diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/README.md b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/README.md new file mode 100644 index 0000000000..f92327e325 --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/README.md @@ -0,0 +1,106 @@ +# SP8192 V21 + Inside-timer N-gram TTT (no Gated XSA) — val_bpb 1.05692 + +**Score: 1.05692 BPB** (3-seed mean, full val partition, seeds 42 / 0 / 1234) + +| Seed | val_bpb | val_loss | train wallclock | eval | artifact | +|------|--------:|---------:|----------------:|-----:|---------:| +| 42 | 1.05610 | 2.31114 | 596.058s | 592.2s | 15,977,032 B | +| 0 | 1.05736 | 2.31390 | 596.017s | 565.2s | 15,975,966 B | +| 1234 | 1.05730 | 2.31377 | 596.094s | 518.8s | 15,972,820 B | +| **mean** | **1.05692** | **2.31294** | **596.06s** | **558.7s** | **15,975,273 B** | +| std (pop) | 0.000580 | — | — | — | — | + +`val_tokens: 47,851,520` on every seed (full validation partition, identical to PR #1855 reference). + +vs current merged SOTA (PR #1855 1.06108): **−0.00416 BPB ≈ 0.0091 nats**, clears the 0.005-nat README threshold. + +## Approach + +This submission stacks the **eval-time recipe from PR #2018** (simon-marcus, val_bpb 1.04616) on top of the **PR #1967 V21 + LeakyReLU 0.3 training base** (ndokutovich), without Gated XSA: + +- **PR #2018's Phased TTT shape** (`PHASED_TTT_NUM_PHASES=1`, `PHASED_TTT_PREFIX_DOCS=1000`) — replaces 3-phase / 2500-prefix with one cheaper score-first phase. Frees eval budget. +- **PR #2018's N-gram tilt INSIDE the eval timer** (`NGRAM_HINT_PRECOMPUTE_OUTSIDE=0`). Per-position hint precompute (~150-160s) is part of the 600s eval budget, not setup. Same accounting as the merged A2 record (`records/track_10min_16mb/2026-04-09_A2_Muon097_3Seed/README.md` line 106: *"Eval under 600s on all 3 seeds (~436-442 s actual: ~8 s roundtrip + ~92 s sliding + ~33 s n-gram precompute + ~330-342 s TTT)"*). +- **PR #2018's LQER top-1** (`LQER_TOP_K=1`) — saves artifact bytes vs top-3. +- **GPTQ_RESERVE_SECONDS=4.0** (PR #2018) — 4 seconds reserved before training stop for GPTQ. +- **PR #1967 V21 base + LeakyReLU 0.3** (ndokutovich) — kept verbatim; preserves `WARMDOWN_FRAC=0.85`, `BETA2=0.99`, `SPARSE_ATTN_GATE_SCALE=0.5` from #1967. +- **PR #1923 AsymLogit Rescale + AWQ-lite** — already in V21 (PR #1908 + PR #1945 lineage). + +**Differences from PR #2018 (which scored 1.04616):** +- This submission **does not include Gated XSA** (PR #2018's training-time per-head attention scalar). The ~0.010 BPB gap to #2018 is the Gated XSA contribution; pre-quant post-EMA matches: this run reports 1.06117 (seed 42) vs PR #2018's 1.04930 (seed 42), a difference of 0.012 BPB. +- This submission uses PR #1967's `WARMDOWN_FRAC=0.85`, `BETA2=0.99`, `SPARSE_ATTN_GATE_SCALE=0.5` overrides; PR #2018 uses defaults (0.75 / 0.95 / 1.0). + +We submit this as additional 3-seed evidence that the inside-timer N-gram + cheap-Phased-TTT eval recipe is reproducible across V21 base variants without Gated XSA. + +## Compliance (Issue #1017) + +- **C1 strict causal dependence:** standard varlen + per-doc `cu_seqlens`; no future-token leakage. +- **C2 full normalized distribution:** standard log-softmax over SP8192 vocab; n-gram tilt is the closed-form `p'(a) = exp(β · 1[a=h]) · p(a) / Z` with `Z = 1 + p(h)(exp(β)-1)`, Σ p'(a) = 1. AsymLogit is a deterministic monotone reshape before softmax. +- **C3 score-before-update:** Phased TTT scores each chunk before any LoRA gradient step; n-gram hints are generated left-to-right from prefix state only. +- **C4 single pass:** each val token contributes exactly one BPB term in `quantized_ttt_phased`. +- **No SLOT, no n-gram cache hashing, no logit bias, no PPM, no pre-quant TTT on val data, no tokenizer change.** +- **Compute caps:** train ≤596.094s (max), eval ≤592.2s (max), all 3 seeds. `MAX_WALLCLOCK_SECONDS=600`. +- **Artifact:** ≤15,977,032 bytes (max). Cap is 16,000,000. + +## Reproduction + +```bash +pip install torch==2.9.1 --index-url https://download.pytorch.org/whl/cu128 +pip install --no-deps flash_attn_3 --find-links https://windreamer.github.io/flash-attention3-wheels/cu128_torch291/ +apt-get install -y lrzip build-essential + +# online_ngram_state.c is auto-compiled by online_ngram_tilt.py on first import +# Set DATA_DIR and TOKENIZER_PATH appropriately or use the defaults in run.sh + +for seed in 42 0 1234; do + SEED=$seed bash run.sh 2>&1 | tee train_seed${seed}.log +done +``` + +## Hyperparameters (additions over PR #1967) + +```bash +# === PR #2018 eval-time recipe additions === +PHASED_TTT_NUM_PHASES=1 # vs #1967's 3 +PHASED_TTT_PREFIX_DOCS=1000 # vs #1967's 2500 +NGRAM_HINT_PRECOMPUTE_OUTSIDE=0 # INSIDE timer (vs default outside) +LQER_TOP_K=1 # vs default top-K=3 +GPTQ_RESERVE_SECONDS=4.0 # vs default 0.5 + +# === Inherited from PR #1967 V21 === +WARMDOWN_FRAC=0.85 +BETA2=0.99 +SPARSE_ATTN_GATE_SCALE=0.5 +TTT_LR=0.75 +QK_GAIN_INIT=5.25 +TTT_NO_QV_MASK=1 +TTT_LORA_RANK=80 +EVAL_SEQ_LEN=2560 +TTT_EVAL_SEQ_LEN=2560 +NGRAM_TILT_ENABLED=1 +ASYM_LOGIT_RESCALE=1 +AWQ_LITE_ENABLED=1 +LeakyReLU squared slope = 0.3 (hardcoded in train_gpt.py) +``` + +See `run.sh` for the full env list. + +## Hardware + +8×H100 SXM 80GB (RunPod CA-MTL-1, $23.95/hr), PyTorch 2.9.1+cu128, FlashAttention 3, NVIDIA Driver 580.126.09, CUDA 13.0. + +## Attribution + +- **simon-marcus** — PR #2018 (inside-timer N-gram + 1-phase TTT + 1000-prefix + LQER_TOP_K=1 eval recipe, GPTQ_RESERVE_SECONDS=4.0) +- **ndokutovich** — PR #1967 (V21 + LeakyReLU 0.3 + n-gram tilt code base; `train_gpt.py` used here) +- **AnirudhRahul** — PR #1145 (closed-form n-gram tilt with Σ P=1 renormalization) +- **TimS-ml, lijuncheng16** — PR #1948 (LeakyReLU squared slope 0.3) +- **alertcat** — PR #1945 (V21 base composition) +- **andrewbaggio1** — PR #1953 (7-knob TTT/QK tuning) +- **romeerp** — PR #1908 (AWQ-lite mixed-precision GPTQ) + PR #1729 (CaseOps tokenizer) +- **classiclarryd** — modded-nanogpt #181 (Asymmetric Logit Rescale) +- **codemath3000** — PR #1855 (9-hparam greedy stack base) +- **dexhunter** — PR #1797 (LQER asym + SmearGate base) +- **nprime06** — PR #1787 (Polar Express NS + MIN_LR + Sparse Attn Gate + Fused CE) +- **MarioPaerle** — PR #1667 (SmearGate origin) +- **renqianluo** — PR #1767 (LoRA TTT improvements) +- **Jorge Asenjo** — PR #1700 (Phased Multi-Phase Global SGD TTT, foundational); PR #1923 (AsymLogit + AWQ-lite stack); this PR (inside-timer recipe transferred to PR #1967 base) diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/online_ngram_state.c b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/online_ngram_state.c new file mode 100644 index 0000000000..f8472a6f05 --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/online_ngram_state.c @@ -0,0 +1,433 @@ +#include +#include +#include + +#define COEFF_COUNT 32 + +static const uint64_t ROLLING_COEFFS[COEFF_COUNT] = { + 36313ULL, 27191ULL, 51647ULL, 81929ULL, 131071ULL, 196613ULL, + 262147ULL, 393241ULL, 524309ULL, 655373ULL, 786433ULL, 917521ULL, + 1048583ULL, 1179653ULL, 1310729ULL, 1441801ULL, 1572869ULL, 1703941ULL, + 1835017ULL, 1966087ULL, 2097169ULL, 2228243ULL, 2359319ULL, 2490389ULL, + 2621471ULL, 2752549ULL, 2883617ULL, 3014687ULL, 3145757ULL, 3276833ULL, + 3407903ULL, 3538973ULL, +}; + +static const uint64_t PAIR_MIX = 1000003ULL; +static const uint64_t PREFIX_BASE = 1099511628211ULL; +static const uint64_t LEN_MIX = 0x9E3779B185EBCA87ULL; +static const uint64_t TABLE_MIX = 0x9e3779b97f4a7c15ULL; + +typedef struct { + uint64_t key; + uint32_t total; + uint32_t top_count; + uint16_t top_tok; + uint16_t _pad; +} CtxBucket; + +typedef struct { + uint64_t key; + uint32_t count; + uint32_t _pad; +} PairBucket; + +typedef struct { + int token_ctx_len; + int token_prefix_len; + int token_head; + uint16_t *token_ring; + + CtxBucket *token_ctx_tbl; + uint8_t *token_ctx_used; + size_t token_ctx_mask; + + PairBucket *token_pair_tbl; + uint8_t *token_pair_used; + size_t token_pair_mask; + + uint64_t within_hash; + uint32_t within_len; + + CtxBucket *within_ctx_tbl; + uint8_t *within_ctx_used; + size_t within_ctx_mask; + + PairBucket *within_pair_tbl; + uint8_t *within_pair_used; + size_t within_pair_mask; +} OnlineNgramState; + +static inline size_t mix_index(uint64_t key, size_t mask) { + return (size_t)((key * TABLE_MIX) & mask); +} + +static inline size_t find_ctx_slot( + CtxBucket *tbl, + uint8_t *used, + size_t mask, + uint64_t key, + int *found +) { + size_t idx = mix_index(key, mask); + for (size_t probe = 0; probe <= mask; ++probe) { + if (!used[idx]) { + *found = 0; + return idx; + } + if (tbl[idx].key == key) { + *found = 1; + return idx; + } + idx = (idx + 1U) & mask; + } + *found = -1; + return 0; +} + +static inline size_t find_pair_slot( + PairBucket *tbl, + uint8_t *used, + size_t mask, + uint64_t key, + int *found +) { + size_t idx = mix_index(key, mask); + for (size_t probe = 0; probe <= mask; ++probe) { + if (!used[idx]) { + *found = 0; + return idx; + } + if (tbl[idx].key == key) { + *found = 1; + return idx; + } + idx = (idx + 1U) & mask; + } + *found = -1; + return 0; +} + +static inline uint64_t token_pair_key(uint64_t ctx_key, uint16_t tok, int ctx_len) { + return (ctx_key * PAIR_MIX) ^ (((uint64_t)tok) * ROLLING_COEFFS[(size_t)ctx_len % COEFF_COUNT]); +} + +static inline uint64_t within_pair_key(uint64_t ctx_key, uint16_t tok) { + return (ctx_key * PAIR_MIX) ^ (((uint64_t)tok) * ROLLING_COEFFS[0]); +} + +static inline uint64_t extend_prefix_hash(uint64_t current_hash, uint16_t tok, uint32_t pos) { + return (current_hash * PREFIX_BASE) ^ (((uint64_t)tok + 1ULL) * ROLLING_COEFFS[(size_t)pos % COEFF_COUNT]); +} + +static inline uint32_t pair_increment( + PairBucket *tbl, + uint8_t *used, + size_t mask, + uint64_t key +) { + int found = 0; + size_t idx = find_pair_slot(tbl, used, mask, key, &found); + if (found < 0) { + return 0U; + } + if (!found) { + used[idx] = 1U; + tbl[idx].key = key; + tbl[idx].count = 1U; + return 1U; + } + tbl[idx].count += 1U; + return tbl[idx].count; +} + +static inline int ctx_increment( + CtxBucket *tbl, + uint8_t *used, + size_t mask, + uint64_t key, + uint16_t tok, + uint32_t pair_count +) { + int found = 0; + size_t idx = find_ctx_slot(tbl, used, mask, key, &found); + if (found < 0) { + return -1; + } + if (!found) { + used[idx] = 1U; + tbl[idx].key = key; + tbl[idx].total = 1U; + tbl[idx].top_count = pair_count; + tbl[idx].top_tok = tok; + return 0; + } + tbl[idx].total += 1U; + if (pair_count > tbl[idx].top_count) { + tbl[idx].top_count = pair_count; + tbl[idx].top_tok = tok; + } + return 0; +} + +static inline uint64_t token_context_hash(const OnlineNgramState *st) { + uint64_t h = 0ULL; + if (st->token_ctx_len <= 0) { + return h; + } + for (int j = 0; j < st->token_ctx_len; ++j) { + const int ring_idx = (st->token_head + j) % st->token_ctx_len; + h ^= ((uint64_t)st->token_ring[ring_idx]) * ROLLING_COEFFS[(size_t)j]; + } + return h; +} + +static inline void token_push(OnlineNgramState *st, uint16_t tok) { + if (st->token_ctx_len <= 0) { + return; + } + if (st->token_prefix_len < st->token_ctx_len) { + st->token_ring[st->token_prefix_len] = tok; + st->token_prefix_len += 1; + return; + } + st->token_ring[st->token_head] = tok; + st->token_head = (st->token_head + 1) % st->token_ctx_len; +} + +static void *xcalloc(size_t count, size_t size) { + if (count == 0 || size == 0) { + return NULL; + } + return calloc(count, size); +} + +static int alloc_tables( + size_t table_bits, + CtxBucket **ctx_tbl, + uint8_t **ctx_used, + size_t *ctx_mask, + PairBucket **pair_tbl, + uint8_t **pair_used, + size_t *pair_mask +) { + const size_t size = 1ULL << table_bits; + *ctx_tbl = (CtxBucket *)xcalloc(size, sizeof(CtxBucket)); + *ctx_used = (uint8_t *)xcalloc(size, sizeof(uint8_t)); + *pair_tbl = (PairBucket *)xcalloc(size, sizeof(PairBucket)); + *pair_used = (uint8_t *)xcalloc(size, sizeof(uint8_t)); + if (!*ctx_tbl || !*ctx_used || !*pair_tbl || !*pair_used) { + return -1; + } + *ctx_mask = size - 1U; + *pair_mask = size - 1U; + return 0; +} + +void *online_ngram_state_create( + int token_ctx_len, + int token_table_bits, + int within_table_bits +) { + if (token_ctx_len < 0 || token_table_bits <= 0 || within_table_bits <= 0) { + return NULL; + } + OnlineNgramState *st = (OnlineNgramState *)calloc(1, sizeof(OnlineNgramState)); + if (!st) { + return NULL; + } + st->token_ctx_len = token_ctx_len; + if (token_ctx_len > 0) { + st->token_ring = (uint16_t *)xcalloc((size_t)token_ctx_len, sizeof(uint16_t)); + if (!st->token_ring) { + free(st); + return NULL; + } + } + if (alloc_tables( + (size_t)token_table_bits, + &st->token_ctx_tbl, + &st->token_ctx_used, + &st->token_ctx_mask, + &st->token_pair_tbl, + &st->token_pair_used, + &st->token_pair_mask + ) != 0) { + free(st->token_ring); + free(st); + return NULL; + } + if (alloc_tables( + (size_t)within_table_bits, + &st->within_ctx_tbl, + &st->within_ctx_used, + &st->within_ctx_mask, + &st->within_pair_tbl, + &st->within_pair_used, + &st->within_pair_mask + ) != 0) { + free(st->token_pair_used); + free(st->token_pair_tbl); + free(st->token_ctx_used); + free(st->token_ctx_tbl); + free(st->token_ring); + free(st); + return NULL; + } + return (void *)st; +} + +void online_ngram_state_destroy(void *ptr) { + OnlineNgramState *st = (OnlineNgramState *)ptr; + if (!st) { + return; + } + free(st->within_pair_used); + free(st->within_pair_tbl); + free(st->within_ctx_used); + free(st->within_ctx_tbl); + free(st->token_pair_used); + free(st->token_pair_tbl); + free(st->token_ctx_used); + free(st->token_ctx_tbl); + free(st->token_ring); + free(st); +} + +void online_ngram_state_seed_prefix_token(void *ptr, uint16_t tok) { + OnlineNgramState *st = (OnlineNgramState *)ptr; + if (!st) { + return; + } + token_push(st, tok); +} + +int online_ngram_state_process_chunk( + void *ptr, + const uint16_t *tokens, + int64_t n_tokens, + const uint8_t *starts_new_word_lut, + const uint8_t *boundary_lut, + uint16_t *token_top_token, + float *token_top_prob, + uint16_t *within_top_token, + float *within_top_prob, + uint8_t *within_valid +) { + OnlineNgramState *st = (OnlineNgramState *)ptr; + if (!st || !tokens || n_tokens < 0) { + return -1; + } + for (int64_t i = 0; i < n_tokens; ++i) { + const uint16_t tok = tokens[i]; + const uint8_t is_boundary = boundary_lut[tok]; + const uint8_t is_new_word = starts_new_word_lut[tok]; + + uint64_t token_ctx_key = 0ULL; + if (st->token_ctx_len == 0 || st->token_prefix_len >= st->token_ctx_len) { + token_ctx_key = token_context_hash(st); + int found = 0; + size_t idx = find_ctx_slot( + st->token_ctx_tbl, + st->token_ctx_used, + st->token_ctx_mask, + token_ctx_key, + &found + ); + if (found > 0) { + token_top_token[i] = st->token_ctx_tbl[idx].top_tok; + token_top_prob[i] = + (float)st->token_ctx_tbl[idx].top_count / (float)st->token_ctx_tbl[idx].total; + } else { + token_top_token[i] = 0U; + token_top_prob[i] = 0.0f; + } + } else { + token_top_token[i] = 0U; + token_top_prob[i] = 0.0f; + } + + uint64_t within_ctx_key = 0ULL; + if (!is_boundary && !is_new_word && st->within_len > 0U) { + within_ctx_key = st->within_hash ^ ((uint64_t)st->within_len * LEN_MIX); + int found = 0; + size_t idx = find_ctx_slot( + st->within_ctx_tbl, + st->within_ctx_used, + st->within_ctx_mask, + within_ctx_key, + &found + ); + within_valid[i] = 1U; + if (found > 0) { + within_top_token[i] = st->within_ctx_tbl[idx].top_tok; + within_top_prob[i] = + (float)st->within_ctx_tbl[idx].top_count / (float)st->within_ctx_tbl[idx].total; + } else { + within_top_token[i] = 0U; + within_top_prob[i] = 0.0f; + } + } else { + within_valid[i] = 0U; + within_top_token[i] = 0U; + within_top_prob[i] = 0.0f; + } + + if (st->token_ctx_len == 0 || st->token_prefix_len >= st->token_ctx_len) { + const uint64_t pair_key = token_pair_key(token_ctx_key, tok, st->token_ctx_len); + const uint32_t pair_count = pair_increment( + st->token_pair_tbl, + st->token_pair_used, + st->token_pair_mask, + pair_key + ); + if (pair_count == 0U) { + return -2; + } + if (ctx_increment( + st->token_ctx_tbl, + st->token_ctx_used, + st->token_ctx_mask, + token_ctx_key, + tok, + pair_count + ) != 0) { + return -3; + } + } + token_push(st, tok); + + if (is_boundary) { + st->within_hash = 0ULL; + st->within_len = 0U; + continue; + } + if (is_new_word || st->within_len == 0U) { + st->within_hash = extend_prefix_hash(0ULL, tok, 0U); + st->within_len = 1U; + continue; + } + const uint32_t within_pair_count = pair_increment( + st->within_pair_tbl, + st->within_pair_used, + st->within_pair_mask, + within_pair_key(within_ctx_key, tok) + ); + if (within_pair_count == 0U) { + return -4; + } + if (ctx_increment( + st->within_ctx_tbl, + st->within_ctx_used, + st->within_ctx_mask, + within_ctx_key, + tok, + within_pair_count + ) != 0) { + return -5; + } + st->within_hash = extend_prefix_hash(st->within_hash, tok, st->within_len); + st->within_len += 1U; + } + return 0; +} diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/online_ngram_tilt.py b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/online_ngram_tilt.py new file mode 100644 index 0000000000..973c21866f --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/online_ngram_tilt.py @@ -0,0 +1,386 @@ +""" +Vendored online n-gram tilt helpers from PR #1145 (AnirudhRahul, valerio-endorsed). + +Provides causal, normalized, prefix-only n-gram experts that propose at most one +hinted token per scored position. Caller obtains q_t = p(h_t | x) from the model +(post-TTT-adapt logits) and applies multiplicative-boost-with-renorm: + + p'(a) = exp(beta * 1[a == h_t]) * p(a) / Z_t + Z_t = 1 - q_t + exp(beta) * q_t = 1 + q_t * (exp(beta) - 1) + -log p'(y_realized) = -log p(y) - beta * 1[y == h_t] + log Z_t + = ptl - beta * is_hit + log1p(q_t * (exp(beta) - 1)) + +Compliance: +- C1 causal: hint h_t computed from strict prefix (tokens 0..t-1 only) +- C2 normalized over Sigma: closed-form Z_t over full vocab softmax +- C3 score-before-update: hints precomputed in single L->R pass; loss uses prefix-only +- C4 single pass: process_chunk advances state monotonically + +Compatible with both #1934/#1855 base architectures via Hyperparameter env-var gates. +""" + +from __future__ import annotations + +import ctypes +import math +import os +import subprocess +from collections import deque +from pathlib import Path + +import numpy as np +import sentencepiece as spm +import torch + + +SCRIPT_DIR = Path(__file__).resolve().parent +ONLINE_NGRAM_SRC = SCRIPT_DIR / "online_ngram_state.c" +ONLINE_NGRAM_LIB = SCRIPT_DIR / "libonline_ngram_state.so" + +WHITESPACE_BYTE_IDS = {9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 36} +EDGE_PUNCT = ".,:;!?