttg: c196/250 lr:0.000112 t:13.3s +tttg: c197/250 lr:0.000108 t:13.3s +tttg: c198/250 lr:0.000104 t:13.4s +tttg: c199/250 lr:0.000100 t:13.5s +tttg: c200/250 lr:0.000096 t:13.5s +tttg: c201/250 lr:0.000093 t:13.6s +tttg: c202/250 lr:0.000089 t:13.7s +tttg: c203/250 lr:0.000085 t:13.7s +tttg: c204/250 lr:0.000082 t:13.8s +tttg: c205/250 lr:0.000078 t:13.9s +tttg: c206/250 lr:0.000075 t:13.9s +tttg: c207/250 lr:0.000072 t:14.0s +tttg: c208/250 lr:0.000069 t:14.1s +tttg: c209/250 lr:0.000065 t:14.1s +tttg: c210/250 lr:0.000062 t:14.2s +tttg: c211/250 lr:0.000059 t:14.3s +tttg: c212/250 lr:0.000056 t:14.3s +tttg: c213/250 lr:0.000053 t:14.4s +tttg: c214/250 lr:0.000051 t:14.5s +tttg: c215/250 lr:0.000048 t:14.5s +tttg: c216/250 lr:0.000045 t:14.6s +tttg: c217/250 lr:0.000043 t:14.7s +tttg: c218/250 lr:0.000040 t:14.7s +tttg: c219/250 lr:0.000038 t:14.8s +tttg: c220/250 lr:0.000035 t:14.9s +tttg: c221/250 lr:0.000033 t:14.9s +tttg: c222/250 lr:0.000031 t:15.0s +tttg: c223/250 lr:0.000029 t:15.1s +tttg: c224/250 lr:0.000027 t:15.2s +tttg: c225/250 lr:0.000025 t:15.2s +tttg: c226/250 lr:0.000023 t:15.3s +tttg: c227/250 lr:0.000021 t:15.4s +tttg: c228/250 lr:0.000019 t:15.4s +tttg: c229/250 lr:0.000017 t:15.5s +tttg: c230/250 lr:0.000016 t:15.6s +tttg: c231/250 lr:0.000014 t:15.6s +tttg: c232/250 lr:0.000013 t:15.7s +tttg: c233/250 lr:0.000011 t:15.8s +tttg: c234/250 lr:0.000010 t:15.8s +tttg: c235/250 lr:0.000009 t:15.9s +tttg: c236/250 lr:0.000008 t:16.0s +tttg: c237/250 lr:0.000007 t:16.0s +tttg: c238/250 lr:0.000006 t:16.1s +tttg: c239/250 lr:0.000005 t:16.2s +tttg: c240/250 lr:0.000004 t:16.2s +tttg: c241/250 lr:0.000003 t:16.3s +tttg: c242/250 lr:0.000003 t:16.4s +tttg: c243/250 lr:0.000002 t:16.4s +tttg: c244/250 lr:0.000001 t:16.5s +tttg: c245/250 lr:0.000001 t:16.6s +tttg: c246/250 lr:0.000001 t:16.6s +tttg: c247/250 lr:0.000000 t:16.7s +tttg: c248/250 lr:0.000000 t:16.8s +tttg: c249/250 lr:0.000000 t:16.8s +ttpr: phase:3/3 t:399.6s +ttp: b736/782 bl:2.2515 bb:1.0608 rl:2.2869 rb:1.0581 dl:2526-2550 gd:1 +ttp: b735/782 bl:2.3910 bb:1.1000 rl:2.2948 rb:1.0613 dl:2495-2526 gd:1 +ttp: b725/782 bl:2.3259 bb:1.0463 rl:2.2968 