diff --git a/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/README.md b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/README.md new file mode 100644 index 0000000000..6e1e1e66a5 --- /dev/null +++ b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/README.md @@ -0,0 +1,149 @@ +# Record: Per-Layer Adaptive GPTQ Clip + int7 Embeddings + MATRIX_LR=0.026 + +**val_bpb: 1.07493** (3-seed mean, std 0.00078) | **2.77666 nats** | **~15.93 MB** | 8xH100 SXM, 600s | TTT (doc-independent LoRA) + +## Results (8xH100 80GB SXM, PyTorch 2.9.1+cu128, doc-TTT LoRA) + +| Seed | Steps | ms/step | Pre-Quant BPB | Post-Quant BPB | Post-TTT BPB | TTT gain | Eval time | Artifact | +|------|-------|---------|---------------|----------------|--------------|----------|-----------|----------| +| 42 | 4888 | 120.1 | 1.07298 | 1.08495 | **1.07437** | -0.01058 | 220s | 15,934,100 | +| 0 | 4905 | 119.7 | 1.07184 | 1.08567 | **1.07460** | -0.01107 | 218s | 15,937,217 | +| 1337 | 4904 | 119.7 | 1.07310 | 1.08663 | **1.07582** | -0.01081 | 213s | 15,928,721 | +| **Mean** | **4899** | **119.8** | **1.07264** | **1.08575** | **1.07493** | **-1.08e-2** | **217s** | **15,933,346** | +| **Std** | | | | | **0.00078** | | | | + +### Supplemental Diagnostics + +| Seed | Pre-Quant BPB | Post-Quant BPB | Post-TTT BPB | val_loss (nats) | Code size | Total submission | Train time | Eval time | +|------|---------------|----------------|--------------|-----------------|-----------|------------------|------------|-----------| +| 42 | 1.07298 | 1.08495 | 1.07437 | 2.77521 | 26,845 | 15,934,100 | 587s | 220s | +| 0 | 1.07184 | 1.08567 | 1.07460 | 2.77579 | 26,845 | 15,937,217 | 587s | 218s | +| 1337 | 1.07310 | 1.08663 | 1.07582 | 2.77897 | 26,845 | 15,928,721 | 587s | 213s | + +**Merged SOTA**: PR #1493 (@bigbag) at val_bpb=1.0810 (2.78932 nats). **Delta: -0.01266 nats** (clears 0.005-nat bar by 2.5x). + +## Key Innovation: Per-Layer Adaptive GPTQ Clip + +Standard GPTQ uses a single `clip_sigmas` for all weight matrices, forcing a one-size-fits-all trade-off between quantization quality and artifact size. We observe that **MLP and attention layers have fundamentally different weight distributions** and respond differently to clipping: + +- **MLP layers** benefit from tighter clipping (`MLP_CLIP_SIGMAS=12.0`) -- preserving more precision for the information-dense feedforward weights +- **Attention layers** tolerate looser clipping (`ATTN_CLIP_SIGMAS=13.0`) -- saving bytes while maintaining attention pattern quality +- **Embeddings** use int7 with tight clip (`EMBED_BITS=7, EMBED_CLIP_SIGMAS=15.0`) -- saves ~530 KB vs int8 with minimal quality loss + +```python +# Per-layer adaptive clip in GPTQ quantization +if "tok_emb" in name: + cs = h.embed_clip_sigmas # 15.0 -- tight for embeddings +elif "mlp" in name: + cs = h.mlp_clip_sigmas # 12.0 -- tight for quality +elif "attn" in name: + cs = h.attn_clip_sigmas # 13.0 -- looser for byte savings +else: + cs = h.matrix_clip_sigmas # 12.85 -- fallback + +bits = h.embed_bits if "tok_emb" in name else h.matrix_bits # int7 embed, int6 matrices +``` + +This adaptive scheme yields **better BPB than uniform clip=12.85** while using fewer bytes for attention weights. + +## Changes from Baseline (PR #1530 v2) + +| Parameter | PR #1530 v2 default | This submission | +|-----------|---------------------|-----------------| +| MLP_CLIP_SIGMAS | N/A (uniform) | **12.0** | +| ATTN_CLIP_SIGMAS | N/A (uniform) | **13.0** | +| EMBED_BITS | 8 | **7** | +| EMBED_CLIP_SIGMAS | 20.0 | **15.0** | +| MATRIX_LR | 0.025 | **0.026** | +| WARMDOWN_FRAC | 0.72 | **0.75** | +| TTT_CHUNK_SIZE | 32 | **48** | + +## Architecture + +11L x 512d x 8H / 4KV, MLP 4x (Triton fused), LeakyReLU(0.5)^2, Partial RoPE (16/64 dims), layerwise LN scale, tied embeddings, logit softcap=30.0. + +- **Triple recurrence**: layers 3-5 looped 2x (NUM_LOOPS=2), activated at 35% of training +- **Parallel residuals**: GPT-J style from layer 8+ +- **VarLen attention**: Flash Attention 3 variable-length for document boundaries +- **Skip gates**: sigmoid-gated U-Net connections + +## Training + +MuonEq-R optimizer (row-normalized, Newton-Schulz 5 steps), AdamW for embeddings/scalars. ~4900 steps in 587s on 8xH100 SXM. MATRIX_LR=0.026 (6-point sweep optimum), linear warmdown over final 75% of training. EMA decay 0.9965. + +## Quantization + +Full-Hessian GPTQ with **per-layer adaptive clip**: +- int6 for MLP matrices (clip_sigmas=12.0 -- tight) +- int6 for attention matrices (clip_sigmas=13.0 -- loose) +- int7 for token embeddings (clip_sigmas=15.0, saves ~530 KB vs int8) +- Brotli-11 compression + +## TTT (Test-Time Training) + +Doc-independent LoRA adaptation at eval time: +- LoRA rank 96 on K, MLP, and O projections +- Per-document: score chunk under `torch.no_grad()`, then adapt LoRA weights +- Adam optimizer (lr=0.0001, chunk_size=48) +- Weight decay 0.5 for regularization +- ~217s eval time (well within 600s budget) + +## Rule Compliance + +Per Issue #1017 (Track B -- legal eval-time adaptation): + +- **Condition 1 (Causality):** All attention is strictly causal. VarLen attention respects document boundaries. Each position scored from prefix tokens only. +- **Condition 2 (Normalized distribution):** Standard softmax over full vocab (8192 tokens). No n-gram cache, no logit biasing. +- **Condition 3 (Score before update):** Each document fully scored under `torch.no_grad()` BEFORE any LoRA update. Training only on already-scored tokens. Doc-independent: each document gets fresh LoRA weights. +- **Condition 4 (Single pass):** Each token scored exactly once. No rescoring, no multi-pass selection. + +Additional: +- No SLOT (standard or causal) +- No pre-quant TTT on val data +- No ETLB (eval-time logit bias) +- No n-gram cache or tilt +- All artifacts under 16,000,000 bytes on all 3 seeds (max: 15,937,217) +- Training under 600s on all 3 seeds (587s actual) +- Eval (TTT) under 600s on all 3 seeds (max: 220s) + +## Requirements + +``` +torch>=2.9.0 +flash-attn-3 +sentencepiece +brotli +triton +numpy +``` + +## Run Command (3-seed loop) + +```bash +pip install brotli sentencepiece +pip install flash_attn_3 --no-deps --find-links https://windreamer.github.io/flash-attention3-wheels/cu128_torch291/ +MATCHED_FINEWEB_REPO_ID=kevclark/parameter-golf python3 data/cached_challenge_fineweb.py --variant sp8192 + +for SEED in 42 0 1337; do + SEED=$SEED torchrun --standalone --nproc_per_node=8 train_gpt.py 2>&1 | tee train_seed${SEED}.log +done +``` + +## Lineage + +PR #1530 v2 (@samacqua) -> PR #1523 (@EthanYangTW) -> PR #1493 (@bigbag) -> PR #1394 (@clarkkev) -> PR #549 (@abaybektursun) + +## Credits + +- **@samacqua** -- VarLen attention, Triton fused MLP, doc-independent LoRA TTT, triple recurrence (PR #1530) +- **@EthanYangTW** -- Parameter banking, fused MLP TMA, triple recurrence (PR #1523) +- **@bigbag** -- Current merged SOTA (PR #1493) +- **@clarkkev** -- SP8192, GPTQ embeddings, SDClip, MuonEq-R, depth recurrence (PR #1394) +- **@abaybektursun** -- Score-first TTT framework (PR #549) + +## Included Files + +- `train_gpt.py` -- Self-contained training + GPTQ + TTT eval script +- `submission.json` -- Machine-readable results +- `train_seed42.log`, `train_seed0.log`, `train_seed1337.log` -- Full training + eval logs +- `README.md` -- This file diff --git a/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/submission.json b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/submission.json new file mode 100644 index 0000000000..fc0691a6a1 --- /dev/null +++ b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/submission.json @@ -0,0 +1,46 @@ +{ + "author": "Dex Hunter", + "github_id": "dexhunter", + "name": "Record: Per-Layer Adaptive GPTQ Clip + int7 Embeddings + MATRIX_LR=0.026", + "blurb": "Per-layer adaptive GPTQ clip sigmas (MLP=12.0 tight for quality, Attn=13.0 loose for bytes), int7 embeddings (saves ~530KB vs int8), tighter embed clip (15.0), MATRIX_LR=0.026 (6-point sweep optimum), WARMDOWN_FRAC=0.75, TTT_CHUNK_SIZE=48. Built on PR #1530 v2 (VarLen attention, Triton fused MLP, doc-TTT LoRA, triple recurrence).", + "date": "2026-04-13T00:00:00Z", + "val_loss": 2.77666, + "val_bpb": 1.07493, + "val_bpb_std": 0.00078, + "seeds": [42, 0, 1337], + "seed_results": { + "42": { + "val_loss": 2.77521, + "val_bpb": 1.07437, + "pre_quant_val_bpb": 1.07298, + "post_quant_val_bpb": 1.08495, + "artifact_bytes": 15934100, + "steps": 4888, + "train_time_seconds": 587, + "eval_time_seconds": 220 + }, + "0": { + "val_loss": 2.77579, + "val_bpb": 1.07460, + "pre_quant_val_bpb": 1.07184, + "post_quant_val_bpb": 1.08567, + "artifact_bytes": 15937217, + "steps": 4905, + "train_time_seconds": 587, + "eval_time_seconds": 218 + }, + "1337": { + "val_loss": 2.77897, + "val_bpb": 1.07582, + "pre_quant_val_bpb": 1.07310, + "post_quant_val_bpb": 1.08663, + "artifact_bytes": 15928721, + "steps": 4904, + "train_time_seconds": 587, + "eval_time_seconds": 213 + } + }, + "hardware": "8xH100 80GB SXM", + "bytes_total": 15933346, + "based_on": "PR #1530 (@samacqua)" +} diff --git a/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_gpt.py b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_gpt.py new file mode 100644 index 0000000000..84d9a24fba --- /dev/null +++ b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_gpt.py @@ -0,0 +1,2852 @@ +import