()[]{}<>\"'`" + + +def normalize_word(text: str, mode: str) -> str: + text = text.strip() + if mode == "lower": + return text.lower() + if mode == "identity": + return text + if mode == "strip_punct_lower": + return text.strip(EDGE_PUNCT).lower() + raise ValueError(f"Unknown word normalization mode: {mode}") + + +def suggest_table_bits(expected_entries: int, load_factor: float) -> int: + if expected_entries <= 0: + return 16 + target = max(int(expected_entries / max(load_factor, 1e-6)), 1) + bits = max(int(math.ceil(math.log2(target))), 12) + return min(bits, 28) + + +def ensure_online_ngram_lib(log0=print) -> ctypes.CDLL: + needs_build = (not ONLINE_NGRAM_LIB.exists()) or ( + ONLINE_NGRAM_SRC.stat().st_mtime_ns > ONLINE_NGRAM_LIB.stat().st_mtime_ns + ) + if needs_build: + log0(f"ngram_tilt:building_native_helper src={ONLINE_NGRAM_SRC.name}") + subprocess.run( + [ + "gcc", "-O3", "-march=native", "-shared", "-fPIC", + "-o", str(ONLINE_NGRAM_LIB), + str(ONLINE_NGRAM_SRC), + ], + check=True, + ) + lib = ctypes.CDLL(str(ONLINE_NGRAM_LIB)) + lib.online_ngram_state_create.restype = ctypes.c_void_p + lib.online_ngram_state_create.argtypes = [ctypes.c_int, ctypes.c_int, ctypes.c_int] + lib.online_ngram_state_destroy.restype = None + lib.online_ngram_state_destroy.argtypes = [ctypes.c_void_p] + lib.online_ngram_state_seed_prefix_token.restype = None + lib.online_ngram_state_seed_prefix_token.argtypes = [ctypes.c_void_p, ctypes.c_uint16] + lib.online_ngram_state_process_chunk.restype = ctypes.c_int + lib.online_ngram_state_process_chunk.argtypes = [ + ctypes.c_void_p, + ctypes.POINTER(ctypes.c_uint16), + ctypes.c_int64, + ctypes.POINTER(ctypes.c_uint8), + ctypes.POINTER(ctypes.c_uint8), + ctypes.POINTER(ctypes.c_uint16), + ctypes.POINTER(ctypes.c_float), + ctypes.POINTER(ctypes.c_uint16), + ctypes.POINTER(ctypes.c_float), + ctypes.POINTER(ctypes.c_uint8), + ] + return lib + + +class OnlineNgramState: + def __init__( + self, *, lib, token_ctx_len, token_table_bits, within_table_bits, + starts_new_word_lut, boundary_lut, seed_prefix_token, + ): + self.lib = lib + self.state = lib.online_ngram_state_create(token_ctx_len, token_table_bits, within_table_bits) + if not self.state: + raise RuntimeError( + f"Native ngram state alloc failed token_table_bits={token_table_bits} within_table_bits={within_table_bits}" + ) + self.starts_new_word_lut = np.ascontiguousarray(starts_new_word_lut.astype(np.uint8, copy=False)) + self.boundary_lut = np.ascontiguousarray(boundary_lut.astype(np.uint8, copy=False)) + self.lib.online_ngram_state_seed_prefix_token(self.state, ctypes.c_uint16(int(seed_prefix_token))) + + def close(self): + if self.state: + self.lib.online_ngram_state_destroy(self.state) + self.state = None + + def __del__(self): + self.close() + + def process_chunk(self, chunk_tokens): + chunk_tokens = np.ascontiguousarray(chunk_tokens.astype(np.uint16, copy=False)) + n = int(chunk_tokens.size) + token_top_token = np.zeros(n, dtype=np.uint16) + token_top_prob = np.zeros(n, dtype=np.float32) + within_top_token = np.zeros(n, dtype=np.uint16) + within_top_prob = np.zeros(n, dtype=np.float32) + within_valid = np.zeros(n, dtype=np.uint8) + rc = self.lib.online_ngram_state_process_chunk( + self.state, + chunk_tokens.ctypes.data_as(ctypes.POINTER(ctypes.c_uint16)), + ctypes.c_int64(n), + self.starts_new_word_lut.ctypes.data_as(ctypes.POINTER(ctypes.c_uint8)), + self.boundary_lut.ctypes.data_as(ctypes.POINTER(ctypes.c_uint8)), + token_top_token.ctypes.data_as(ctypes.POINTER(ctypes.c_uint16)), + token_top_prob.ctypes.data_as(ctypes.POINTER(ctypes.c_float)), + within_top_token.ctypes.data_as(ctypes.POINTER(ctypes.c_uint16)), + within_top_prob.ctypes.data_as(ctypes.POINTER(ctypes.c_float)), + within_valid.ctypes.data_as(ctypes.POINTER(ctypes.c_uint8)), + ) + if rc != 0: + raise RuntimeError(f"Native ngram process_chunk failed rc={rc}") + return token_top_token, token_top_prob, within_top_token, within_top_prob, within_valid.astype(bool) + + +class WordStartState: + def __init__(self, *, sp, order, normalize_mode): + self.sp = sp + self.ctx_w = max(order - 1, 0) + self.normalize_mode = normalize_mode + self.prev_word_ids: deque = deque(maxlen=self.ctx_w) + self.current_word_tokens: list = [] + self.word_to_id: dict = {} + self.next_word_id = 1 + self.ctx_total: dict = {} + self.pair_count: dict = {} + self.ctx_best_token: dict = {} + self.ctx_best_count: dict = {} + + def _flush_current_word(self): + if not self.current_word_tokens: + return + text = normalize_word(self.sp.decode(self.current_word_tokens), self.normalize_mode) + if text: + wid = self.word_to_id.get(text) + if wid is None: + wid = self.next_word_id + self.word_to_id[text] = wid + self.next_word_id += 1 + if self.ctx_w > 0: + self.prev_word_ids.append(wid) + self.current_word_tokens = [] + + def process_chunk(self, chunk_tokens, *, starts_new_word_lut, boundary_lut): + chunk_tokens = np.ascontiguousarray(chunk_tokens.astype(np.uint16, copy=False)) + top_token = np.zeros(chunk_tokens.size, dtype=np.uint16) + top_prob = np.zeros(chunk_tokens.size, dtype=np.float32) + for i, tok_u16 in enumerate(chunk_tokens): + tok = int(tok_u16) + is_boundary = bool(boundary_lut[tok]) + is_word_start = bool(starts_new_word_lut[tok]) or not self.current_word_tokens + if is_boundary: + self._flush_current_word() + continue + if bool(starts_new_word_lut[tok]): + self._flush_current_word() + ctx_key = None + if is_word_start and len(self.prev_word_ids) >= self.ctx_w: + ctx_key = tuple(self.prev_word_ids) if self.ctx_w > 0 else () + total = self.ctx_total.get(ctx_key, 0) + if total > 0: + top_token[i] = np.uint16(self.ctx_best_token[ctx_key]) + top_prob[i] = np.float32(self.ctx_best_count[ctx_key] / total) + if is_word_start: + if ctx_key is not None: + pair_key = (ctx_key, tok) + pair = self.pair_count.get(pair_key, 0) + 1 + self.pair_count[pair_key] = pair + total = self.ctx_total.get(ctx_key, 0) + 1 + self.ctx_total[ctx_key] = total + best_count = self.ctx_best_count.get(ctx_key, 0) + if pair > best_count: + self.ctx_best_count[ctx_key] = pair + self.ctx_best_token[ctx_key] = tok + self.current_word_tokens = [tok] + else: + self.current_word_tokens.append(tok) + return top_token, top_prob + + +def build_piece_luts(*, tokenizer_path, vocab_size): + sp = spm.SentencePieceProcessor(model_file=tokenizer_path) + pieces = [sp.id_to_piece(i) for i in range(sp.vocab_size())] + starts_new_word_lut = np.zeros(vocab_size, dtype=np.uint8) + for i, piece in enumerate(pieces): + starts_new_word_lut[i] = 1 if piece.startswith("▁") else 0 + boundary_lut = np.zeros(vocab_size, dtype=np.uint8) + bos_id = sp.bos_id() + if bos_id >= 0 and bos_id < vocab_size: + boundary_lut[bos_id] = 1 + for tok in range(min(sp.vocab_size(), vocab_size)): + if sp.is_byte(tok) and tok in WHITESPACE_BYTE_IDS: + boundary_lut[tok] = 1 + return sp, starts_new_word_lut, boundary_lut + + +def build_hints_for_targets( + *, target_token_ids_np, tokenizer_path, vocab_size, log0=print, + token_order=16, token_threshold=0.800, token_boost=2.625, + within_tau=0.450, within_boost=0.750, + word_order=4, word_normalize="strip_punct_lower", + word_tau=0.650, word_boost=0.750, + agree_add_boost=0.500, +): + """Single L->R pass. Returns dict with hint_ids, gate_mask, boost_per_pos. + + target_token_ids_np: np.uint16 array of realized targets (length = total_targets). + Output arrays are aligned to target_token_ids_np indexing. + + For each scored position t we pick at most one hint h_t: + - prefer the expert with highest expected gain = p_top * boost - log1p(p_top * (exp(boost)-1)) + - if multiple experts agree on the same h_t, additive boost agree_add_boost + - gate (don't tilt) when no expert clears its threshold + + The realized loss formula used by the caller: + ptl' = ptl - beta * 1[y == h_t] + log1p(q_t * (exp(beta) - 1)) when gate_mask == True + ptl' = ptl when gate_mask == False + """ + sp, starts_new_word_lut, boundary_lut = build_piece_luts( + tokenizer_path=tokenizer_path, vocab_size=vocab_size + ) + total = int(target_token_ids_np.size) + if total == 0: + return { + "hint_ids": np.zeros(0, dtype=np.int64), + "gate_mask": np.zeros(0, dtype=bool), + "boost": np.zeros(0, dtype=np.float32), + "sp": sp, + "starts_new_word_lut": starts_new_word_lut, + "boundary_lut": boundary_lut, + } + + token_table_bits = suggest_table_bits(total, load_factor=0.55) + within_table_bits = suggest_table_bits(max(total // 2, 1), load_factor=0.60) + online_lib = ensure_online_ngram_lib(log0) + ngram_state = OnlineNgramState( + lib=online_lib, + token_ctx_len=max(token_order - 1, 0), + token_table_bits=token_table_bits, + within_table_bits=within_table_bits, + starts_new_word_lut=starts_new_word_lut, + boundary_lut=boundary_lut, + seed_prefix_token=int(target_token_ids_np[0]), + ) + word_state = WordStartState(sp=sp, order=word_order, normalize_mode=word_normalize) + + token_top_tok, token_top_prob, within_top_tok, within_top_prob, within_valid = ( + ngram_state.process_chunk(target_token_ids_np) + ) + word_top_tok, word_top_prob = word_state.process_chunk( + target_token_ids_np, + starts_new_word_lut=starts_new_word_lut, + boundary_lut=boundary_lut, + ) + + def _expected_gain(p_top, boost): + # E[ -log p'(y) under -log p(y)] when y ~ p + # = p_top * boost - log1p(p_top * (exp(boost) - 1)) + # Maximizing this over experts => pick the most informative hint. + log_norm = np.log1p(p_top * (math.exp(boost) - 1.0)) + return p_top * boost - log_norm + + token_gate = token_top_prob >= np.float32(token_threshold) + within_gate = within_valid & (within_top_prob >= np.float32(within_tau)) + word_gate = word_top_prob >= np.float32(word_tau) + + token_gain = np.where(token_gate, _expected_gain(token_top_prob.astype(np.float64), token_boost), -np.inf) + within_gain = np.where(within_gate, _expected_gain(within_top_prob.astype(np.float64), within_boost), -np.inf) + word_gain = np.where(word_gate, _expected_gain(word_top_prob.astype(np.float64), word_boost), -np.inf) + + stack = np.stack([token_gain, within_gain, word_gain], axis=1) + best_idx = np.argmax(stack, axis=1) + best_gain = np.max(stack, axis=1) + any_gate = best_gain > -np.inf + + hint_ids = np.zeros(total, dtype=np.int64) + boost = np.zeros(total, dtype=np.float32) + base_boost_per_expert = np.array([token_boost, within_boost, word_boost], dtype=np.float32) + hint_per_expert = np.stack([ + token_top_tok.astype(np.int64), + within_top_tok.astype(np.int64), + word_top_tok.astype(np.int64), + ], axis=1) + + rows = np.arange(total) + hint_ids[any_gate] = hint_per_expert[rows[any_gate], best_idx[any_gate]] + boost[any_gate] = base_boost_per_expert[best_idx[any_gate]] + + # Agreement bonus: if 2+ experts agree on the same hint as best, add agree_add_boost + gate_mask_each = np.stack([token_gate, within_gate, word_gate], axis=1) + expert_hints = hint_per_expert.copy() + expert_hints[~gate_mask_each] = -1 + agreements = (expert_hints == hint_ids[:, None]).sum(axis=1) + agreement_extra = np.where(agreements >= 2, np.float32(agree_add_boost), np.float32(0.0)) + boost = (boost + agreement_extra).astype(np.float32) + + log0( + f"ngram_tilt:hints total={total} gated={int(any_gate.sum())} " + f"token_gate={int(token_gate.sum())} within_gate={int(within_gate.sum())} word_gate={int(word_gate.sum())} " + f"agree2plus={int((agreements >= 2).sum())}" + ) + + return { + "hint_ids": hint_ids, + "gate_mask": any_gate, + "boost": boost, + "sp": sp, + "starts_new_word_lut": starts_new_word_lut, + "boundary_lut": boundary_lut, + } + + +def apply_tilt_to_ptl_torch( + ptl: torch.Tensor, + log_q_hint: torch.Tensor, + target_ids: torch.Tensor, + hint_ids: torch.Tensor, + gate_mask: torch.Tensor, + boost: torch.Tensor, +): + """Closed-form tilt applied to per-token NLL. + + All tensors same shape [..., L]. + ptl_tilted = ptl - beta * 1[y == h] + log1p(q * (exp(beta) - 1)) if gate else ptl + """ + boost64 = boost.to(torch.float64) + q = log_q_hint.to(torch.float64).clamp_(max=0.0).exp() + is_hit = (target_ids == hint_ids).to(torch.float64) + log_Z = torch.log1p(q * (torch.expm1(boost64))) + ptl_tilted = ptl.to(torch.float64) - boost64 * is_hit + log_Z + return torch.where(gate_mask, ptl_tilted, ptl.to(torch.float64)).to(ptl.dtype) + + +def apply_tilt_to_ptl_torch_fast( + ptl: torch.Tensor, + log_q_hint: torch.Tensor, + target_ids: torch.Tensor, + hint_ids: torch.Tensor, + gate_mask: torch.Tensor, + boost: torch.Tensor, +): + """fp32 variant of apply_tilt — cast removed where safe. + + BPB downstream accumulator is fp64, so per-token tilt computation in + fp32 has no impact on final precision. Saves ~10-15s per eval pass on + H100 (avoids fp64 ALU + double memory traffic). + """ + boost32 = boost.to(torch.float32) + q = log_q_hint.to(torch.float32).clamp_(max=0.0).exp() + is_hit = (target_ids == hint_ids).to(torch.float32) + log_Z = torch.log1p(q * (torch.expm1(boost32))) + ptl_f32 = ptl.to(torch.float32) + ptl_tilted = ptl_f32 - boost32 * is_hit + log_Z + return torch.where(gate_mask, ptl_tilted, ptl_f32).to(ptl.dtype) diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/run.sh b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/run.sh new file mode 100644 index 0000000000..cbcffe12c7 --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/run.sh @@ -0,0 +1,46 @@ +#!/bin/bash +# #2018 emulation (without Gated XSA): V21 + ngram_inside + LeakyReLU 0.3 + Phased TTT 1ph/1000pref + LQER_TOP_K=1 +# Predicted: ~1.048-1.050 (vs #2018 1.04616 = ours minus Gated XSA contribution ~0.002) +set -e +cd "$(dirname "$0")" + +DATA_DIR="${DATA_DIR:-/workspace/pg-data/datasets}" +DATA_PATH="${DATA_PATH:-$DATA_DIR/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved}" +TOKENIZER_PATH="${TOKENIZER_PATH:-$DATA_DIR/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model}" +SEED="${SEED:-42}" + +env_vars=( + DATA_DIR="$DATA_DIR" DATA_PATH="$DATA_PATH" TOKENIZER_PATH="$TOKENIZER_PATH" + CASEOPS_ENABLED=1 VOCAB_SIZE=8192 ITERATIONS=20000 MAX_WALLCLOCK_SECONDS=600 + WARMUP_STEPS=20 WARMDOWN_FRAC=0.85 BETA2=0.99 GRAD_CLIP_NORM=0.3 + MIN_LR=0.1 MATRIX_LR=0.026 GLOBAL_TTT_MOMENTUM=0.9 + SPARSE_ATTN_GATE_ENABLED=1 SPARSE_ATTN_GATE_SCALE=0.5 + SMEAR_GATE_ENABLED=1 GATE_WINDOW=12 GATED_ATTN_QUANT_GATE=1 + FUSED_CE_ENABLED=1 EMBED_BITS=7 + MLP_CLIP_SIGMAS=11.5 ATTN_CLIP_SIGMAS=13.0 EMBED_CLIP_SIGMAS=14.0 + GPTQ_RESERVE_SECONDS=4.0 GPTQ_CALIBRATION_BATCHES=16 + COMPRESSOR=pergroup + LQER_ENABLED=1 LQER_RANK=4 LQER_TOP_K=1 LQER_GROUP_SIZE=64 LQER_ASYM_ENABLED=1 LQER_ASYM_GROUP=64 + ASYM_LOGIT_RESCALE=1 AWQ_LITE_ENABLED=1 + # === #2018 key changes vs Stack-B === + PHASED_TTT_ENABLED=1 + PHASED_TTT_NUM_PHASES=1 + PHASED_TTT_PREFIX_DOCS=1000 + TTT_LORA_RANK=80 + # === #1953 7 knobs === + TTT_LR=0.75 QK_GAIN_INIT=5.25 TTT_NO_QV_MASK=1 + TTT_Q_LORA=0 TTT_V_LORA=0 TTT_LOCAL_LR_MULT=0.75 + EVAL_SEQ_LEN=2560 TTT_EVAL_SEQ_LEN=2560 + # === N-gram tilt INSIDE timer (key #2018 trick) === + NGRAM_TILT_ENABLED=1 + NGRAM_HINT_PRECOMPUTE_OUTSIDE=0 + TOKEN_ORDER=16 TOKEN_THRESHOLD=0.800 TOKEN_BOOST=2.625 + WITHIN_TAU=0.450 WITHIN_BOOST=0.750 + WORD_ORDER=4 WORD_NORMALIZE=strip_punct_lower WORD_TAU=0.650 WORD_BOOST=0.750 + AGREE_ADD_BOOST=0.500 + NCCL_NET=Socket SEED="$SEED" +) +echo "=== #2018 emulation (no Gated XSA) seed=$SEED ===" +echo "PHASED_TTT_NUM_PHASES=1 PHASED_TTT_PREFIX_DOCS=1000 NGRAM_INSIDE=1 RANK=80" +date +env "${env_vars[@]}" torchrun --standalone --nproc_per_node=8 train_gpt.py diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/submission.json b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/submission.json new file mode 100644 index 0000000000..54062d4843 --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/submission.json @@ -0,0 +1,11 @@ +{ + "author": "Jorge Asenjo", + "github_id": "jorge-asenjo", + "name": "SP8192 V21 + Inside-timer N-gram TTT (no Gated XSA)", + "blurb": "PR #1967 V21+LeakyReLU 0.3 training base + PR #2018 eval-time recipe (Phased TTT 1 phase / 1000 prefix, NGRAM_HINT_PRECOMPUTE_OUTSIDE=0 inside timer, LQER_TOP_K=1, GPTQ_RESERVE_SECONDS=4.0) without Gated XSA. Inherits PR #1923 AsymLogit + PR #1908 AWQ-lite. Validates that the inside-timer N-gram + cheap-Phased-TTT eval recipe from PR #2018 reproduces on the PR #1967 training stack across 3 seeds. 