rb:1.0603 dl:2232-2254 gd:1 +ttp: b715/782 bl:2.3634 bb:1.0304 rl:2.3004 rb:1.0586 dl:2036-2053 gd:1 +ttp: b709/782 bl:2.4496 bb:1.0957 rl:2.3077 rb:1.0605 dl:1937-1952 gd:1 +ttp: b698/782 bl:2.2566 bb:1.0329 rl:2.3055 rb:1.0592 dl:1803-1814 gd:1 +ttp: b693/782 bl:2.3435 bb:1.0528 rl:2.3071 rb:1.0590 dl:1746-1757 gd:1 +ttp: b683/782 bl:2.2796 bb:1.0611 rl:2.3060 rb:1.0591 dl:1646-1657 gd:1 +ttp: b675/782 bl:2.3703 bb:1.0600 rl:2.3082 rb:1.0591 dl:1578-1586 gd:1 +ttp: b664/782 bl:2.3435 bb:1.0285 rl:2.3093 rb:1.0581 dl:1493-1499 gd:1 +ttp: b657/782 bl:2.3336 bb:1.0605 rl:2.3101 rb:1.0582 dl:1445-1452 gd:1 +ttp: b649/782 bl:2.2910 bb:1.0186 rl:2.3095 rb:1.0570 dl:1392-1398 gd:1 +ttp: b640/782 bl:2.3170 bb:1.0555 rl:2.3097 rb:1.0570 dl:1337-1343 gd:1 +ttp: b639/782 bl:2.3150 bb:1.0338 rl:2.3099 rb:1.0564 dl:1331-1337 gd:1 +ttp: b629/782 bl:2.3583 bb:1.0149 rl:2.3110 rb:1.0554 dl:1276-1280 gd:1 +ttp: b621/782 bl:2.3031 bb:1.0517 rl:2.3108 rb:1.0553 dl:1231-1237 gd:1 +ttp: b612/782 bl:2.2445 bb:1.0169 rl:2.3094 rb:1.0545 dl:1186-1190 gd:1 +ttp: b605/782 bl:2.2556 bb:1.0287 rl:2.3084 rb:1.0540 dl:1154-1159 gd:1 +ttp: b593/782 bl:2.2960 bb:1.0135 rl:2.3081 rb:1.0532 dl:1103-1107 gd:1 +ttp: b584/782 bl:2.3057 bb:1.0424 rl:2.3081 rb:1.0530 dl:1064-1069 gd:1 +ttp: b576/782 bl:2.3839 bb:1.0965 rl:2.3094 rb:1.0537 dl:1033-1037 gd:1 +ttp: b568/782 bl:2.3619 bb:1.0843 rl:2.3102 rb:1.0542 dl:1004-1007 gd:1 +ttp: b566/782 bl:2.3076 bb:1.0307 rl:2.3102 rb:1.0538 dl:997-1001 gd:1 +ttp: b558/782 bl:2.3771 bb:1.0631 rl:2.3112 rb:1.0540 dl:968-972 gd:1 +ttp: b547/782 bl:2.3409 bb:1.0521 rl:2.3116 rb:1.0540 dl:934-937 gd:1 +ttp: b539/782 bl:2.3413 bb:1.0379 rl:2.3120 rb:1.0537 dl:909-912 gd:1 +ttp: b535/782 bl:2.3836 bb:1.0338 rl:2.3130 rb:1.0534 dl:896-899 gd:1 +ttp: b527/782 bl:2.3491 bb:1.0310 rl:2.3134 rb:1.0531 dl:872-875 gd:1 +ttp: b516/782 bl:2.3601 bb:1.0473 rl:2.3140 rb:1.0531 dl:841-843 gd:1 +ttp: b508/782 bl:2.3972 bb:1.0539 rl:2.3150 rb:1.0531 dl:817-820 gd:1 +ttp: b498/782 bl:2.3618 bb:1.0555 rl:2.3155 rb:1.0531 dl:791-794 gd:1 +ttp: b490/782 bl:2.3982 bb:1.0591 rl:2.3164 rb:1.0532 dl:771-773 gd:1 +ttp: b482/782 bl:2.3376 bb:1.0509 