base64, collections, copy, fcntl, glob, io, json, 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 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 + + +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)) + 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)) + sliding_window_enabled = bool(int(os.environ.get("SLIDING_WINDOW_ENABLED", "0"))) + 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)) + embedding_dim = int(os.environ.get("EMBEDDING_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)) + 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)) + head_lr = float(os.environ.get("HEAD_LR", 0.008)) + 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)) + muon_beta2 = float(os.environ.get("MUON_BETA2", 0.95)) + 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_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", "") + ttt_output_dir = os.environ.get("TTT_OUTPUT_DIR", "") + 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", 64)) + gptq_reserve_seconds = float(os.environ.get("GPTQ_RESERVE_SECONDS", 13.0)) + matrix_bits = int(os.environ.get("MATRIX_BITS", 6)) + embed_bits = int(os.environ.get("EMBED_BITS", 7)) + matrix_clip_sigmas = float(os.environ.get("MATRIX_CLIP_SIGMAS", 12.85)) + embed_clip_sigmas = float(os.environ.get("EMBED_CLIP_SIGMAS", 15.0)) + mlp_clip_sigmas = float(os.environ.get("MLP_CLIP_SIGMAS", 12.0)) + attn_clip_sigmas = float(os.environ.get("ATTN_CLIP_SIGMAS", 13.0)) + 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 + datasets_dir = os.path.join(data_dir, "datasets", f"fineweb10B_sp{vocab_size}") + train_files = os.path.join(datasets_dir, "fineweb_train_*.bin") + val_files = os.path.join(datasets_dir, "fineweb_val_*.bin") + tokenizer_path = os.path.join( + data_dir, "tokenizers", f"fineweb_{vocab_size}_bpe.model" + ) + artifact_dir = os.environ.get("ARTIFACT_DIR", "") + eval_only_path = os.environ.get("EVAL_ONLY_PATH", "") + 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) + + +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): + files = [Path(p) for p in sorted(glob.glob(pattern))] + 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_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): + 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 + ): + 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") + 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 + + 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 + q = F.linear(x, q_w.to(x.dtype)).reshape(bsz, seqlen, self.num_heads, self.head_dim) + k = F.linear(x, k_w.to(x.dtype)).reshape(bsz, seqlen, self.num_kv_heads, self.head_dim) + v = F.linear(x, 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) + y = y.reshape(bsz, seqlen, dim) + self._last_proj_input = y.detach() if getattr(self, "_calib", False) else None + 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): + if self.training and self.use_fused: + 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() + self._last_down_input = hidden.detach() if getattr(self, "_calib", False) else None + 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, + ): + 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 + ) + 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.tok_emb = nn.Embedding(h.vocab_size, h.embedding_dim) + if h.embedding_dim != h.model_dim: + self.embed_proj = CastedLinear(h.embedding_dim, h.model_dim, bias=False) + self.head_proj = CastedLinear(h.model_dim, h.embedding_dim, bias=False) + else: + self.embed_proj = None + self.head_proj = None + 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, + ) + 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.embedding_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 + 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:] + else: + self.encoder_indices = list(range(self.num_encoder_layers)) + self.decoder_indices = list(range(self.num_encoder_layers, h.num_layers)) + 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) + ) + 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) + 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_logits(self, input_ids, cu_seqlens=None, max_seqlen=0): + x = self.tok_emb(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + if self.embed_proj is not None: + x = self.embed_proj(x) + x0 = x + skips = [] + enc_iter = ( + self.encoder_indices + if self.looping_active + else range(self.num_encoder_layers) + ) + dec_iter = ( + self.decoder_indices + if self.looping_active + else range( + self.num_encoder_layers, + self.num_encoder_layers + self.num_decoder_layers, + ) + ) + for i in enc_iter: + q_w, k_w, v_w, out_w, up_w, down_w = self._bank_weights(i) + x = self.blocks[i](x, x0, q_w, k_w, v_w, out_w, up_w, down_w, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + skips.append(x) + psl = self.parallel_start_layer + lane0 = None + lane1 = 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 = self.blocks[i](x, x0, q_w, k_w, v_w, out_w, up_w, down_w, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + if lane0 is not None: + x = self._final_parallel_hidden(lane0, lane1) + x = self.final_norm(x) + if self.head_proj is not None: + x = self.head_proj(x) + if self.tie_embeddings: + logits_proj = F.linear(x, self.tok_emb.weight) + else: + logits_proj = self.lm_head(x) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward(self, input_ids, target_ids, cu_seqlens=None, max_seqlen=0): + logits = self.forward_logits( + input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen + ) + return F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + target_ids.reshape(-1), + reduction="mean", + ) + + def forward_ttt(self, input_ids, target_ids, lora): + x = self.tok_emb(input_ids) + x = F.rms_norm(x, (x.size(-1),)) + if self.embed_proj is not None: + x = self.embed_proj(x) + 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, + ) + ) + ) + slot = 0 + for i in enc_iter: + q_w, k_w, v_w, out_w, up_w, down_w = self._bank_weights(i) + x = self._block_with_lora(self.blocks[i], x, x0, lora, slot, q_w, k_w, v_w, out_w, up_w, down_w) + slot += 1 + skips.append(x) + psl = self.parallel_start_layer + lane0 = None + lane1 = 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 = self._block_with_lora(self.blocks[i], x, x0, lora, slot, q_w, k_w, v_w, out_w, up_w, down_w) + slot += 1 + if lane0 is not None: + x = self._final_parallel_hidden(lane0, lane1) + x = self.final_norm(x) + if self.head_proj is not None: + x = self.head_proj(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 + q = (F.linear(n, q_w.to(n.dtype)) + lora.q_loras[slot](n)).reshape( + bsz, seqlen, attn.num_heads, attn.head_dim + ) + k = F.linear(n, 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, 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) + y = y.reshape(bsz, seqlen, dim) + attn_out = F.linear(y, 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 + q = (F.linear(n, q_w.to(n.dtype)) + lora.q_loras[slot](n)).reshape( + bsz, seqlen, attn.num_heads, attn.head_dim + ) + k = F.linear(n, 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, 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) + y = y.reshape(bsz, seqlen, dim) + attn_out = F.linear(y, 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): + super().__init__() + self._bound = 1.0 / math.sqrt(in_features) + 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(): + self.A.uniform_(-self._bound, self._bound) + self.B.zero_() + + def forward(self, x): + return (x @ self.A.transpose(1, 2)) @ self.B.transpose(1, 2) + + +class BatchedTTTLoRA(nn.Module): + def __init__(self, bsz, model, rank, 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) + self.q_loras = nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, dim, rank) for _ in range(num_slots)] + ) + self.v_loras = nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, kv_dim, rank) for _ in range(num_slots)] + ) + self.k_loras = ( + nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, kv_dim, rank) for _ in range(num_slots)] + ) + if k_lora + else None + ) + self.mlp_loras = ( + nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, dim, rank) for _ in range(num_slots)] + ) + if mlp_lora + else None + ) + self.o_loras = ( + nn.ModuleList( + [BatchedLinearLoRA(bsz, dim, dim, rank) 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() + + +def classify_param(name): + if "tok_emb" in name or "lm_head" in name: + return "embed" + if ".mlp." in name: + return "mlp" + if ".attn." in name or ".proj." in name and ".mlp." not in name: + return "attn" + return "other" + + +@torch.compile +def zeropower_via_newtonschulz5(G, steps=10, eps=1e-07): + a, b, c = 3.4445, -4.775, 2.0315 + 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) + for _ in range(steps): + 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", + ).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) + 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, + ] + if base_model.lm_head is not None: + self.optimizer_head = torch.optim.Adam( + [ + { + "params": [base_model.lm_head.weight], + "lr": h.head_lr, + "base_lr": h.head_lr, + } + ], + betas=(h.beta1, h.beta2), + eps=h.adam_eps, + fused=True, + ) + self.optimizers.insert(1, self.optimizer_head) + else: + self.optimizer_head = None + self.replicated_params = list(tok_params[0]["params"]) + self.replicated_params.extend(scalar_params) + if base_model.lm_head is not None: + self.replicated_params.append(base_model.lm_head.weight) + 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() + if self.optimizer_head is not None: + self.optimizer_head.