3-seed mean val_bpb 1.05692 (population std 0.000580), full validation partition (47,851,520 tokens). All 3 seeds: train ≤596.094s, eval ≤592.2s, max artifact 15,977,032 bytes. N-gram precompute (~150-160s) measured INSIDE the eval timer per the merged A2 record convention.", + "date": "2026-05-01T00:20:00Z", + "val_loss": 2.31294, + "val_bpb": 1.05692, + "bytes_total": 15977032, + "bytes_code": 35707 +} diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_gpt.py b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_gpt.py new file mode 100644 index 0000000000..b9583fa832 --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_gpt.py @@ -0,0 +1,4293 @@ +import base64, collections, copy, fcntl, glob, io, lzma, math, os +from pathlib import Path +import random, re, subprocess, sys, time, uuid, numpy as np, sentencepiece as spm, torch, torch.distributed as dist, torch.nn.functional as F +from torch import Tensor, nn +from flash_attn_interface import ( + flash_attn_func as flash_attn_3_func, + flash_attn_varlen_func, +) +from concurrent.futures import ThreadPoolExecutor +import triton +import triton.language as tl +from triton.tools.tensor_descriptor import TensorDescriptor + + +# ===== Fused softcapped cross-entropy (Triton) — training-only path ===== +# Replaces the eager +# logits_softcap = softcap * tanh(logits / softcap) +# F.cross_entropy(logits_softcap.float(), targets, reduction="mean") +# sequence with a single fused kernel that reads logits_proj once, applies +# softcap in-register, and computes (LSE, loss) in one streaming pass. The +# backward kernel mirrors the forward so there's no stored softcapped logits. +# Numerically identical to the eager path up to fp32 accumulation differences. +_FUSED_CE_LIBRARY = "pgsubmission1draft7fusedce" +_FUSED_CE_BLOCK_SIZE = 1024 +_FUSED_CE_NUM_WARPS = 4 + + +@triton.jit +def _softcapped_ce_fwd_kernel( + logits_ptr, losses_ptr, lse_ptr, targets_ptr, + stride_logits_n, stride_logits_v, + n_rows, n_cols, softcap, + block_size: tl.constexpr, +): + row_idx = tl.program_id(0).to(tl.int64) + logits_row_ptr = logits_ptr + row_idx * stride_logits_n + max_val = -float("inf") + sum_exp = 0.0 + A = 2.0 * softcap + inv_C = 2.0 / softcap + for off in range(0, n_cols, block_size): + cols = off + tl.arange(0, block_size) + mask = cols < n_cols + val = tl.load( + logits_row_ptr + cols * stride_logits_v, + mask=mask, other=-float("inf"), + ).to(tl.float32) + z = A * tl.sigmoid(val * inv_C) + z = tl.where(mask, z, -float("inf")) + curr_max = tl.max(z, axis=0) + new_max = tl.maximum(max_val, curr_max) + sum_exp = sum_exp * tl.exp(max_val - new_max) + tl.sum(tl.exp(z - new_max), axis=0) + max_val = new_max + lse = max_val + tl.log(sum_exp) + tl.store(lse_ptr + row_idx, lse) + target = tl.load(targets_ptr + row_idx).to(tl.int32) + target_val = tl.load(logits_row_ptr + target * stride_logits_v).to(tl.float32) + target_z = A * tl.sigmoid(target_val * inv_C) + tl.store(losses_ptr + row_idx, lse - target_z) + + +@triton.jit +def _softcapped_ce_bwd_kernel( + grad_logits_ptr, grad_losses_ptr, lse_ptr, logits_ptr, targets_ptr, + stride_logits_n, stride_logits_v, + stride_grad_n, stride_grad_v, + n_rows, n_cols, softcap, + block_size: tl.constexpr, +): + row_idx = tl.program_id(0).to(tl.int64) + logits_row_ptr = logits_ptr + row_idx * stride_logits_n + grad_row_ptr = grad_logits_ptr + row_idx * stride_grad_n + lse = tl.load(lse_ptr + row_idx) + grad_loss = tl.load(grad_losses_ptr + row_idx).to(tl.float32) + target = tl.load(targets_ptr + row_idx).to(tl.int32) + A = 2.0 * softcap + inv_C = 2.0 / softcap + dz_dx_scale = A * inv_C + for off in range(0, n_cols, block_size): + cols = off + tl.arange(0, block_size) + mask = cols < n_cols + val = tl.load( + logits_row_ptr + cols * stride_logits_v, + mask=mask, other=0.0, + ).to(tl.float32) + sigmoid_u = tl.sigmoid(val * inv_C) + z = A * sigmoid_u + probs = tl.exp(z - lse) + grad_z = grad_loss * (probs - tl.where(cols == target, 1.0, 0.0)) + grad_x = grad_z * (dz_dx_scale * sigmoid_u * (1.0 - sigmoid_u)) + tl.store(grad_row_ptr + cols * stride_grad_v, grad_x, mask=mask) + + +def _validate_softcapped_ce_inputs( + logits: Tensor, targets: Tensor, softcap: float, +) -> tuple[Tensor, Tensor]: + if logits.ndim != 2: + raise ValueError(f"Expected logits.ndim=2, got {logits.ndim}") + if targets.ndim != 1: + raise ValueError(f"Expected targets.ndim=1, got {targets.ndim}") + if logits.shape[0] != targets.shape[0]: + raise ValueError( + f"Expected matching rows, got logits={tuple(logits.shape)} targets={tuple(targets.shape)}" + ) + if not logits.is_cuda or not targets.is_cuda: + raise ValueError("softcapped_cross_entropy requires CUDA tensors") + if softcap <= 0.0: + raise ValueError(f"softcap must be positive, got {softcap}") + if logits.dtype not in (torch.float16, torch.bfloat16, torch.float32): + raise ValueError(f"Unsupported logits dtype: {logits.dtype}") + logits = logits.contiguous() + targets = targets.contiguous() + if targets.dtype != torch.int64: + targets = targets.to(dtype=torch.int64) + return logits, targets + + +@torch.library.custom_op(f"{_FUSED_CE_LIBRARY}::softcapped_ce", mutates_args=()) +def softcapped_ce_op(logits: Tensor, targets: Tensor, softcap: float) -> tuple[Tensor, Tensor]: + logits, targets = _validate_softcapped_ce_inputs(logits, targets, float(softcap)) + n_rows, n_cols = logits.shape + losses = torch.empty((n_rows,), device=logits.device, dtype=torch.float32) + lse = torch.empty((n_rows,), device=logits.device, dtype=torch.float32) + _softcapped_ce_fwd_kernel[(n_rows,)]( + logits, losses, lse, targets, + logits.stride(0), logits.stride(1), + n_rows, n_cols, float(softcap), + block_size=_FUSED_CE_BLOCK_SIZE, num_warps=_FUSED_CE_NUM_WARPS, + ) + return losses, lse + + +@softcapped_ce_op.register_fake +def _(logits: Tensor, targets: Tensor, softcap: float): + if logits.ndim != 2 or targets.ndim != 1: + raise ValueError("softcapped_ce fake impl expects 2D logits and 1D targets") + if logits.shape[0] != targets.shape[0]: + raise ValueError( + f"Expected matching rows, got logits={tuple(logits.shape)} targets={tuple(targets.shape)}" + ) + n_rows = logits.shape[0] + return ( + logits.new_empty((n_rows,), dtype=torch.float32), + logits.new_empty((n_rows,), dtype=torch.float32), + ) + + +@torch.library.custom_op(f"{_FUSED_CE_LIBRARY}::softcapped_ce_backward", mutates_args=()) +def softcapped_ce_backward_op( + logits: Tensor, targets: Tensor, lse: Tensor, grad_losses: Tensor, softcap: float, +) -> Tensor: + logits, targets = _validate_softcapped_ce_inputs(logits, targets, float(softcap)) + lse = lse.contiguous() + grad_losses = grad_losses.contiguous().to(dtype=torch.float32) + if lse.ndim != 1 or grad_losses.ndim != 1: + raise ValueError("Expected 1D lse and grad_losses") + if lse.shape[0] != logits.shape[0] or grad_losses.shape[0] != logits.shape[0]: + raise ValueError( + f"Expected row-aligned lse/grad_losses, got logits={tuple(logits.shape)} " + f"lse={tuple(lse.shape)} grad_losses={tuple(grad_losses.shape)}" + ) + grad_logits = torch.empty_like(logits) + n_rows, n_cols = logits.shape + _softcapped_ce_bwd_kernel[(n_rows,)]( + grad_logits, grad_losses, lse, logits, targets, + logits.stride(0), logits.stride(1), + grad_logits.stride(0), grad_logits.stride(1), + n_rows, n_cols, float(softcap), + block_size=_FUSED_CE_BLOCK_SIZE, num_warps=_FUSED_CE_NUM_WARPS, + ) + return grad_logits + + +@softcapped_ce_backward_op.register_fake +def _(logits: Tensor, targets: Tensor, lse: Tensor, grad_losses: Tensor, softcap: float): + if logits.ndim != 2 or targets.ndim != 1 or lse.ndim != 1 or grad_losses.ndim != 1: + raise ValueError("softcapped_ce_backward fake impl expects 2D logits and 1D row tensors") + if ( + logits.shape[0] != targets.shape[0] + or logits.shape[0] != lse.shape[0] + or logits.shape[0] != grad_losses.shape[0] + ): + raise ValueError("softcapped_ce_backward fake impl expects row-aligned tensors") + return logits.new_empty(logits.shape) + + +def _softcapped_ce_setup_context( + ctx: torch.autograd.function.FunctionCtx, inputs, output, +) -> None: + logits, targets, softcap = inputs + _losses, lse = output + ctx.save_for_backward(logits, targets, lse) + ctx.softcap = float(softcap) + + +def _softcapped_ce_backward( + ctx: torch.autograd.function.FunctionCtx, grad_losses: Tensor, grad_lse: "Tensor | None", +): + del grad_lse + logits, targets, lse = ctx.saved_tensors + grad_logits = torch.ops.pgsubmission1draft7fusedce.softcapped_ce_backward( + logits, targets, lse, grad_losses, ctx.softcap + ) + return grad_logits, None, None + + +softcapped_ce_op.register_autograd( + _softcapped_ce_backward, setup_context=_softcapped_ce_setup_context, +) + + +def softcapped_cross_entropy( + logits: Tensor, targets: Tensor, softcap: float, reduction: str = "mean", +) -> Tensor: + losses, _lse = torch.ops.pgsubmission1draft7fusedce.softcapped_ce( + logits, targets, float(softcap) + ) + if reduction == "none": + return losses + if reduction == "sum": + return losses.sum() + if reduction == "mean": + return losses.mean() + raise ValueError(f"Unsupported reduction={reduction!r}") + + +class Hyperparameters: + data_dir = os.environ.get("DATA_DIR", "./data/") + seed = int(os.environ.get("SEED", 