rl:2.3166 rb:1.0532 dl:752-754 gd:1 +ttp: b474/782 bl:2.3463 bb:1.0743 rl:2.3169 rb:1.0534 dl:733-735 gd:1 +ttp: b466/782 bl:2.3918 bb:1.0312 rl:2.3177 rb:1.0531 dl:714-717 gd:1 +ttp: b458/782 bl:2.2083 bb:1.0242 rl:2.3166 rb:1.0529 dl:697-700 gd:1 +ttp: b450/782 bl:2.3679 bb:1.0380 rl:2.3171 rb:1.0527 dl:680-682 gd:1 +ttp: b442/782 bl:2.2612 bb:1.0319 rl:2.3166 rb:1.0525 dl:664-666 gd:1 +ttp: b434/782 bl:2.3762 bb:1.0549 rl:2.3171 rb:1.0526 dl:647-648 gd:1 +ttp: b426/782 bl:2.2645 bb:1.0479 rl:2.3167 rb:1.0525 dl:632-634 gd:1 +ttp: b418/782 bl:2.2907 bb:1.0401 rl:2.3165 rb:1.0524 dl:617-618 gd:1 +ttp: b412/782 bl:2.3385 bb:1.0485 rl:2.3166 rb:1.0524 dl:605-607 gd:1 +ttp: b403/782 bl:2.3326 bb:1.0466 rl:2.3168 rb:1.0524 dl:588-590 gd:1 +ttp: b394/782 bl:2.2564 bb:0.9934 rl:2.3163 rb:1.0519 dl:571-573 gd:1 +ttp: b387/782 bl:2.3655 bb:1.0850 rl:2.3167 rb:1.0521 dl:559-561 gd:1 +ttp: b379/782 bl:2.4309 bb:1.0928 rl:2.3175 rb:1.0524 dl:545-547 gd:1 +ttp: b371/782 bl:2.2616 bb:1.1043 rl:2.3171 rb:1.0527 dl:532-533 gd:1 +ttp: b363/782 bl:2.3864 bb:1.0682 rl:2.3175 rb:1.0528 dl:518-521 gd:1 +ttp: b355/782 bl:2.3145 bb:1.0740 rl:2.3175 rb:1.0530 dl:504-506 gd:1 +ttp: b347/782 bl:2.3434 bb:1.1137 rl:2.3177 rb:1.0533 dl:492-494 gd:1 +ttp: b339/782 bl:2.3405 bb:1.0807 rl:2.3178 rb:1.0535 dl:480-482 gd:1 +ttp: b330/782 bl:2.2500 bb:1.0722 rl:2.3174 rb:1.0536 dl:466-468 gd:1 +ttp: b322/782 bl:2.3767 bb:1.0608 rl:2.3177 rb:1.0536 dl:455-457 gd:1 +ttp: b314/782 bl:2.2502 bb:1.0613 rl:2.3174 rb:1.0537 dl:442-444 gd:1 +ttp: b306/782 bl:2.4019 bb:1.0678 rl:2.3178 rb:1.0537 dl:430-432 gd:1 +ttp: b298/782 bl:2.4243 bb:1.1039 rl:2.3183 rb:1.0540 dl:418-420 gd:1 +ttp: b290/782 bl:2.3468 bb:1.0750 rl:2.3185 rb:1.0541 dl:406-407 gd:1 +ttp: b282/782 bl:2.3232 bb:1.0721 rl:2.3185 rb:1.0542 dl:395-396 gd:1 +ttp: b274/782 bl:2.3022 bb:1.0702 rl:2.3184 rb:1.0542 dl:384-385 gd:1 +ttp: b266/782 bl:2.3869 bb:1.1106 rl:2.3187 rb:1.0545 dl:374-375 gd:1 +ttp: b259/782 bl:2.3490 bb:1.1016 rl:2.3188 rb:1.0547 dl:365-366 gd:1 +ttp: b252/782 bl:2.3903 bb:1.0714 rl:2.3191 rb:1.0547 dl:356-357 gd:1 +ttp: b245/782 bl:2.3800 bb:1.1142 rl:2.3194 rb:1.0550 dl:347-349 