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"): + 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))) + if model.tie_embeddings: + hook_module = ( + model.head_proj if model.head_proj is not None else 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 gptq_mixed_quantize(state_dict, hessians, h): + result = {} + meta = {} + for (name, tensor) in state_dict.items(): + t = tensor.detach().cpu().contiguous() + 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 + q, s = gptq_quantize_weight( + t, hessians[name], clip_sigmas=cs, clip_range=2 ** (bits - 1) - 1 + ) + 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 + 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 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: + 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) + 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["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"][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 + + +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): + code_bytes_uncompressed, code_bytes = _compressed_code_size(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") + 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) + log("GPTQ:collecting Hessians from calibration data...") + t0 = time.perf_counter() + calib_loader = ShuffledSequenceLoader(h, device) + 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") + 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) + 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 + 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 eval_val_sliding(h, device, val_data, base_model, forward_logits_fn=None, batch_seqs=32): + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + base_model.eval() + run_forward_logits = base_model.forward_logits if forward_logits_fn is None else forward_logits_fn + seq_len = h.eval_seq_len + stride = h.eval_stride + total_tokens = val_data.val_tokens.numel() - 1 + context_size = seq_len - stride + window_starts = [ws for ws in range(0, total_tokens, stride) + if ws + context_size < total_tokens] + total_windows = len(window_starts) + my_s = (total_windows * h.rank) // h.world_size + my_e = (total_windows * (h.rank + 1)) // h.world_size + my_windows = window_starts[my_s:my_e] + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + byte_count = torch.zeros((), device=device, dtype=torch.float64) + total_batches = (len(my_windows) + batch_seqs - 1) // batch_seqs + is_master = h.rank == 0 + cu_bucket = 64 + t_sw_start = time.perf_counter() + with torch.no_grad(): + for bi in range(0, len(my_windows), batch_seqs): + batch_idx = bi // batch_seqs + if is_master and (batch_idx % 50 == 0 or batch_idx == total_batches - 1): + elapsed = time.perf_counter() - t_sw_start + rl = float(loss_sum.item() / token_count.item()) if token_count.item() > 0 else 0.0 + rb = float((rl / math.log(2.0)) * token_count.item() / byte_count.item()) if byte_count.item() > 0 else 0.0 + log(f"sliding_progress: batch {batch_idx+1}/{total_batches} " + f"tokens:{int(token_count.item())} running_loss:{rl:.4f} running_bpb:{rb:.4f} " + f"elapsed:{elapsed:.1f}s") + batch_ws = my_windows[bi:bi + batch_seqs] + x_parts = [] + y_parts = [] + cu_starts = [] + score_ranges = [] + offset = 0 + for ws in batch_ws: + end = min(ws + seq_len, total_tokens) + wlen = end - ws + chunk_cpu = val_data.val_tokens[ws:end + 1] + bos_pos = (chunk_cpu[:-1] == BOS_ID).nonzero(as_tuple=True)[0].tolist() + if not bos_pos or bos_pos[0] != 0: + bos_pos = [0] + bos_pos + cu_starts.extend(offset + pos for pos in bos_pos) + chunk = chunk_cpu.to(dtype=torch.int64, device=device) + x_parts.append(chunk[:-1]) + y_parts.append(chunk[1:]) + score_ranges.append((offset, wlen, ws)) + offset += wlen + x_cat = torch.cat(x_parts, dim=0)[None] + y_cat = torch.cat(y_parts, dim=0) + boundaries = cu_starts + [offset] + padded_len = get_next_multiple_of_n(len(boundaries), cu_bucket) + cu_seqlens = torch.full((padded_len,), offset, dtype=torch.int32, device=device) + cu_seqlens[:len(boundaries)] = torch.tensor(boundaries, dtype=torch.int32, device=device) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits = run_forward_logits(x_cat, cu_seqlens=cu_seqlens, max_seqlen=seq_len) + flat_nll = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y_cat, + reduction="none", + ) + flat_x = x_cat.reshape(-1) + for off, wlen, ws in score_ranges: + s = 0 if ws == 0 else context_size + lo = off + s + hi = off + wlen + scored_nll = flat_nll[lo:hi].to(torch.float64) + loss_sum += scored_nll.sum() + token_count += float(hi - lo) + tgt = y_cat[lo:hi] + prev = flat_x[lo:hi] + tb = val_data.base_bytes_lut[tgt].to(torch.float64) + tb += (val_data.has_leading_space_lut[tgt] & ~val_data.is_boundary_token_lut[prev]).to(torch.float64) + byte_count += tb.sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_count, op=dist.ReduceOp.SUM) + base_model.train() + return _loss_bpb(loss_sum, token_count, 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, +): + 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) + 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 eval_val_ttt_lora(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 = 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) + ) + doc_entries = [(i, docs[i]) for i in sampled_indices] + log( + f"ttt_lora:docs:{len(doc_entries)} rank:{h.ttt_lora_rank} lr:{h.ttt_lora_lr} chunk:{h.ttt_chunk_size}" + ) + if os.environ.get("TTT_DEBUG_BYPASS") and h.rank == 0: + test_doc = doc_entries[0][1] + ds, dl = test_doc + log(f"DEBUG: test doc start={ds} len={dl}") + toks = all_tokens_idx[ds : ds + dl].to(device=device, dtype=torch.int64) + x_d = toks[:-1].unsqueeze(0) + y_d = toks[1:].unsqueeze(0) + with torch.no_grad(), torch.autocast(device_type="cuda", dtype=torch.bfloat16): + logits_d = base_model.forward_logits(x_d) + ptl_d = F.cross_entropy( + logits_d.float().reshape(-1, logits_d.size(-1)), + y_d.reshape(-1), reduction="none", + ) + direct_loss = ptl_d.mean().item() + direct_bpb = direct_loss / math.log(2.0) + log(f"DEBUG: direct forward_logits loss={direct_loss:.6f} bpb={direct_bpb:.6f} ntokens={y_d.numel()}") + toks_first5 = toks[:5].tolist() + ptl_first5 = ptl_d[:5].tolist() + log(f"DEBUG: first 5 tokens={toks_first5} ptl={[f'{v:.4f}' for v in ptl_first5]}") + 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}" + if h.rank == 0: + _init_batch_counter(counter_path) + if dist.is_available() and dist.is_initialized(): + path_list = [counter_path] + dist.broadcast_object_list(path_list, src=0) + counter_path = path_list[0] + 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, + 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) + progress_f = None + if h.ttt_output_dir and h.rank == 0: + os.makedirs(h.ttt_output_dir, exist_ok=True) + progress_f = open(os.path.join(h.ttt_output_dir, "progress.jsonl"), "w") + 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, + 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) + 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, + ) + 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 = False + if eval_batch_set is not None: + should_report = batch_num in eval_batch_set + else: + # should_report = local_batch_count % 10 == 0 + should_report = 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 + if dt > 0: + b_loss = (cur_loss_val - prev_loss) / dt + b_bpb = b_loss / math.log(2.0) * (dt / (cur_bytes_val - prev_bytes)) + 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"ttt_progress: batch {batch_num}/{queue_len} batch_loss:{b_loss:.4f} " + f"batch_bpb:{b_bpb:.4f} running_loss:{r_loss:.4f} running_bpb:{r_bpb:.4f} " + f"doc_len:{min(doc_lens)}-{max(doc_lens)}" + ) + if progress_f is not None: + progress_f.write( + json.dumps({ + "batch": batch_num, "total_batches": queue_len, + "batch_loss": round(b_loss, 8), "batch_bpb": round(b_bpb, 8), + "running_loss": round(r_loss, 8), "running_bpb": round(r_bpb, 8), + "doc_len_min": min(doc_lens), "doc_len_max": max(doc_lens), + "chunk_size": chunk_size, + "elapsed_s": round(elapsed, 3), + "batch_t_s": round(elapsed, 3), + }) + "\n" + ) + progress_f.flush() + del cur_lora, cur_opt + finally: + if progress_f is not None: + progress_f.close() + 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() + 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 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())}") + 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) + optimizers.step(distributed=h.distributed) + return train_loss + + 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() + 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}") + 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} encoder:{base_model.encoder_indices} decoder:{base_model.decoder_indices}" + ) + train_loss = 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}" + ) + 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) + if h.eval_only_path: + log(f"eval_only:loading checkpoint from {h.eval_only_path}") + base_model = GPT(h).to(device).bfloat16() + restore_fp32_params(base_model) + base_model.load_state_dict(torch.load(h.eval_only_path, map_location=device)) + if h.num_loops > 0: + base_model.looping_active = True + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + compiled_forward_logits = torch.compile( + base_model.forward_logits, dynamic=False, fullgraph=True + ) + else: + 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 + ) + _skip_training = bool(h.eval_only_path) + torch._dynamo.reset() + timed_eval( + "diagnostic pre-quantization post-ema", + eval_val, + h, + device, + val_data, + compiled_model, + compiled_forward_logits, + ) + if not _skip_training: + serialize(h, base_model, Path(__file__).read_text(encoding="utf-8")) + else: + log("eval_only: skipping serialize (already have quantized model)") + if not os.path.exists(h.quantized_model_path): + log("eval_only: no quantized model found, running serialize anyway") + 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 + 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.sliding_window_enabled: + timed_eval( + "diagnostic quantized_sliding_window", + eval_val_sliding, + h, + device, + val_data, + eval_model, + forward_logits_fn=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 + 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) + + _ttt_debug_bypass = bool(os.environ.get("TTT_DEBUG_BYPASS")) + if _ttt_debug_bypass: + def _fwd_ttt_bypass(input_ids, target_ids, lora): + logits = ttt_model.forward_logits(input_ids) + dummy = lora.q_loras[0].B.sum() * 0 + logits = logits + dummy + bsz, sl, V = logits.shape + return F.cross_entropy( + logits.float().reshape(-1, V), target_ids.reshape(-1), reduction="none" + ).reshape(bsz, sl) + fwd_ttt_compiled = _fwd_ttt_bypass + log("ttt_lora:DEBUG BYPASS active - using forward_logits directly (no compile warmup)") + else: + 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, + 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_lora( + h, ttt_model, device, val_data, forward_ttt_train=fwd_ttt_compiled + ) + torch.cuda.synchronize() + ttt_eval_elapsed = time.perf_counter() - t_ttt + log( + f"quantized_ttt_lora val_loss:{ttt_val_loss:.8f} val_bpb:{ttt_val_bpb:.8f} 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( + subprocess.run( + ["nvidia-smi"], + stdout=subprocess.PIPE, + stderr=subprocess.PIPE, + text=True, + check=False, + ).stdout, + 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/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed0.log b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed0.log new file mode 100644 index 0000000000..95c35f6276 --- /dev/null +++ b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed0.log @@ -0,0 +1,261 @@ + +***************************************** +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. +***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: + attn_clip_sigmas: 13.0 + beta1: 0.9 + beta2: 0.95 + compressor: brotli + data_dir: ./data + datasets_dir: ./data/datasets/fineweb10B_sp8192 + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + embedding_dim: 512 + enable_looping_at: 0.35 + eval_only_path: + eval_seq_len: 2048 + eval_stride: 64 + gptq_calibration_batches: 64 + gptq_reserve_seconds: 13.0 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + head_lr: 0.008 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/PR1530_v2_aclip_m12a13_emb7_ec15_s0.txt + logit_softcap: 30.0 + 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: 0.0 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_beta2: 0.95 + 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 + qk_gain_init: 5.0 + quantized_model_path: final_model.int6.ptz + rank: 0 + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: PR1530_v2_aclip_m12a13_emb7_ec15_s0 + scalar_lr: 0.02 + seed: 0 + skip_gates_enabled: True + sliding_window_enabled: False + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: ./data/tokenizers/fineweb_8192_bpe.model + train_batch_tokens: 786432 + train_files: ./data/datasets/fineweb10B_sp8192/fineweb_train_*.bin + train_log_every: 500 + 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_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_output_dir: + ttt_weight_decay: 0.5 + val_batch_tokens: 524288 + val_doc_fraction: 1.0 + val_files: ./data/datasets/fineweb10B_sp8192/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 40540160 +model_params:35944602 +gptq:reserving 13s, effective=587000ms +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 +0/20000 val_loss: 9.0086 val_bpb: 3.4874 +1/20000 train_loss: 9.0089 train_time: 0.0m tok/s: 16259342 +2/20000 train_loss: 12.2648 train_time: 0.0m tok/s: 12917395 +3/20000 train_loss: 11.2034 train_time: 0.0m tok/s: 11115359 +4/20000 train_loss: 9.5517 train_time: 0.0m tok/s: 10266395 +5/20000 train_loss: 8.1529 train_time: 0.0m tok/s: 9860692 +500/20000 train_loss: 3.2562 train_time: 0.8m tok/s: 8334818 +1000/20000 train_loss: 3.0188 train_time: 1.6m tok/s: 8321557 +1500/20000 train_loss: 3.0309 train_time: 2.4m tok/s: 8314820 +2000/20000 train_loss: 2.9845 train_time: 3.2m tok/s: 8313022 +layer_loop:enabled step:2170 frac:0.350 encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +2500/20000 train_loss: 3.0675 train_time: 4.2m tok/s: 7825851 +3000/20000 train_loss: 2.9104 train_time: 5.3m tok/s: 7357724 +3500/20000 train_loss: 2.9765 train_time: 6.5m tok/s: 7018063 +4000/20000 train_loss: 2.9066 train_time: 7.7m tok/s: 6815052 +4000/20000 val_loss: 2.8832 val_bpb: 1.1161 +4500/20000 train_loss: 2.8592 train_time: 8.8m tok/s: 6664801 +4905/20000 val_loss: 2.7699 val_bpb: 1.0723 +stopping_early: wallclock_cap train_time: 587168ms step: 4905/20000 +peak memory allocated: 40019 MiB reserved: 44090 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.76876767 val_bpb:1.07184170 eval_time:5778ms +Serialized model: 135409136 bytes +Code size (uncompressed): 116730 bytes +Code size (compressed): 26845 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 12.6s +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.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, skip_gates, skip_weights +Serialized model quantized+brotli: 15910372 bytes +Total submission size quantized+brotli: 15937217 bytes +diagnostic quantized val_loss:2.80447660 val_bpb:1.08566530 eval_time:8875ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (80.2s) + +beginning TTT eval timer +ttt_lora:docs:50000 rank:96 lr:0.0001 chunk:48 +ttt_progress: batch 781/782 batch_loss:2.5689 batch_bpb:1.0605 running_loss:2.5689 running_bpb:1.0605 doc_len:14510-25988 +ttt_progress: batch 723/782 batch_loss:2.7845 batch_bpb:1.0623 running_loss:2.5884 running_bpb:1.0607 doc_len:1861-1885 +ttt_progress: batch 716/782 batch_loss:2.8115 batch_bpb:1.0375 running_loss:2.6058 running_bpb:1.0587 doc_len:1739-1754 +ttt_progress: batch 709/782 batch_loss:2.7901 batch_bpb:1.0601 running_loss:2.6185 running_bpb:1.0588 doc_len:1649-1661 +ttt_progress: batch 702/782 batch_loss:2.8010 batch_bpb:1.0655 running_loss:2.6297 running_bpb:1.0592 doc_len:1572-1581 +ttt_progress: batch 695/782 batch_loss:2.7831 batch_bpb:1.0790 running_loss:2.6382 running_bpb:1.0604 doc_len:1504-1513 +ttt_progress: batch 688/782 batch_loss:2.7505 batch_bpb:1.0493 running_loss:2.6439 running_bpb:1.0598 doc_len:1441-1450 +ttt_progress: batch 681/782 batch_loss:2.8258 batch_bpb:1.0729 running_loss:2.6523 running_bpb:1.0604 doc_len:1383-1393 +ttt_progress: batch 677/782 batch_loss:2.8684 batch_bpb:1.1119 running_loss:2.6617 running_bpb:1.0627 doc_len:1353-1360 +ttt_progress: batch 669/782 batch_loss:2.7829 batch_bpb:1.0553 running_loss:2.6665 running_bpb:1.0624 doc_len:1301-1308 +ttt_progress: batch 662/782 batch_loss:2.8101 batch_bpb:1.0723 running_loss:2.6719 running_bpb:1.0628 doc_len:1258-1263 +ttt_progress: batch 656/782 batch_loss:2.7494 batch_bpb:1.0380 running_loss:2.6746 running_bpb:1.0619 doc_len:1220-1227 +ttt_progress: batch 648/782 batch_loss:2.7512 batch_bpb:1.0429 running_loss:2.6770 running_bpb:1.0612 doc_len:1177-1182 +ttt_progress: batch 641/782 batch_loss:2.7654 batch_bpb:1.0412 running_loss:2.6797 running_bpb:1.0606 doc_len:1140-1144 +ttt_progress: batch 629/782 batch_loss:2.7298 batch_bpb:1.0459 running_loss:2.6811 running_bpb:1.0602 doc_len:1082-1086 +ttt_progress: batch 623/782 batch_loss:2.7901 batch_bpb:1.0739 running_loss:2.6841 running_bpb:1.0605 doc_len:1055-1060 +ttt_progress: batch 614/782 batch_loss:2.7896 batch_bpb:1.0670 running_loss:2.6867 running_bpb:1.0607 doc_len:1016-1020 +ttt_progress: batch 606/782 batch_loss:2.8216 batch_bpb:1.0856 running_loss:2.6899 running_bpb:1.0613 doc_len:982-986 +ttt_progress: batch 599/782 batch_loss:2.7452 batch_bpb:1.0543 running_loss:2.6911 running_bpb:1.0611 doc_len:954-958 +ttt_progress: batch 592/782 batch_loss:2.7881 batch_bpb:1.0497 running_loss:2.6932 running_bpb:1.0609 doc_len:930-933 +ttt_progress: batch 584/782 batch_loss:2.7656 batch_bpb:1.0391 running_loss:2.6947 running_bpb:1.0604 doc_len:904-907 +ttt_progress: batch 576/782 batch_loss:2.7819 batch_bpb:1.0477 running_loss:2.6964 running_bpb:1.0602 doc_len:877-880 +ttt_progress: batch 567/782 batch_loss:2.6802 batch_bpb:1.0323 running_loss:2.6961 running_bpb:1.0596 doc_len:849-852 +ttt_progress: batch 559/782 batch_loss:2.7549 batch_bpb:1.0470 running_loss:2.6971 running_bpb:1.0594 doc_len:824-827 +ttt_progress: batch 550/782 batch_loss:2.8120 batch_bpb:1.0790 running_loss:2.6990 running_bpb:1.0597 doc_len:798-801 +ttt_progress: batch 543/782 batch_loss:2.7841 batch_bpb:1.0452 running_loss:2.7004 running_bpb:1.0595 doc_len:779-782 +ttt_progress: batch 535/782 batch_loss:2.7891 batch_bpb:1.0575 running_loss:2.7018 running_bpb:1.0595 doc_len:759-762 +ttt_progress: batch 527/782 batch_loss:2.7416 batch_bpb:1.0418 running_loss:2.7024 running_bpb:1.0592 doc_len:739-742 +ttt_progress: batch 519/782 batch_loss:2.7411 batch_bpb:1.0395 running_loss:2.7029 running_bpb:1.0589 doc_len:720-723 +ttt_progress: batch 507/782 batch_loss:2.7574 batch_bpb:1.0410 running_loss:2.7036 running_bpb:1.0587 doc_len:690-693 +ttt_progress: batch 499/782 batch_loss:2.7813 batch_bpb:1.0496 running_loss:2.7047 running_bpb:1.0585 doc_len:673-675 +ttt_progress: batch 491/782 batch_loss:2.7374 batch_bpb:1.0314 running_loss:2.7051 running_bpb:1.0582 doc_len:655-657 +ttt_progress: batch 483/782 batch_loss:2.7472 batch_bpb:1.0506 running_loss:2.7056 running_bpb:1.0581 doc_len:639-641 +ttt_progress: batch 475/782 batch_loss:2.7292 batch_bpb:1.0232 running_loss:2.7058 running_bpb:1.0577 doc_len:622-623 +ttt_progress: batch 472/782 batch_loss:2.8034 batch_bpb:1.0717 running_loss:2.7069 running_bpb:1.0578 doc_len:616-618 +ttt_progress: batch 464/782 batch_loss:2.7272 batch_bpb:1.0808 running_loss:2.7072 running_bpb:1.0581 doc_len:600-602 +ttt_progress: batch 456/782 batch_loss:2.8200 batch_bpb:1.0710 running_loss:2.7083 running_bpb:1.0582 doc_len:586-587 +ttt_progress: batch 448/782 batch_loss:2.7284 batch_bpb:1.0365 running_loss:2.7085 running_bpb:1.0580 doc_len:571-573 +ttt_progress: batch 440/782 batch_loss:2.8705 batch_bpb:1.0959 running_loss:2.7101 running_bpb:1.0584 doc_len:556-559 +ttt_progress: batch 432/782 batch_loss:2.7603 batch_bpb:1.0502 running_loss:2.7106 running_bpb:1.0583 doc_len:542-544 +ttt_progress: batch 424/782 batch_loss:2.8038 batch_bpb:1.0836 running_loss:2.7114 running_bpb:1.0585 doc_len:528-530 +ttt_progress: batch 413/782 batch_loss:2.6465 batch_bpb:0.9979 running_loss:2.7109 running_bpb:1.0580 doc_len:510-511 +ttt_progress: batch 405/782 batch_loss:2.8317 batch_bpb:1.0703 running_loss:2.7119 running_bpb:1.0581 doc_len:497-498 +ttt_progress: batch 397/782 batch_loss:2.8915 batch_bpb:1.0985 running_loss:2.7134 running_bpb:1.0584 doc_len:484-486 +ttt_progress: batch 389/782 batch_loss:2.7951 batch_bpb:1.0646 running_loss:2.7140 running_bpb:1.0585 doc_len:471-473 +ttt_progress: batch 381/782 batch_loss:2.9088 batch_bpb:1.0923 running_loss:2.7155 running_bpb:1.0587 doc_len:460-461 +ttt_progress: batch 373/782 batch_loss:2.7692 batch_bpb:1.0809 running_loss:2.7159 running_bpb:1.0589 doc_len:449-450 +ttt_progress: batch 365/782 batch_loss:2.7857 batch_bpb:1.0857 running_loss:2.7164 running_bpb:1.0591 doc_len:437-439 +ttt_progress: batch 357/782 batch_loss:2.8639 batch_bpb:1.0836 running_loss:2.7174 running_bpb:1.0593 doc_len:426-427 +ttt_progress: batch 349/782 batch_loss:2.9269 batch_bpb:1.1121 running_loss:2.7188 running_bpb:1.0596 doc_len:415-417 +ttt_progress: batch 341/782 batch_loss:2.8735 batch_bpb:1.1001 running_loss:2.7198 running_bpb:1.0599 doc_len:404-406 +ttt_progress: batch 333/782 batch_loss:2.9087 batch_bpb:1.1328 running_loss:2.7210 running_bpb:1.0604 doc_len:394-395 +ttt_progress: batch 325/782 batch_loss:2.8443 batch_bpb:1.0926 running_loss:2.7217 running_bpb:1.0605 doc_len:384-385 +ttt_progress: batch 317/782 batch_loss:2.8764 batch_bpb:1.1120 running_loss:2.7226 running_bpb:1.0609 doc_len:373-374 +ttt_progress: batch 309/782 batch_loss:2.8370 batch_bpb:1.1069 running_loss:2.7233 running_bpb:1.0611 doc_len:363-364 +ttt_progress: batch 301/782 batch_loss:2.7947 batch_bpb:1.0868 running_loss:2.7237 running_bpb:1.0613 doc_len:353-354 +ttt_progress: batch 294/782 batch_loss:2.8484 batch_bpb:1.1015 running_loss:2.7243 running_bpb:1.0615 doc_len:345-345 +ttt_progress: batch 285/782 batch_loss:2.8883 batch_bpb:1.1304 running_loss:2.7252 running_bpb:1.0618 doc_len:334-335 +ttt_progress: batch 277/782 batch_loss:2.8213 batch_bpb:1.1112 running_loss:2.7256 running_bpb:1.0621 doc_len:325-326 +ttt_progress: batch 269/782 batch_loss:2.9245 batch_bpb:1.1272 running_loss:2.7266 running_bpb:1.0624 doc_len:316-318 +ttt_progress: batch 261/782 batch_loss:2.8738 batch_bpb:1.1241 running_loss:2.7273 running_bpb:1.0627 doc_len:308-309 +ttt_progress: batch 253/782 batch_loss:2.7555 batch_bpb:1.0821 running_loss:2.7274 running_bpb:1.0628 doc_len:298-299 +ttt_progress: batch 244/782 batch_loss:2.9564 batch_bpb:1.1595 running_loss:2.7284 running_bpb:1.0632 doc_len:289-290 +ttt_progress: batch 235/782 batch_loss:2.9334 batch_bpb:1.1150 running_loss:2.7293 running_bpb:1.0634 doc_len:280-281 +ttt_progress: batch 227/782 batch_loss:2.8132 batch_bpb:1.0910 running_loss:2.7296 running_bpb:1.0635 doc_len:272-273 +ttt_progress: batch 219/782 batch_loss:2.9076 batch_bpb:1.1345 running_loss:2.7303 running_bpb:1.0638 doc_len:264-265 +ttt_progress: batch 211/782 batch_loss:2.8975 batch_bpb:1.1543 running_loss:2.7309 running_bpb:1.0641 doc_len:256-257 +ttt_progress: batch 203/782 batch_loss:2.7762 batch_bpb:1.0906 running_loss:2.7311 running_bpb:1.0642 doc_len:249-250 +ttt_progress: batch 195/782 batch_loss:2.8503 batch_bpb:1.1156 running_loss:2.7315 running_bpb:1.0644 doc_len:242-243 +ttt_progress: batch 187/782 batch_loss:2.9006 batch_bpb:1.1184 running_loss:2.7321 running_bpb:1.0646 doc_len:235-236 +ttt_progress: batch 179/782 batch_loss:2.9521 batch_bpb:1.1531 running_loss:2.7328 running_bpb:1.0649 doc_len:228-229 +ttt_progress: batch 171/782 batch_loss:2.8958 batch_bpb:1.1138 running_loss:2.7333 running_bpb:1.0650 doc_len:221-222 +ttt_progress: batch 163/782 batch_loss:2.8912 batch_bpb:1.1348 running_loss:2.7338 running_bpb:1.0652 doc_len:214-215 +ttt_progress: batch 155/782 batch_loss:2.8847 batch_bpb:1.1338 running_loss:2.7343 running_bpb:1.0654 doc_len:207-208 +ttt_progress: batch 147/782 batch_loss:2.9348 batch_bpb:1.1611 running_loss:2.7349 running_bpb:1.0657 doc_len:201-202 +ttt_progress: batch 138/782 batch_loss:2.9256 batch_bpb:1.1645 running_loss:2.7354 running_bpb:1.0660 doc_len:194-195 +ttt_progress: batch 132/782 batch_loss:2.9549 batch_bpb:1.1372 running_loss:2.7360 running_bpb:1.0662 doc_len:189-189 +ttt_progress: batch 123/782 batch_loss:2.9565 batch_bpb:1.1812 running_loss:2.7366 running_bpb:1.0665 doc_len:182-183 +ttt_progress: batch 115/782 batch_loss:2.8777 batch_bpb:1.1612 running_loss:2.7369 running_bpb:1.0667 doc_len:176-177 +ttt_progress: batch 107/782 batch_loss:2.9224 batch_bpb:1.1472 running_loss:2.7374 running_bpb:1.0669 doc_len:171-171 +ttt_progress: batch 97/782 batch_loss:3.0106 batch_bpb:1.1761 running_loss:2.7380 running_bpb:1.0671 doc_len:163-164 +ttt_progress: batch 89/782 batch_loss:3.0199 batch_bpb:1.2044 running_loss:2.7386 running_bpb:1.0674 doc_len:157-158 +ttt_progress: batch 80/782 batch_loss:2.9225 batch_bpb:1.1971 running_loss:2.7390 running_bpb:1.0677 doc_len:150-151 +ttt_progress: batch 73/782 batch_loss:3.0565 batch_bpb:1.2091 running_loss:2.7396 running_bpb:1.0680 doc_len:144-145 +ttt_progress: batch 68/782 batch_loss:3.1022 batch_bpb:1.2052 running_loss:2.7404 running_bpb:1.0683 doc_len:141-142 +ttt_progress: batch 60/782 batch_loss:3.0757 batch_bpb:1.2343 running_loss:2.7410 running_bpb:1.0686 doc_len:134-135 +ttt_progress: batch 52/782 batch_loss:3.0650 batch_bpb:1.1993 running_loss:2.7416 running_bpb:1.0688 doc_len:128-129 +ttt_progress: batch 44/782 batch_loss:3.1443 batch_bpb:1.2252 running_loss:2.7422 running_bpb:1.0691 doc_len:122-122 +ttt_progress: batch 34/782 batch_loss:3.0810 batch_bpb:1.2473 running_loss:2.7428 running_bpb:1.0693 doc_len:114-115 +ttt_progress: batch 25/782 batch_loss:3.3027 batch_bpb:1.3089 running_loss:2.7436 running_bpb:1.0697 doc_len:106-107 +ttt_progress: batch 18/782 batch_loss:3.1507 batch_bpb:1.2763 running_loss:2.7442 running_bpb:1.0699 doc_len:99-100 +ttt_progress: batch 10/782 batch_loss:3.1420 batch_bpb:1.2416 running_loss:2.7446 running_bpb:1.0702 doc_len:89-90 +ttt_progress: batch 1/782 batch_loss:3.3569 batch_bpb:1.2458 running_loss:2.7452 running_bpb:1.0703 doc_len:45-70 +quantized_ttt_lora val_loss:2.77579059 val_bpb:1.07459504 eval_time:218228ms +total_eval_time:218.2s diff --git a/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed1337.log b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed1337.log new file mode 100644 index 0000000000..d921489d2f --- /dev/null +++ b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed1337.log @@ -0,0 +1,251 @@ + +***************************************** +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. +***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: + attn_clip_sigmas: 13.0 + beta1: 0.9 + beta2: 0.95 + compressor: brotli + data_dir: ./data + datasets_dir: ./data/datasets/fineweb10B_sp8192 + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + embedding_dim: 512 + enable_looping_at: 0.35 + eval_only_path: + eval_seq_len: 2048 + eval_stride: 64 + gptq_calibration_batches: 64 + gptq_reserve_seconds: 13.0 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + head_lr: 0.008 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/PR1530_v2_aclip_m12a13_emb7_ec15_s1337.txt + logit_softcap: 30.0 + 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: 0.0 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_beta2: 0.95 + 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 + qk_gain_init: 5.0 + quantized_model_path: final_model.int6.ptz + rank: 0 + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: PR1530_v2_aclip_m12a13_emb7_ec15_s1337 + scalar_lr: 0.02 + seed: 1337 + skip_gates_enabled: True + sliding_window_enabled: False + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: ./data/tokenizers/fineweb_8192_bpe.model + train_batch_tokens: 786432 + train_files: ./data/datasets/fineweb10B_sp8192/fineweb_train_*.bin + train_log_every: 500 + 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_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_output_dir: + ttt_weight_decay: 0.5 + val_batch_tokens: 524288 + val_doc_fraction: 1.0 + val_files: ./data/datasets/fineweb10B_sp8192/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 40540160 +model_params:35944602 +gptq:reserving 13s, effective=587000ms +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 +0/20000 val_loss: 9.0095 val_bpb: 3.4877 +1/20000 train_loss: 9.0094 train_time: 0.0m tok/s: 16543558 +2/20000 train_loss: 12.2047 train_time: 0.0m tok/s: 12946538 +3/20000 train_loss: 11.2056 train_time: 0.0m tok/s: 11092672 +4/20000 train_loss: 9.5535 train_time: 0.0m tok/s: 10322596 +5/20000 train_loss: 8.1632 train_time: 0.0m tok/s: 9916221 +500/20000 train_loss: 3.2689 train_time: 0.8m tok/s: 8338406 +1000/20000 train_loss: 3.0317 train_time: 1.6m tok/s: 8321498 +1500/20000 train_loss: 3.0332 train_time: 2.4m tok/s: 8312589 +2000/20000 train_loss: 2.9901 train_time: 3.2m tok/s: 8309814 +layer_loop:enabled step:2169 frac:0.350 encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +2500/20000 train_loss: 3.0730 train_time: 4.2m tok/s: 7823275 +3000/20000 train_loss: 2.9123 train_time: 5.3m tok/s: 7355602 +3500/20000 train_loss: 2.9792 train_time: 6.5m tok/s: 7017844 +4000/20000 train_loss: 2.9086 train_time: 7.7m tok/s: 6814690 +4000/20000 val_loss: 2.8860 val_bpb: 1.1172 +4500/20000 train_loss: 2.8623 train_time: 8.9m tok/s: 6664656 +4904/20000 val_loss: 2.7731 val_bpb: 1.0735 +stopping_early: wallclock_cap train_time: 587058ms step: 4904/20000 +peak memory allocated: 40019 MiB reserved: 44090 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.77202502 val_bpb:1.07310269 eval_time:5770ms +Serialized model: 135409136 bytes +Code size (uncompressed): 116730 bytes +Code size (compressed): 26845 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 12.6s +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.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, skip_gates, skip_weights +Serialized model quantized+brotli: 15901876 bytes +Total submission size quantized+brotli: 15928721 bytes +diagnostic quantized val_loss:2.80698023 val_bpb:1.08663450 eval_time:8982ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (85.0s) + +beginning TTT eval timer +ttt_lora:docs:50000 rank:96 lr:0.0001 chunk:48 +ttt_progress: batch 782/782 batch_loss:2.5706 batch_bpb:1.0378 running_loss:2.5706 running_bpb:1.0378 doc_len:26524-79464 +ttt_progress: batch 741/782 batch_loss:2.8119 batch_bpb:1.1069 running_loss:2.6136 running_bpb:1.0504 doc_len:2286-2319 +ttt_progress: batch 646/782 batch_loss:2.7797 batch_bpb:1.0763 running_loss:2.6274 running_bpb:1.0526 doc_len:1166-1171 +ttt_progress: batch 638/782 batch_loss:2.8480 batch_bpb:1.0505 running_loss:2.6437 running_bpb:1.0525 doc_len:1123-1129 +ttt_progress: batch 632/782 batch_loss:2.7436 batch_bpb:1.0305 running_loss:2.6504 running_bpb:1.0509 doc_len:1096-1101 +ttt_progress: batch 625/782 batch_loss:2.6788 batch_bpb:1.0064 running_loss:2.6522 running_bpb:1.0480 doc_len:1064-1068 +ttt_progress: batch 617/782 batch_loss:2.7464 batch_bpb:1.0395 running_loss:2.6574 running_bpb:1.0475 doc_len:1027-1031 +ttt_progress: batch 609/782 batch_loss:2.7943 batch_bpb:1.0607 running_loss:2.6645 running_bpb:1.0482 doc_len:994-999 +ttt_progress: batch 601/782 batch_loss:2.7710 batch_bpb:1.0649 running_loss:2.6695 running_bpb:1.0490 doc_len:963-966 +ttt_progress: batch 593/782 batch_loss:2.8004 batch_bpb:1.0473 running_loss:2.6753 running_bpb:1.0490 doc_len:933-937 +ttt_progress: batch 585/782 batch_loss:2.7688 batch_bpb:1.0676 running_loss:2.6791 running_bpb:1.0497 doc_len:908-911 +ttt_progress: batch 578/782 batch_loss:2.8113 batch_bpb:1.0711 running_loss:2.6842 running_bpb:1.0506 doc_len:884-887 +ttt_progress: batch 571/782 batch_loss:2.7198 batch_bpb:1.0375 running_loss:2.6854 running_bpb:1.0501 doc_len:862-865 +ttt_progress: batch 563/782 batch_loss:2.8066 batch_bpb:1.0647 running_loss:2.6895 running_bpb:1.0506 doc_len:837-840 +ttt_progress: batch 555/782 batch_loss:2.7680 batch_bpb:1.0564 running_loss:2.6920 running_bpb:1.0508 doc_len:812-815 +ttt_progress: batch 547/782 batch_loss:2.7388 batch_bpb:1.0343 running_loss:2.6934 running_bpb:1.0503 doc_len:790-793 +ttt_progress: batch 539/782 batch_loss:2.7340 batch_bpb:1.0468 running_loss:2.6946 running_bpb:1.0502 doc_len:769-771 +ttt_progress: batch 530/782 batch_loss:2.8171 batch_bpb:1.0428 running_loss:2.6979 running_bpb:1.0500 doc_len:747-750 +ttt_progress: batch 522/782 batch_loss:2.8314 batch_bpb:1.0884 running_loss:2.7012 running_bpb:1.0510 doc_len:727-730 +ttt_progress: batch 514/782 batch_loss:2.9153 batch_bpb:1.0996 running_loss:2.7064 running_bpb:1.0522 doc_len:707-710 +ttt_progress: batch 506/782 batch_loss:2.8157 batch_bpb:1.0786 running_loss:2.7089 running_bpb:1.0528 doc_len:688-690 +ttt_progress: batch 498/782 batch_loss:2.6852 batch_bpb:1.0395 running_loss:2.7084 running_bpb:1.0525 doc_len:671-673 +ttt_progress: batch 491/782 batch_loss:2.7470 batch_bpb:1.0350 running_loss:2.7092 running_bpb:1.0521 doc_len:655-657 +ttt_progress: batch 483/782 batch_loss:2.7485 batch_bpb:1.0511 running_loss:2.7100 running_bpb:1.0521 doc_len:639-641 +ttt_progress: batch 475/782 batch_loss:2.7329 batch_bpb:1.0246 running_loss:2.7104 running_bpb:1.0516 doc_len:622-623 +ttt_progress: batch 469/782 batch_loss:2.8027 batch_bpb:1.1145 running_loss:2.7121 running_bpb:1.0527 doc_len:610-611 +ttt_progress: batch 460/782 batch_loss:2.7961 batch_bpb:1.0605 running_loss:2.7136 running_bpb:1.0528 doc_len:593-595 +ttt_progress: batch 452/782 batch_loss:2.7502 batch_bpb:1.0609 running_loss:2.7142 running_bpb:1.0530 doc_len:579-580 +ttt_progress: batch 443/782 batch_loss:2.7752 batch_bpb:1.0571 running_loss:2.7152 running_bpb:1.0530 doc_len:562-564 +ttt_progress: batch 435/782 batch_loss:2.7361 batch_bpb:1.0535 running_loss:2.7155 running_bpb:1.0530 doc_len:547-549 +ttt_progress: batch 426/782 batch_loss:2.7408 batch_bpb:1.0726 running_loss:2.7159 running_bpb:1.0533 doc_len:532-533 +ttt_progress: batch 418/782 batch_loss:2.8168 batch_bpb:1.0744 running_loss:2.7173 running_bpb:1.0536 doc_len:517-519 +ttt_progress: batch 410/782 batch_loss:2.7846 batch_bpb:1.0573 running_loss:2.7182 running_bpb:1.0537 doc_len:505-507 +ttt_progress: batch 402/782 batch_loss:2.7578 batch_bpb:1.0390 running_loss:2.7187 running_bpb:1.0535 doc_len:492-493 +ttt_progress: batch 394/782 batch_loss:2.9013 batch_bpb:1.1189 running_loss:2.7211 running_bpb:1.0543 doc_len:479-481 +ttt_progress: batch 392/782 batch_loss:2.8082 batch_bpb:1.0844 running_loss:2.7221 running_bpb:1.0547 doc_len:476-478 +ttt_progress: batch 384/782 batch_loss:2.8554 batch_bpb:1.0955 running_loss:2.7237 running_bpb:1.0552 doc_len:464-466 +ttt_progress: batch 375/782 batch_loss:2.8144 batch_bpb:1.1091 running_loss:2.7248 running_bpb:1.0558 doc_len:452-453 +ttt_progress: batch 367/782 batch_loss:2.8382 batch_bpb:1.0660 running_loss:2.7260 running_bpb:1.0559 doc_len:440-441 +ttt_progress: batch 359/782 batch_loss:2.8081 batch_bpb:1.0853 running_loss:2.7269 running_bpb:1.0562 doc_len:429-430 +ttt_progress: batch 351/782 batch_loss:2.8388 batch_bpb:1.0926 running_loss:2.7280 running_bpb:1.0566 doc_len:418-419 +ttt_progress: batch 341/782 batch_loss:2.8780 batch_bpb:1.1018 running_loss:2.7295 running_bpb:1.0570 doc_len:404-406 +ttt_progress: batch 333/782 batch_loss:2.9051 batch_bpb:1.1314 running_loss:2.7312 running_bpb:1.0577 doc_len:394-395 +ttt_progress: batch 325/782 batch_loss:2.8570 batch_bpb:1.0975 running_loss:2.7323 running_bpb:1.0581 doc_len:384-385 +ttt_progress: batch 320/782 batch_loss:2.7771 batch_bpb:1.0834 running_loss:2.7327 running_bpb:1.0583 doc_len:377-378 +ttt_progress: batch 312/782 batch_loss:2.7388 batch_bpb:1.0691 running_loss:2.7328 running_bpb:1.0584 doc_len:367-368 +ttt_progress: batch 304/782 batch_loss:2.9217 batch_bpb:1.1379 running_loss:2.7343 running_bpb:1.0591 doc_len:357-358 +ttt_progress: batch 296/782 batch_loss:2.8175 batch_bpb:1.0896 running_loss:2.7350 running_bpb:1.0593 doc_len:347-348 +ttt_progress: batch 287/782 batch_loss:2.8717 batch_bpb:1.1202 running_loss:2.7360 running_bpb:1.0598 doc_len:336-337 +ttt_progress: batch 279/782 batch_loss:2.8581 batch_bpb:1.0924 running_loss:2.7370 running_bpb:1.0600 doc_len:327-329 +ttt_progress: batch 272/782 batch_loss:2.8681 batch_bpb:1.1126 running_loss:2.7379 running_bpb:1.0604 doc_len:320-321 +ttt_progress: batch 264/782 batch_loss:2.9088 batch_bpb:1.1513 running_loss:2.7391 running_bpb:1.0610 doc_len:311-312 +ttt_progress: batch 256/782 batch_loss:2.9024 batch_bpb:1.1378 running_loss:2.7402 running_bpb:1.0615 doc_len:301-302 +ttt_progress: batch 247/782 batch_loss:2.7948 batch_bpb:1.0799 running_loss:2.7405 running_bpb:1.0617 doc_len:292-293 +ttt_progress: batch 237/782 batch_loss:2.9240 batch_bpb:1.1496 running_loss:2.7417 running_bpb:1.0622 doc_len:282-283 +ttt_progress: batch 229/782 batch_loss:2.9013 