gd:1 +ttp: b238/782 bl:2.3314 bb:1.1119 rl:2.3194 rb:1.0552 dl:338-340 gd:1 +ttp: b231/782 bl:2.3113 bb:1.0857 rl:2.3194 rb:1.0553 dl:330-331 gd:1 +ttp: b224/782 bl:2.3907 bb:1.0955 rl:2.3197 rb:1.0554 dl:322-323 gd:1 +ttp: b217/782 bl:2.3753 bb:1.1341 rl:2.3198 rb:1.0557 dl:314-315 gd:1 +ttp: b210/782 bl:2.2620 bb:1.0846 rl:2.3197 rb:1.0558 dl:306-307 gd:1 +ttp: b202/782 bl:2.3689 bb:1.1087 rl:2.3198 rb:1.0560 dl:298-299 gd:1 +ttp: b194/782 bl:2.4451 bb:1.1202 rl:2.3202 rb:1.0562 dl:289-290 gd:1 +ttp: b185/782 bl:2.4394 bb:1.1185 rl:2.3206 rb:1.0563 dl:279-280 gd:1 +ttp: b177/782 bl:2.4122 bb:1.1114 rl:2.3209 rb:1.0565 dl:271-272 gd:1 +ttp: b169/782 bl:2.3763 bb:1.1169 rl:2.3210 rb:1.0567 dl:263-264 gd:1 +ttp: b161/782 bl:2.3686 bb:1.1401 rl:2.3212 rb:1.0569 dl:256-256 gd:1 +ttp: b153/782 bl:2.2622 bb:1.0464 rl:2.3210 rb:1.0569 dl:248-249 gd:1 +ttp: b144/782 bl:2.3589 bb:1.1088 rl:2.3211 rb:1.0570 dl:239-240 gd:1 +ttp: b135/782 bl:2.4417 bb:1.1832 rl:2.3214 rb:1.0573 dl:231-232 gd:1 +ttp: b127/782 bl:2.4886 bb:1.1937 rl:2.3218 rb:1.0576 dl:223-224 gd:1 +ttp: b118/782 bl:2.4596 bb:1.1227 rl:2.3221 rb:1.0578 dl:215-216 gd:1 +ttp: b109/782 bl:2.5059 bb:1.1942 rl:2.3225 rb:1.0581 dl:207-208 gd:1 +ttp: b103/782 bl:2.4597 bb:1.1840 rl:2.3228 rb:1.0583 dl:202-202 gd:1 +ttp: b94/782 bl:2.5692 bb:1.2140 rl:2.3234 rb:1.0586 dl:193-194 gd:1 +ttp: b84/782 bl:2.5301 bb:1.2030 rl:2.3238 rb:1.0589 dl:184-185 gd:1 +ttp: b76/782 bl:2.4966 bb:1.1725 rl:2.3241 rb:1.0591 dl:177-178 gd:1 +ttp: b68/782 bl:2.5223 bb:1.1772 rl:2.3245 rb:1.0593 dl:170-171 gd:1 +ttp: b60/782 bl:2.4750 bb:1.1896 rl:2.3247 rb:1.0595 dl:163-164 gd:1 +ttp: b52/782 bl:2.6826 bb:1.2522 rl:2.3253 rb:1.0599 dl:155-156 gd:1 +ttp: b44/782 bl:2.5707 bb:1.1996 rl:2.3257 rb:1.0601 dl:147-148 gd:1 +ttp: b32/782 bl:2.6119 bb:1.2178 rl:2.3261 rb:1.0603 dl:135-136 gd:1 +ttp: b24/782 bl:2.4641 bb:1.1621 rl:2.3263 rb:1.0604 dl:127-128 gd:1 +ttp: b16/782 bl:2.6308 bb:1.2606 rl:2.3267 rb:1.0607 dl:117-118 gd:1 +ttp: b8/782 bl:2.8027 bb:1.3009 rl:2.3272 rb:1.0609 dl:103-105 gd:1 +quantized_ttt_phased val_loss:2.32767903 val_bpb:1.06365848 eval_time:498524ms +total_eval_time:498.5s