batch_bpb:1.1412 running_loss:2.7426 running_bpb:1.0627 doc_len:274-275 +ttt_progress: batch 221/782 batch_loss:2.8572 batch_bpb:1.1467 running_loss:2.7433 running_bpb:1.0631 doc_len:266-267 +ttt_progress: batch 213/782 batch_loss:3.0055 batch_bpb:1.1727 running_loss:2.7447 running_bpb:1.0637 doc_len:258-259 +ttt_progress: batch 205/782 batch_loss:2.8495 batch_bpb:1.1118 running_loss:2.7453 running_bpb:1.0640 doc_len:251-252 +ttt_progress: batch 196/782 batch_loss:2.9108 batch_bpb:1.1663 running_loss:2.7462 running_bpb:1.0645 doc_len:243-244 +ttt_progress: batch 189/782 batch_loss:2.9750 batch_bpb:1.2074 running_loss:2.7473 running_bpb:1.0652 doc_len:237-237 +ttt_progress: batch 181/782 batch_loss:2.8877 batch_bpb:1.1603 running_loss:2.7480 running_bpb:1.0656 doc_len:230-230 +ttt_progress: batch 172/782 batch_loss:3.0137 batch_bpb:1.1852 running_loss:2.7492 running_bpb:1.0662 doc_len:222-223 +ttt_progress: batch 164/782 batch_loss:2.9834 batch_bpb:1.1537 running_loss:2.7503 running_bpb:1.0666 doc_len:215-216 +ttt_progress: batch 156/782 batch_loss:2.9069 batch_bpb:1.1147 running_loss:2.7510 running_bpb:1.0668 doc_len:208-209 +ttt_progress: batch 148/782 batch_loss:2.9907 batch_bpb:1.1624 running_loss:2.7520 running_bpb:1.0672 doc_len:202-203 +ttt_progress: batch 140/782 batch_loss:2.9713 batch_bpb:1.1732 running_loss:2.7529 running_bpb:1.0676 doc_len:195-196 +ttt_progress: batch 131/782 batch_loss:3.0514 batch_bpb:1.2128 running_loss:2.7540 running_bpb:1.0682 doc_len:188-189 +ttt_progress: batch 124/782 batch_loss:2.8931 batch_bpb:1.1573 running_loss:2.7545 running_bpb:1.0685 doc_len:183-184 +ttt_progress: batch 117/782 batch_loss:2.8606 batch_bpb:1.1466 running_loss:2.7549 running_bpb:1.0687 doc_len:178-178 +ttt_progress: batch 108/782 batch_loss:2.8743 batch_bpb:1.1039 running_loss:2.7553 running_bpb:1.0689 doc_len:171-172 +ttt_progress: batch 100/782 batch_loss:2.9507 batch_bpb:1.1583 running_loss:2.7560 running_bpb:1.0692 doc_len:165-166 +ttt_progress: batch 93/782 batch_loss:2.9620 batch_bpb:1.1882 running_loss:2.7567 running_bpb:1.0695 doc_len:160-160 +ttt_progress: batch 85/782 batch_loss:3.0059 batch_bpb:1.2068 running_loss:2.7574 running_bpb:1.0700 doc_len:154-154 +ttt_progress: batch 77/782 batch_loss:3.0389 batch_bpb:1.1742 running_loss:2.7583 running_bpb:1.0703 doc_len:148-148 +ttt_progress: batch 64/782 batch_loss:3.0097 batch_bpb:1.2475 running_loss:2.7590 running_bpb:1.0707 doc_len:138-139 +ttt_progress: batch 57/782 batch_loss:3.0619 batch_bpb:1.2342 running_loss:2.7598 running_bpb:1.0711 doc_len:132-133 +ttt_progress: batch 48/782 batch_loss:3.0096 batch_bpb:1.1766 running_loss:2.7604 running_bpb:1.0714 doc_len:125-126 +ttt_progress: batch 41/782 batch_loss:3.1535 batch_bpb:1.2888 running_loss:2.7613 running_bpb:1.0719 doc_len:119-120 +ttt_progress: batch 33/782 batch_loss:3.1220 batch_bpb:1.2222 running_loss:2.7621 running_bpb:1.0722 doc_len:113-114 +ttt_progress: batch 24/782 batch_loss:3.0672 batch_bpb:1.2135 running_loss:2.7627 running_bpb:1.0725 doc_len:105-106 +ttt_progress: batch 16/782 batch_loss:3.0856 batch_bpb:1.2303 running_loss:2.7634 running_bpb:1.0728 doc_len:97-98 +ttt_progress: batch 5/782 batch_loss:3.3315 batch_bpb:1.2995 running_loss:2.7643 running_bpb:1.0732 doc_len:80-82 +quantized_ttt_lora val_loss:2.77896748 val_bpb:1.07582491 eval_time:213210ms +total_eval_time:213.2s diff --git a/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed42.log b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed42.log new file mode 100644 index 0000000000..d407fdef7a --- /dev/null +++ b/records/track_10min_16mb/2026-04-13_AdaptiveClip_Emb7_MLR026/train_seed42.log @@ -0,0 +1,262 @@ + +***************************************** +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. +***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: + attn_clip_sigmas: 13.0 + beta1: 0.9 + beta2: 0.95 + compressor: brotli + data_dir: ./data + datasets_dir: ./data/datasets/fineweb10B_sp8192 + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + embedding_dim: 512 + enable_looping_at: 0.35 + eval_only_path: + eval_seq_len: 2048 + eval_stride: 64 + gptq_calibration_batches: 64 + gptq_reserve_seconds: 13.0 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + head_lr: 0.008 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/PR1530_v2_aclip_m12a13_emb7_ec15_s42.txt + logit_softcap: 30.0 + 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: 0.0 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_beta2: 0.95 + 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 + qk_gain_init: 5.0 + quantized_model_path: final_model.int6.ptz + rank: 0 + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: PR1530_v2_aclip_m12a13_emb7_ec15_s42 + scalar_lr: 0.02 + seed: 42 + skip_gates_enabled: True + sliding_window_enabled: False + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: ./data/tokenizers/fineweb_8192_bpe.model + train_batch_tokens: 786432 + train_files: ./data/datasets/fineweb10B_sp8192/fineweb_train_*.bin + train_log_every: 500 + 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_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_output_dir: + ttt_weight_decay: 0.5 + val_batch_tokens: 524288 + val_doc_fraction: 1.0 + val_files: ./data/datasets/fineweb10B_sp8192/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 40540160 +model_params:35944602 +gptq:reserving 13s, effective=587000ms +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 +0/20000 val_loss: 9.0078 val_bpb: 3.4871 +1/20000 train_loss: 9.0072 train_time: 0.0m tok/s: 16670435 +2/20000 train_loss: 12.3395 train_time: 0.0m tok/s: 12980022 +3/20000 train_loss: 11.3031 train_time: 0.0m tok/s: 11155122 +4/20000 train_loss: 9.6455 train_time: 0.0m tok/s: 10324651 +5/20000 train_loss: 8.2401 train_time: 0.0m tok/s: 9917503 +500/20000 train_loss: 3.2675 train_time: 0.8m tok/s: 8297149 +1000/20000 train_loss: 3.0266 train_time: 1.6m tok/s: 8273523 +1500/20000 train_loss: 3.0380 train_time: 2.4m tok/s: 8265451 +2000/20000 train_loss: 2.9872 train_time: 3.2m tok/s: 8265629 +layer_loop:enabled step:2160 frac:0.350 encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +2500/20000 train_loss: 3.0734 train_time: 4.2m tok/s: 7776533 +3000/20000 train_loss: 2.9096 train_time: 5.4m tok/s: 7318892 +3500/20000 train_loss: 2.9786 train_time: 6.6m tok/s: 6987006 +4000/20000 train_loss: 2.9071 train_time: 7.7m tok/s: 6787252 +4000/20000 val_loss: 2.8834 val_bpb: 1.1162 +4500/20000 train_loss: 2.8582 train_time: 8.9m tok/s: 6638550 +4888/20000 val_loss: 2.7727 val_bpb: 1.0734 +stopping_early: wallclock_cap train_time: 587073ms step: 4888/20000 +peak memory allocated: 40019 MiB reserved: 44090 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.77169854 val_bpb:1.07297630 eval_time:6374ms +Serialized model: 135409136 bytes +Code size (uncompressed): 116730 bytes +Code size (compressed): 26845 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 12.6s +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.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, skip_gates, skip_weights +Serialized model quantized+brotli: 15907255 bytes +Total submission size quantized+brotli: 15934100 bytes +diagnostic quantized val_loss:2.80263104 val_bpb:1.08495085 eval_time:10096ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (85.3s) + +beginning TTT eval timer +ttt_lora:docs:50000 rank:96 lr:0.0001 chunk:48 +ttt_progress: batch 781/782 batch_loss:2.5691 batch_bpb:1.0606 running_loss:2.5691 running_bpb:1.0606 doc_len:14510-25988 +ttt_progress: batch 724/782 batch_loss:2.7503 batch_bpb:1.0513 running_loss:2.5856 running_bpb:1.0597 doc_len:1885-1900 +ttt_progress: batch 717/782 batch_loss:2.7944 batch_bpb:1.0524 running_loss:2.6020 running_bpb:1.0590 doc_len:1754-1773 +ttt_progress: batch 710/782 batch_loss:2.7586 batch_bpb:1.0695 running_loss:2.6129 running_bpb:1.0598 doc_len:1661-1673 +ttt_progress: batch 705/782 batch_loss:2.7842 batch_bpb:1.0724 running_loss:2.6236 running_bpb:1.0606 doc_len:1606-1617 +ttt_progress: batch 700/782 batch_loss:2.6793 batch_bpb:1.0457 running_loss:2.6268 running_bpb:1.0598 doc_len:1552-1562 +ttt_progress: batch 691/782 batch_loss:2.7075 batch_bpb:1.0454 running_loss:2.6309 running_bpb:1.0590 doc_len:1467-1476 +ttt_progress: batch 685/782 batch_loss:2.7761 batch_bpb:1.0640 running_loss:2.6377 running_bpb:1.0592 doc_len:1414-1422 +ttt_progress: batch 678/782 batch_loss:2.7857 batch_bpb:1.0486 running_loss:2.6441 running_bpb:1.0587 doc_len:1361-1368 +ttt_progress: batch 671/782 batch_loss:2.8805 batch_bpb:1.1164 running_loss:2.6536 running_bpb:1.0611 doc_len:1316-1321 +ttt_progress: batch 665/782 batch_loss:2.7402 batch_bpb:1.0327 running_loss:2.6569 running_bpb:1.0600 doc_len:1275-1282 +ttt_progress: batch 658/782 batch_loss:2.8130 batch_bpb:1.0767 running_loss:2.6623 running_bpb:1.0606 doc_len:1234-1239 +ttt_progress: batch 651/782 batch_loss:2.7222 batch_bpb:1.0455 running_loss:2.6643 running_bpb:1.0601 doc_len:1193-1198 +ttt_progress: batch 644/782 batch_loss:2.7378 batch_bpb:1.0330 running_loss:2.6665 running_bpb:1.0592 doc_len:1155-1160 +ttt_progress: batch 629/782 batch_loss:2.7308 batch_bpb:1.0462 running_loss:2.6683 running_bpb:1.0589 doc_len:1082-1086 +ttt_progress: batch 621/782 batch_loss:2.8367 batch_bpb:1.0864 running_loss:2.6728 running_bpb:1.0596 doc_len:1046-1050 +ttt_progress: batch 615/782 batch_loss:2.8382 batch_bpb:1.0656 running_loss:2.6769 running_bpb:1.0598 doc_len:1020-1023 +ttt_progress: batch 607/782 batch_loss:2.6988 batch_bpb:1.0402 running_loss:2.6774 running_bpb:1.0593 doc_len:986-990 +ttt_progress: batch 599/782 batch_loss:2.7446 batch_bpb:1.0541 running_loss:2.6789 running_bpb:1.0592 doc_len:954-958 +ttt_progress: batch 592/782 batch_loss:2.7852 batch_bpb:1.0486 running_loss:2.6812 running_bpb:1.0589 doc_len:930-933 +ttt_progress: batch 585/782 batch_loss:2.7639 batch_bpb:1.0657 running_loss:2.6829 running_bpb:1.0591 doc_len:908-911 +ttt_progress: batch 579/782 batch_loss:2.6427 batch_bpb:1.0072 running_loss:2.6821 running_bpb:1.0580 doc_len:888-891 +ttt_progress: batch 571/782 batch_loss:2.7170 batch_bpb:1.0364 running_loss:2.6827 running_bpb:1.0576 doc_len:862-865 +ttt_progress: batch 563/782 batch_loss:2.8011 batch_bpb:1.0626 running_loss:2.6848 running_bpb:1.0577 doc_len:837-840 +ttt_progress: batch 555/782 batch_loss:2.7685 batch_bpb:1.0565 running_loss:2.6862 running_bpb:1.0577 doc_len:812-815 +ttt_progress: batch 547/782 batch_loss:2.7332 batch_bpb:1.0322 running_loss:2.6870 running_bpb:1.0572 doc_len:790-793 +ttt_progress: batch 538/782 batch_loss:2.6963 batch_bpb:1.0428 running_loss:2.6871 running_bpb:1.0570 doc_len:767-769 +ttt_progress: batch 531/782 batch_loss:2.7796 batch_bpb:1.0543 running_loss:2.6885 running_bpb:1.0570 doc_len:750-752 +ttt_progress: batch 523/782 batch_loss:2.8130 batch_bpb:1.0564 running_loss:2.6903 running_bpb:1.0570 doc_len:730-732 +ttt_progress: batch 515/782 batch_loss:2.7865 batch_bpb:1.0740 running_loss:2.6916 running_bpb:1.0572 doc_len:710-713 +ttt_progress: batch 507/782 batch_loss:2.7594 batch_bpb:1.0417 running_loss:2.6925 running_bpb:1.0570 doc_len:690-693 +ttt_progress: batch 499/782 batch_loss:2.7842 batch_bpb:1.0507 running_loss:2.6937 running_bpb:1.0569 doc_len:673-675 +ttt_progress: batch 487/782 batch_loss:2.8243 batch_bpb:1.0790 running_loss:2.6952 running_bpb:1.0572 doc_len:647-649 +ttt_progress: batch 479/782 batch_loss:2.7113 batch_bpb:1.0348 running_loss:2.6954 running_bpb:1.0569 doc_len:630-632 +ttt_progress: batch 471/782 batch_loss:2.8425 batch_bpb:1.0710 running_loss:2.6971 running_bpb:1.0571 doc_len:614-616 +ttt_progress: batch 463/782 batch_loss:2.7976 batch_bpb:1.0739 running_loss:2.6981 running_bpb:1.0573 doc_len:599-600 +ttt_progress: batch 455/782 batch_loss:2.7993 batch_bpb:1.0735 running_loss:2.6992 running_bpb:1.0574 doc_len:584-586 +ttt_progress: batch 447/782 batch_loss:2.8277 batch_bpb:1.0874 running_loss:2.7005 running_bpb:1.0577 doc_len:569-571 +ttt_progress: batch 440/782 batch_loss:2.8682 batch_bpb:1.0951 running_loss:2.7021 running_bpb:1.0581 doc_len:556-559 +ttt_progress: batch 432/782 batch_loss:2.7602 batch_bpb:1.0501 running_loss:2.7027 running_bpb:1.0580 doc_len:542-544 +ttt_progress: batch 424/782 batch_loss:2.8025 batch_bpb:1.0832 running_loss:2.7036 running_bpb:1.0583 doc_len:528-530 +ttt_progress: batch 415/782 batch_loss:2.8492 batch_bpb:1.0825 running_loss:2.7048 running_bpb:1.0585 doc_len:513-514 +ttt_progress: batch 407/782 batch_loss:2.7878 batch_bpb:1.0613 running_loss:2.7055 running_bpb:1.0585 doc_len:500-501 +ttt_progress: batch 399/782 batch_loss:2.7479 batch_bpb:1.0409 running_loss:2.7059 running_bpb:1.0584 doc_len:487-489 +ttt_progress: batch 391/782 batch_loss:2.8210 batch_bpb:1.0986 running_loss:2.7068 running_bpb:1.0587 doc_len:475-476 +ttt_progress: batch 383/782 batch_loss:2.8354 batch_bpb:1.0859 running_loss:2.7078 running_bpb:1.0589 doc_len:463-464 +ttt_progress: batch 375/782 batch_loss:2.8078 batch_bpb:1.1065 running_loss:2.7085 running_bpb:1.0592 doc_len:452-453 +ttt_progress: batch 367/782 batch_loss:2.8363 batch_bpb:1.0653 running_loss:2.7094 running_bpb:1.0593 doc_len:440-441 +ttt_progress: batch 359/782 batch_loss:2.7964 batch_bpb:1.0808 running_loss:2.7100 running_bpb:1.0594 doc_len:429-430 +ttt_progress: batch 352/782 batch_loss:2.7585 batch_bpb:1.0966 running_loss:2.7103 running_bpb:1.0597 doc_len:419-420 +ttt_progress: batch 344/782 batch_loss:2.8888 batch_bpb:1.1074 running_loss:2.7115 running_bpb:1.0600 doc_len:408-410 +ttt_progress: batch 335/782 batch_loss:2.7216 batch_bpb:1.0908 running_loss:2.7115 running_bpb:1.0602 doc_len:396-398 +ttt_progress: batch 327/782 batch_loss:2.7790 batch_bpb:1.0789 running_loss:2.7120 running_bpb:1.0603 doc_len:387-388 +ttt_progress: batch 320/782 batch_loss:2.7684 batch_bpb:1.0800 running_loss:2.7123 running_bpb:1.0604 doc_len:377-378 +ttt_progress: batch 311/782 batch_loss:2.8551 batch_bpb:1.0937 running_loss:2.7131 running_bpb:1.0606 doc_len:365-367 +ttt_progress: batch 303/782 batch_loss:2.8151 batch_bpb:1.0906 running_loss:2.7136 running_bpb:1.0608 doc_len:355-357 +ttt_progress: batch 295/782 batch_loss:2.8429 batch_bpb:1.1209 running_loss:2.7143 running_bpb:1.0611 doc_len:345-347 +ttt_progress: batch 287/782 batch_loss:2.8633 batch_bpb:1.1170 running_loss:2.7151 running_bpb:1.0614 doc_len:336-337 +ttt_progress: batch 281/782 batch_loss:2.9253 batch_bpb:1.1534 running_loss:2.7161 running_bpb:1.0618 doc_len:329-330 +ttt_progress: batch 273/782 batch_loss:2.7746 batch_bpb:1.0628 running_loss:2.7164 running_bpb:1.0618 doc_len:321-322 +ttt_progress: batch 265/782 batch_loss:2.8328 batch_bpb:1.0904 running_loss:2.7170 running_bpb:1.0620 doc_len:312-313 +ttt_progress: batch 257/782 batch_loss:2.9283 batch_bpb:1.1149 running_loss:2.7179 running_bpb:1.0622 doc_len:302-304 +ttt_progress: batch 249/782 batch_loss:2.8964 batch_bpb:1.1536 running_loss:2.7187 running_bpb:1.0626 doc_len:294-295 +ttt_progress: batch 241/782 batch_loss:2.9153 batch_bpb:1.1293 running_loss:2.7195 running_bpb:1.0629 doc_len:286-287 +ttt_progress: batch 234/782 batch_loss:2.9200 batch_bpb:1.1575 running_loss:2.7204 running_bpb:1.0633 doc_len:279-280 +ttt_progress: batch 226/782 batch_loss:2.9334 batch_bpb:1.1412 running_loss:2.7212 running_bpb:1.0636 doc_len:271-272 +ttt_progress: batch 217/782 batch_loss:2.8850 batch_bpb:1.1301 running_loss:2.7218 running_bpb:1.0638 doc_len:262-263 +ttt_progress: batch 209/782 batch_loss:2.9272 batch_bpb:1.1588 running_loss:2.7226 running_bpb:1.0642 doc_len:254-255 +ttt_progress: batch 201/782 batch_loss:2.8767 batch_bpb:1.1213 running_loss:2.7232 running_bpb:1.0644 doc_len:247-248 +ttt_progress: batch 193/782 batch_loss:2.8845 batch_bpb:1.1622 running_loss:2.7237 running_bpb:1.0647 doc_len:240-241 +ttt_progress: batch 187/782 batch_loss:2.9036 batch_bpb:1.1196 running_loss:2.7243 running_bpb:1.0649 doc_len:235-236 +ttt_progress: batch 181/782 batch_loss:2.8904 batch_bpb:1.1614 running_loss:2.7249 running_bpb:1.0652 doc_len:230-230 +ttt_progress: batch 171/782 batch_loss:2.8919 batch_bpb:1.1123 running_loss:2.7254 running_bpb:1.0654 doc_len:221-222 +ttt_progress: batch 162/782 batch_loss:2.9579 batch_bpb:1.1477 running_loss:2.7261 running_bpb:1.0656 doc_len:213-214 +ttt_progress: batch 153/782 batch_loss:3.0235 batch_bpb:1.1663 running_loss:2.7270 running_bpb:1.0659 doc_len:206-207 +ttt_progress: batch 146/782 batch_loss:2.8964 batch_bpb:1.1498 running_loss:2.7275 running_bpb:1.0662 doc_len:200-201 +ttt_progress: batch 139/782 batch_loss:2.9999 batch_bpb:1.1610 running_loss:2.7282 running_bpb:1.0664 doc_len:195-195 +ttt_progress: batch 130/782 batch_loss:3.1429 batch_bpb:1.2354 running_loss:2.7293 running_bpb:1.0669 doc_len:187-188 +ttt_progress: batch 122/782 batch_loss:2.8987 batch_bpb:1.1598 running_loss:2.7297 running_bpb:1.0671 doc_len:181-182 +ttt_progress: batch 113/782 batch_loss:3.0433 batch_bpb:1.1966 running_loss:2.7305 running_bpb:1.0674 doc_len:175-176 +ttt_progress: batch 107/782 batch_loss:2.9232 batch_bpb:1.1475 running_loss:2.7310 running_bpb:1.0676 doc_len:171-171 +ttt_progress: batch 97/782 batch_loss:3.0098 batch_bpb:1.1758 running_loss:2.7316 running_bpb:1.0679 doc_len:163-164 +ttt_progress: batch 89/782 batch_loss:3.0280 batch_bpb:1.2076 running_loss:2.7323 running_bpb:1.0682 doc_len:157-158 +ttt_progress: batch 82/782 batch_loss:2.9786 batch_bpb:1.1990 running_loss:2.7328 running_bpb:1.0684 doc_len:151-152 +ttt_progress: batch 74/782 batch_loss:3.1319 batch_bpb:1.2802 running_loss:2.7336 running_bpb:1.0688 doc_len:145-146 +ttt_progress: batch 65/782 batch_loss:3.0516 batch_bpb:1.2252 running_loss:2.7342 running_bpb:1.0691 doc_len:139-139 +ttt_progress: batch 55/782 batch_loss:3.0866 batch_bpb:1.2396 running_loss:2.7348 running_bpb:1.0694 doc_len:130-131 +ttt_progress: batch 47/782 batch_loss:2.9482 batch_bpb:1.1782 running_loss:2.7352 running_bpb:1.0696 doc_len:124-125 +ttt_progress: batch 40/782 batch_loss:3.0220 batch_bpb:1.2155 running_loss:2.7357 running_bpb:1.0698 doc_len:119-119 +ttt_progress: batch 31/782 batch_loss:3.1864 batch_bpb:1.2627 running_loss:2.7364 running_bpb:1.0701 doc_len:111-112 +ttt_progress: batch 23/782 batch_loss:3.1615 batch_bpb:1.2601 running_loss:2.7370 running_bpb:1.0704 doc_len:104-105 +ttt_progress: batch 15/782 batch_loss:3.2505 batch_bpb:1.2435 running_loss:2.7376 running_bpb:1.0706 doc_len:95-97 +ttt_progress: batch 8/782 batch_loss:3.2717 batch_bpb:1.2647 running_loss:2.7383 running_bpb:1.0709 doc_len:86-87 +ttt_progress: batch 1/782 batch_loss:3.3458 batch_bpb:1.2417 running_loss:2.7388 running_bpb:1.0710 doc_len:45-70 +quantized_ttt_lora val_loss:2.77520810 val_bpb:1.07436954 eval_time:220397ms +total_eval_time:220.4s