1337)) + run_id = os.environ.get("RUN_ID", str(uuid.uuid4())) + iterations = int(os.environ.get("ITERATIONS", 20000)) + warmdown_frac = float(os.environ.get("WARMDOWN_FRAC", 0.75)) + warmup_steps = int(os.environ.get("WARMUP_STEPS", 20)) + train_batch_tokens = int(os.environ.get("TRAIN_BATCH_TOKENS", 786432)) + # Fused softcapped CE (Triton). Training-only — forward_logits eval path still uses + # eager softcap+F.cross_entropy. Default ON since validated as at-worst neutral. + fused_ce_enabled = bool(int(os.environ.get("FUSED_CE_ENABLED", "1"))) + train_seq_len = int(os.environ.get("TRAIN_SEQ_LEN", 2048)) + train_log_every = int(os.environ.get("TRAIN_LOG_EVERY", 500)) + max_wallclock_seconds = float(os.environ.get("MAX_WALLCLOCK_SECONDS", 6e2)) + val_batch_tokens = int(os.environ.get("VAL_BATCH_TOKENS", 524288)) + eval_seq_len = int(os.environ.get("EVAL_SEQ_LEN", 2048)) + val_loss_every = int(os.environ.get("VAL_LOSS_EVERY", 4000)) + vocab_size = int(os.environ.get("VOCAB_SIZE", 8192)) + num_layers = int(os.environ.get("NUM_LAYERS", 11)) + xsa_last_n = int(os.environ.get("XSA_LAST_N", 11)) + model_dim = int(os.environ.get("MODEL_DIM", 512)) + num_kv_heads = int(os.environ.get("NUM_KV_HEADS", 4)) + num_heads = int(os.environ.get("NUM_HEADS", 8)) + mlp_mult = float(os.environ.get("MLP_MULT", 4.0)) + skip_gates_enabled = bool(int(os.environ.get("SKIP_GATES_ENABLED", "1"))) + tie_embeddings = bool(int(os.environ.get("TIE_EMBEDDINGS", "1"))) + logit_softcap = float(os.environ.get("LOGIT_SOFTCAP", 3e1)) + rope_base = float(os.environ.get("ROPE_BASE", 1e4)) + rope_dims = int(os.environ.get("ROPE_DIMS", 16)) + rope_train_seq_len = int(os.environ.get("ROPE_TRAIN_SEQ_LEN", 2048)) + rope_yarn = bool(int(os.environ.get("ROPE_YARN", "0"))) + ln_scale = bool(int(os.environ.get("LN_SCALE", "1"))) + qk_gain_init = float(os.environ.get("QK_GAIN_INIT", 5.0)) + num_loops = int(os.environ.get("NUM_LOOPS", 2)) + loop_start = int(os.environ.get("LOOP_START", 3)) + loop_end = int(os.environ.get("LOOP_END", 5)) + enable_looping_at = float(os.environ.get("ENABLE_LOOPING_AT", 0.35)) + parallel_start_layer = int(os.environ.get("PARALLEL_START_LAYER", 8)) + parallel_final_lane = os.environ.get("PARALLEL_FINAL_LANE", "mean") + min_lr = float(os.environ.get("MIN_LR", 0.0)) + embed_lr = float(os.environ.get("EMBED_LR", 0.6)) + tied_embed_lr = float(os.environ.get("TIED_EMBED_LR", 0.03)) + tied_embed_init_std = float(os.environ.get("TIED_EMBED_INIT_STD", 0.005)) + matrix_lr = float(os.environ.get("MATRIX_LR", 0.026)) + scalar_lr = float(os.environ.get("SCALAR_LR", 0.02)) + muon_momentum = float(os.environ.get("MUON_MOMENTUM", 0.97)) + muon_backend_steps = int(os.environ.get("MUON_BACKEND_STEPS", 5)) + muon_momentum_warmup_start = float( + os.environ.get("MUON_MOMENTUM_WARMUP_START", 0.92) + ) + muon_momentum_warmup_steps = int(os.environ.get("MUON_MOMENTUM_WARMUP_STEPS", 1500)) + muon_row_normalize = bool(int(os.environ.get("MUON_ROW_NORMALIZE", "1"))) + beta1 = float(os.environ.get("BETA1", 0.9)) + beta2 = float(os.environ.get("BETA2", 0.95)) + adam_eps = float(os.environ.get("ADAM_EPS", 1e-08)) + grad_clip_norm = float(os.environ.get("GRAD_CLIP_NORM", 0.3)) + eval_stride = int(os.environ.get("EVAL_STRIDE", 64)) + adam_wd = float(os.environ.get("ADAM_WD", 0.02)) + muon_wd = float(os.environ.get("MUON_WD", 0.095)) + embed_wd = float(os.environ.get("EMBED_WD", 0.085)) + ema_decay = float(os.environ.get("EMA_DECAY", 0.9965)) + ttt_enabled = bool(int(os.environ.get("TTT_ENABLED", "1"))) + ttt_lora_rank = int(os.environ.get("TTT_LORA_RANK", 96)) + ttt_lora_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)) + # V19: PR #1886 (renqianluo) + sunnypatneedi research log 2026-04-28 found that + # the Triton fused-CE kernel's fp32-accumulation interacts with warm-start LoRA-A + # to destabilize seeds 314/1337 at TTT_WEIGHT_DECAY=1.0. Raising the default to + # 2.0 prevents seed collapse without measurably moving stable seeds. + ttt_weight_decay = float(os.environ.get("TTT_WEIGHT_DECAY", 2.0)) + ttt_beta1 = float(os.environ.get("TTT_BETA1", 0)) + ttt_beta2 = float(os.environ.get("TTT_BETA2", 0.999)) + ttt_k_lora = bool(int(os.environ.get("TTT_K_LORA", "1"))) + ttt_mlp_lora = bool(int(os.environ.get("TTT_MLP_LORA", "1"))) + ttt_o_lora = bool(int(os.environ.get("TTT_O_LORA", "1"))) + ttt_optimizer = os.environ.get("TTT_OPTIMIZER", "adam") + ttt_eval_batches = os.environ.get("TTT_EVAL_BATCHES", "") + val_doc_fraction = float(os.environ.get("VAL_DOC_FRACTION", 1.0)) + compressor = os.environ.get("COMPRESSOR", "brotli") + gptq_calibration_batches = int(os.environ.get("GPTQ_CALIBRATION_BATCHES", 16)) + gptq_reserve_seconds = float(os.environ.get("GPTQ_RESERVE_SECONDS", 4.0)) + phased_ttt_prefix_docs = int(os.environ.get("PHASED_TTT_PREFIX_DOCS", 2000)) + phased_ttt_num_phases = int(os.environ.get("PHASED_TTT_NUM_PHASES", 1)) + global_ttt_lr = float(os.environ.get("GLOBAL_TTT_LR", 0.001)) + global_ttt_momentum = float(os.environ.get("GLOBAL_TTT_MOMENTUM", 0.9)) + global_ttt_epochs = int(os.environ.get("GLOBAL_TTT_EPOCHS", 1)) + global_ttt_chunk_tokens = int(os.environ.get("GLOBAL_TTT_CHUNK_TOKENS", 32768)) + global_ttt_batch_seqs = int(os.environ.get("GLOBAL_TTT_BATCH_SEQS", 32)) + global_ttt_warmup_start_lr = float(os.environ.get("GLOBAL_TTT_WARMUP_START_LR", 0.0)) + global_ttt_warmup_chunks = int(os.environ.get("GLOBAL_TTT_WARMUP_CHUNKS", 0)) + global_ttt_grad_clip = float(os.environ.get("GLOBAL_TTT_GRAD_CLIP", 1.0)) + global_ttt_respect_doc_boundaries = bool(int(os.environ.get("GLOBAL_TTT_RESPECT_DOC_BOUNDARIES", "1"))) + matrix_bits = int(os.environ.get("MATRIX_BITS", 6)) + embed_bits = int(os.environ.get("EMBED_BITS", 8)) + matrix_clip_sigmas = float(os.environ.get("MATRIX_CLIP_SIGMAS", 12.85)) + embed_clip_sigmas = float(os.environ.get("EMBED_CLIP_SIGMAS", 2e1)) + mlp_clip_sigmas = float(os.environ.get("MLP_CLIP_SIGMAS", 10.0)) + attn_clip_sigmas = float(os.environ.get("ATTN_CLIP_SIGMAS", 13.0)) + # AttnOutGate (per-head multiplicative output gate, PR #1667 MarioPaerle). + # Zero-init weight: 2*sigmoid(0)=1 -> transparent at start. Source defaults to + # block input x ('proj'); 'q' uses raw Q projection output. + attn_out_gate_enabled = bool(int(os.environ.get("ATTN_OUT_GATE_ENABLED", "0"))) + attn_out_gate_src = os.environ.get("ATTN_OUT_GATE_SRC", "proj") + # SmearGate (input-dependent forward-1 token smear, modded-nanogpt @classiclarryd + # via PR #1667). x_t <- x_t + lam * sigmoid(W*x_t[:gate_window]) * x_{t-1}. + # lam=0 + W=0 -> transparent at init. + smear_gate_enabled = bool(int(os.environ.get("SMEAR_GATE_ENABLED", "0"))) + # Window: first GATE_WINDOW dims of the source feed the gate projection. + gate_window = int(os.environ.get("GATE_WINDOW", 12)) + # Gated Attention (Qwen, NeurIPS 2025 Best Paper, arXiv:2505.06708; + # qiuzh20/gated_attention). Per-head sigmoid gate on SDPA output, BEFORE + # out_proj. Gate input = full block input x (paper's headwise G1 variant + # driven from hidden_states). W_g shape (num_heads, dim), plain sigmoid. + # Near-zero init gives g~0.5 at step 0 (half attention output); per-block + # attn_scale (init 1.0) compensates during training. Name contains + # "attn_gate" so CONTROL_TENSOR_NAME_PATTERNS routes it to scalar AdamW. + gated_attn_enabled = bool(int(os.environ.get("GATED_ATTN_ENABLED", "0"))) + gated_attn_init_std = float(os.environ.get("GATED_ATTN_INIT_STD", 0.01)) + # Dedicated int8-per-row quantization for `attn_gate_w` tensors. These are + # small ((num_heads, dim) = (8, 512) = 4096 params) and bypass GPTQ via the + # numel<=65536 passthrough branch -> stored as fp16 (8 KB/layer, ~65 KB total + # compressed). int8-per-row cuts the raw tensor in half with negligible BPB + # impact: scales per head (8 values), symmetric quant over [-127, 127]. + # No Hessian needed (gate weights not in collect_hessians()). + gated_attn_quant_gate = bool(int(os.environ.get("GATED_ATTN_QUANT_GATE", "0"))) + # Sparse Attention Gate (modded-nanogpt-style). Keeps dense SDPA and only + # swaps the output-gate input to the first GATE_WINDOW residual dims. + # W_g: (num_heads, gate_window) = (8, 12) = 96 params/layer (~44K total), + # vs dense GatedAttn's (8, 512) = 4K/layer (~44K diff). Name "attn_gate_w" + # is shared so quant routing and int8 gate passthrough Just Work. Gate + # passthrough int8 still applies via GATED_ATTN_QUANT_GATE=1. + # Mutually exclusive with ATTN_OUT_GATE_ENABLED and GATED_ATTN_ENABLED. + sparse_attn_gate_enabled = bool(int(os.environ.get("SPARSE_ATTN_GATE_ENABLED", "0"))) + sparse_attn_gate_init_std = float(os.environ.get("SPARSE_ATTN_GATE_INIT_STD", 0.0)) + sparse_attn_gate_scale = float(os.environ.get("SPARSE_ATTN_GATE_SCALE", 1.0)) + # LQER asymmetric rank-k correction on top-K quant-error tensors (PR #1530 v2 port). + # Computes SVD of E = W_fp - W_quant, packs top-r A,B as INT2/INT4 (asym) or INTk (sym). + lqer_enabled = bool(int(os.environ.get("LQER_ENABLED", "1"))) + lqer_rank = int(os.environ.get("LQER_RANK", 4)) + lqer_top_k = int(os.environ.get("LQER_TOP_K", 3)) + lqer_factor_bits = int(os.environ.get("LQER_FACTOR_BITS", 4)) + lqer_asym_enabled = bool(int(os.environ.get("LQER_ASYM_ENABLED", "1"))) + lqer_asym_group = int(os.environ.get("LQER_ASYM_GROUP", "64")) + lqer_scope = os.environ.get("LQER_SCOPE", "all") + lqer_gain_select = bool(int(os.environ.get("LQER_GAIN_SELECT", "0"))) + awq_lite_enabled = bool(int(os.environ.get("AWQ_LITE_ENABLED", "0"))) + awq_lite_bits = int(os.environ.get("AWQ_LITE_BITS", "8")) + awq_lite_group_top_k = int(os.environ.get("AWQ_LITE_GROUP_TOP_K", "1")) + awq_lite_group_size = int(os.environ.get("AWQ_LITE_GROUP_SIZE", "64")) + # PR #1145 online n-gram tilt (AnirudhRahul, valerio-endorsed). Causal, + # normalized, prefix-only experts; closed-form multiplicative-boost-with-renorm + # applied to per-token NLL. See online_ngram_tilt.py for math + compliance. + ngram_tilt_enabled = bool(int(os.environ.get("NGRAM_TILT_ENABLED", "0"))) + token_order = int(os.environ.get("TOKEN_ORDER", "16")) + token_threshold = float(os.environ.get("TOKEN_THRESHOLD", "0.800")) + token_boost = float(os.environ.get("TOKEN_BOOST", "2.625")) + within_tau = float(os.environ.get("WITHIN_TAU", "0.450")) + within_boost = float(os.environ.get("WITHIN_BOOST", "0.750")) + word_order = int(os.environ.get("WORD_ORDER", "4")) + word_normalize = os.environ.get("WORD_NORMALIZE", "strip_punct_lower") + word_tau = float(os.environ.get("WORD_TAU", "0.650")) + word_boost = float(os.environ.get("WORD_BOOST", "0.750")) + agree_add_boost = float(os.environ.get("AGREE_ADD_BOOST", "0.500")) + # === v5 Stage 1 optimizations (env-gated) === + # 1A: Move ngram hint precompute OUTSIDE eval timer (single causal pass over val tokens). + # Compliance: still inside validate(), single-pass causal, val tokens only. + # Save: ~168s (measured in v2 fulltilt) — enough alone to fit cap. + ngram_hint_precompute_outside = bool(int(os.environ.get("NGRAM_HINT_PRECOMPUTE_OUTSIDE", "1"))) + # 2C: Temperature scaling on logits before softcap. Σ P=1 preserved. + # Default 1.0 = no-op. Tune on train holdout, apply at eval. + temperature_scale = float(os.environ.get("TEMPERATURE_SCALE", "1.0")) + 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 + # CaseOps integration: optional override of dataset root + tokenizer path. + # When CASEOPS_ENABLED=1, the wrapper loads a per-token byte sidecar + # (fineweb_val_bytes_*.bin, identical shard layout to val_*.bin) and uses + # it as the canonical raw-byte budget for BPB accounting. The sidecar + # REPLACES the build_sentencepiece_luts byte-counting path entirely. + caseops_enabled = bool(int(os.environ.get("CASEOPS_ENABLED", "0"))) + _default_caseops_data = os.path.join( + data_dir, + "datasets", + "fineweb10B_sp8192_caseops", + "datasets", + "datasets", + "fineweb10B_sp8192_lossless_caps_caseops_v1_reserved", + ) + _default_caseops_tok = os.path.join( + data_dir, + "datasets", + "fineweb10B_sp8192_caseops", + "datasets", + "tokenizers", + "fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model", + ) + if caseops_enabled: + datasets_dir = os.environ.get("DATA_PATH", _default_caseops_data) + tokenizer_path = os.environ.get("TOKENIZER_PATH", _default_caseops_tok) + else: + datasets_dir = os.environ.get( + "DATA_PATH", + os.path.join(data_dir, "datasets", f"fineweb10B_sp{vocab_size}"), + ) + tokenizer_path = os.environ.get( + "TOKENIZER_PATH", + os.path.join(data_dir, "tokenizers", f"fineweb_{vocab_size}_bpe.model"), + ) + train_files = os.path.join(datasets_dir, "fineweb_train_*.bin") + val_files = os.path.join(datasets_dir, "fineweb_val_*.bin") + val_bytes_files = os.path.join(datasets_dir, "fineweb_val_bytes_*.bin") + artifact_dir = os.environ.get("ARTIFACT_DIR", "") + logfile = ( + os.path.join(artifact_dir, f"{run_id}.txt") + if artifact_dir + else f"logs/{run_id}.txt" + ) + model_path = ( + os.path.join(artifact_dir, "final_model.pt") + if artifact_dir + else "final_model.pt" + ) + quantized_model_path = ( + os.path.join(artifact_dir, "final_model.int6.ptz") + if artifact_dir + else "final_model.int6.ptz" + ) + + +_logger_hparams = None + + +def set_logging_hparams(h): + global _logger_hparams + _logger_hparams = h + + +def log(msg, console=True): + if _logger_hparams is None: + print(msg) + return + if _logger_hparams.is_main_process: + if console: + print(msg) + if _logger_hparams.logfile is not None: + with open(_logger_hparams.logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + + +class ValidationData: + def __init__(self, h, device): + self.sp = spm.SentencePieceProcessor(model_file=h.tokenizer_path) + if int(self.sp.vocab_size()) != h.vocab_size: + raise ValueError( + f"VOCAB_SIZE={h.vocab_size} does not match tokenizer vocab_size={int(self.sp.vocab_size())}" + ) + self.val_tokens = load_validation_tokens(h.val_files, h.eval_seq_len) + self.caseops_enabled = bool(getattr(h, "caseops_enabled", False)) + if self.caseops_enabled: + self.base_bytes_lut = None + self.has_leading_space_lut = None + self.is_boundary_token_lut = None + else: + ( + self.base_bytes_lut, + self.has_leading_space_lut, + self.is_boundary_token_lut, + ) = build_sentencepiece_luts(self.sp, h.vocab_size, device) + self.val_bytes = None + if self.caseops_enabled: + self.val_bytes = load_validation_byte_sidecar( + h.val_bytes_files, h.eval_seq_len, self.val_tokens.numel() + ) + + +def build_sentencepiece_luts(sp, vocab_size, device): + sp_vocab_size = int(sp.vocab_size()) + assert ( + sp.piece_to_id("▁") != sp.unk_id() + ), "Tokenizer must have '▁' (space) as its own token for correct BPB byte counting" + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("▁"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + + +def load_validation_tokens(pattern, seq_len): + # Filter out CaseOps byte sidecar shards which share the val_*.bin glob. + files = [ + Path(p) + for p in sorted(glob.glob(pattern)) + if "_bytes_" not in Path(p).name + ] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = (tokens.numel() - 1) // seq_len * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] + + +def load_validation_byte_sidecar(pattern, seq_len, expected_len): + """Load CaseOps per-token byte sidecar(s). Same shard layout as token shards + (256 int32 header + uint16 array). Each entry = canonical raw-text byte + budget for that token in the corresponding val shard. Returns a CPU + int16 tensor sliced to match expected_len (i.e. val_tokens length).""" + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No byte sidecar files for pattern: {pattern}") + shards = [load_data_shard(file) for file in files] + # load_data_shard returns uint16 — that's exactly what the sidecar stores. + bytes_full = torch.cat(shards).contiguous() + if bytes_full.numel() < expected_len: + raise ValueError( + f"Byte sidecar too short: {bytes_full.numel()} < val_tokens {expected_len}" + ) + return bytes_full[:expected_len].to(torch.int32) + + +def load_data_shard(file): + header_bytes = 256 * np.dtype(" 0: + pos = start + while pos < end: + seg_starts.append(pos) + pos += max_doc_len + else: + seg_starts.append(start) + boundaries = seg_starts + [total_len] + padded_len = get_next_multiple_of_n(len(boundaries), bucket_size) + cu = torch.full((padded_len,), total_len, dtype=torch.int32, device=device) + cu[: len(boundaries)] = torch.tensor(boundaries, dtype=torch.int32, device=device) + seg_ends = seg_starts[1:] + [total_len] + max_seqlen = max(end - start for start, end in zip(seg_starts, seg_ends)) + return cu, max_seqlen + +class DocumentPackingLoader: + _shard_pool = ThreadPoolExecutor(1) + + def __init__(self, h, device, cu_bucket_size=64): + self.rank = h.rank + self.world_size = h.world_size + self.device = device + self.cu_bucket_size = cu_bucket_size + self.max_seq_len = h.train_seq_len + all_files = [Path(p) for p in sorted(glob.glob(h.train_files))] + if not all_files: + raise FileNotFoundError(f"No files found for pattern: {h.train_files}") + self.files = all_files + self.file_iter = iter(self.files) + self._init_shard(load_data_shard(next(self.file_iter))) + self._next_shard = self._submit_next_shard() + self._batch_pool = ThreadPoolExecutor(1) + self._prefetch_queue = [] + + 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() + ) + 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 = torch.empty(per_rank_span - 1, dtype=torch.int64, pin_memory=True) + targets = torch.empty(per_rank_span - 1, dtype=torch.int64, pin_memory=True) + inputs.copy_(buf[:-1]) + targets.copy_(buf[1:]) + 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) + while len(self._prefetch_queue) < 2: + self._prefetch_queue.append( + self._batch_pool.submit(self._prepare_batch, num_tokens_local, self.max_seq_len)) + inputs, targets, cu_seqlens, max_seqlen = self._prefetch_queue.pop(0).result() + self._prefetch_queue.append( + 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 fused_log_softmax_dual_gather_kernel( + logits_ptr, + target_ids_ptr, + hint_ids_ptr, + log_p_y_out_ptr, + log_q_h_out_ptr, + BT, + V, + BLOCK_V: tl.constexpr, +): + """Fused log_softmax + dual gather. Single pass over [BT, V] logits per row, + extracts log p(target_id) and log p(hint_id) via online logsumexp. + Replaces F.log_softmax (which materializes [BT, V] fp32) + 2 gather ops. + """ + pid = tl.program_id(0) + if pid >= BT: + return + + target = tl.load(target_ids_ptr + pid) + hint = tl.load(hint_ids_ptr + pid) + row_offset = pid * V + + target_logit = tl.load(logits_ptr + row_offset + target).to(tl.float32) + hint_logit = tl.load(logits_ptr + row_offset + hint).to(tl.float32) + + NEG_INF = float("-inf") + max_val = NEG_INF + for v_start in tl.range(0, V, BLOCK_V): + v_offsets = v_start + tl.arange(0, BLOCK_V) + mask = v_offsets < V + chunk = tl.load( + logits_ptr + row_offset + v_offsets, mask=mask, other=NEG_INF + ).to(tl.float32) + block_max = tl.max(chunk, axis=0) + max_val = tl.maximum(max_val, block_max) + + sum_exp = tl.zeros((), dtype=tl.float32) + for v_start in tl.range(0, V, BLOCK_V): + v_offsets = v_start + tl.arange(0, BLOCK_V) + mask = v_offsets < V + chunk = tl.load( + logits_ptr + row_offset + v_offsets, mask=mask, other=0.0 + ).to(tl.float32) + chunk_centered = chunk - max_val + exp_chunk = tl.where(mask, tl.exp(chunk_centered), 0.0) + sum_exp += tl.sum(exp_chunk, axis=0) + + log_sum_exp = max_val + tl.log(sum_exp) + log_p_y = target_logit - log_sum_exp + log_p_h = hint_logit - log_sum_exp + + tl.store(log_p_y_out_ptr + pid, log_p_y) + tl.store(log_q_h_out_ptr + pid, log_p_h) + + +def fused_log_softmax_dual_gather(logits, target_ids, hint_ids): + """Triton wrapper — replaces F.log_softmax + 2 gather pattern. + Returns (log_p_y, log_q_h) where p = softmax(logits). + """ + bsz, sl, V = logits.shape + BT = bsz * sl + logits_flat = logits.reshape(BT, V).contiguous() + target_flat = target_ids.reshape(BT).contiguous() + hint_flat = hint_ids.reshape(BT).contiguous() + + log_p_y_out = torch.empty(BT, dtype=torch.float32, device=logits.device) + log_q_h_out = torch.empty(BT, dtype=torch.float32, device=logits.device) + + BLOCK_V = 1024 + grid = (BT,) + fused_log_softmax_dual_gather_kernel[grid]( + logits_flat, + target_flat, + hint_flat, + log_p_y_out, + log_q_h_out, + BT, + V, + BLOCK_V=BLOCK_V, + num_warps=8, + ) + return log_p_y_out.reshape(bsz, sl), log_q_h_out.reshape(bsz, sl) + + +@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, 0.18 * pre0) + c1 = c1 * tl.where(pre1 > 0, 2.0 * pre1, 0.18 * 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, 0.3 * c0) + aux1 = tl.where(c1 > 0, c1, 0.3 * 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 = 256, 128, 64 + num_stages = 4 if forward else 3 + a_desc = TensorDescriptor.from_tensor(a, [BLOCK_SIZE_M, BLOCK_SIZE_K]) + b_desc = TensorDescriptor.from_tensor(b, [BLOCK_SIZE_N, BLOCK_SIZE_K]) + c_desc = TensorDescriptor.from_tensor(c, [BLOCK_SIZE_M, BLOCK_SIZE_N // 2]) + aux_desc = TensorDescriptor.from_tensor(aux, [BLOCK_SIZE_M, BLOCK_SIZE_N // 2]) + grid = lambda _meta: ( + min(num_sms, triton.cdiv(M, BLOCK_SIZE_M) * triton.cdiv(N, BLOCK_SIZE_N)), + ) + linear_leaky_relu_square_kernel[grid]( + a_desc, + b_desc, + c_desc, + aux_desc, + M, + N, + K, + BLOCK_SIZE_M=BLOCK_SIZE_M, + BLOCK_SIZE_N=BLOCK_SIZE_N, + BLOCK_SIZE_K=BLOCK_SIZE_K, + NUM_SMS=num_sms, + FORWARD=forward, + num_stages=num_stages, + num_warps=8, + ) + if forward: + return c, aux + return c + + +class FusedLinearLeakyReLUSquareFunction(torch.autograd.Function): + @staticmethod + def forward(ctx, x, w1, w2): + x_flat = x.reshape(-1, x.shape[-1]) + pre, post = linear_leaky_relu_square(x_flat, w1) + out = F.linear(post, w2) + ctx.save_for_backward(x, w1, w2, pre, post) + return out.view(*x.shape[:-1], out.shape[-1]) + + @staticmethod + def backward(ctx, grad_output): + x, w1, w2, pre, post = ctx.saved_tensors + x_flat = x.reshape(-1, x.shape[-1]) + grad_output_flat = grad_output.reshape(-1, grad_output.shape[-1]) + dw2 = grad_output_flat.T @ post + dpre = linear_leaky_relu_square(grad_output_flat, w2.T.contiguous(), aux=pre) + dw1 = dpre.T @ x_flat + dx = dpre @ w1 + return dx.view_as(x), dw1, dw2 + + +FusedLeakyReLUSquareMLP = FusedLinearLeakyReLUSquareFunction.apply + + +class Rotary(nn.Module): + def __init__(self, dim, base=1e4, train_seq_len=1024, rope_dims=0, yarn=True): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.yarn = yarn + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / base ** ( + torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims + ) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached = None + self._sin_cached = None + + def forward(self, seq_len, device, dtype): + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached < seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if self.yarn and seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * scale ** (rd / (rd - 2)) + inv_freq = 1.0 / new_base ** ( + torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd + ) + else: + inv_freq = self.inv_freq.float().to(device) + t = torch.arange(seq_len, device=device, dtype=torch.float32) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached[:, :seq_len].to(dtype=dtype), self._sin_cached[:, :seq_len].to(dtype=dtype) + + +def apply_rotary_emb(x, cos, sin, rope_dims=0): + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * -sin + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * -sin + x2 * cos), dim=-1) + + +class CausalSelfAttention(nn.Module): + def __init__( + self, dim, num_heads, num_kv_heads, rope_base, qk_gain_init, train_seq_len, yarn=True, + attn_out_gate=False, attn_out_gate_src="proj", gate_window=12, + gated_attn=False, gated_attn_init_std=0.01, + sparse_attn_gate=False, sparse_attn_gate_init_std=0.0, sparse_attn_gate_scale=1.0, + ): + super().__init__() + if dim % num_heads != 0: + raise ValueError("model_dim must be divisible by num_heads") + if num_heads % num_kv_heads != 0: + raise ValueError("num_heads must be divisible by num_kv_heads") + if int(attn_out_gate) + int(gated_attn) + int(sparse_attn_gate) > 1: + raise ValueError( + "attn_out_gate, gated_attn, and sparse_attn_gate are mutually exclusive" + ) + self.num_heads = num_heads + self.num_kv_heads = num_kv_heads + self.head_dim = dim // num_heads + if self.head_dim % 2 != 0: + raise ValueError("head_dim must be even for RoPE") + self.q_gain = nn.Parameter( + torch.full((num_heads,), qk_gain_init, dtype=torch.float32) + ) + self.rope_dims = 0 + self.rotary = Rotary(self.head_dim, base=rope_base, train_seq_len=train_seq_len, yarn=yarn) + self.use_xsa = False + # AttnOutGate (PR #1667 MarioPaerle): per-head multiplicative gate on attention + # output. CastedLinear so restore_fp32_params casts back to fp32 for GPTQ. + # _zero_init -> 2*sigmoid(0)=1 -> transparent at init. + self.attn_out_gate = attn_out_gate + self.attn_out_gate_src = attn_out_gate_src + self.gate_window = gate_window + if attn_out_gate: + self.attn_gate_proj = CastedLinear(gate_window, num_heads, bias=False) + self.attn_gate_proj._zero_init = True + # Gated Attention (arXiv:2505.06708, Qwen, NeurIPS 2025). Per-head sigmoid + # gate on SDPA output, BEFORE out_proj. Gate projection W_g: (num_heads, dim). + # Name "attn_gate_w" contains "attn_gate" substring so it matches + # CONTROL_TENSOR_NAME_PATTERNS and routes to the scalar AdamW group. + # fp32 Parameter -> restore_fp32_params path covers it via the ndim<2 OR + # name-pattern check (name matches "attn_gate"). Cast to x.dtype on use. + self.gated_attn = gated_attn + if gated_attn: + W = torch.empty(num_heads, dim, dtype=torch.float32) + nn.init.normal_(W, mean=0.0, std=gated_attn_init_std) + self.attn_gate_w = nn.Parameter(W) + # Sparse attention head-output gate (modded-nanogpt style). Keeps dense SDPA + # and only narrows the gate input to the first gate_window residual dims. + # W_g: (num_heads, gate_window). y_{t,h} <- sigmoid(scale * W_g_h @ x_t[:gate_window]) * y_{t,h}. + # Shares attn_gate_w name with dense GatedAttn so the quant routing + # (CONTROL_TENSOR_NAME_PATTERNS / attn_gate_w int8 passthrough) is unchanged. + self.sparse_attn_gate = sparse_attn_gate + self.sparse_attn_gate_scale = sparse_attn_gate_scale + if sparse_attn_gate: + W = torch.empty(num_heads, gate_window, dtype=torch.float32) + if sparse_attn_gate_init_std > 0: + nn.init.normal_(W, mean=0.0, std=sparse_attn_gate_init_std) + else: + nn.init.zeros_(W) + self.attn_gate_w = nn.Parameter(W) + + def _xsa_efficient(self, y, v): + B, T, H, D = y.shape + Hkv = v.size(-2) + group = H // Hkv + y_g = y.reshape(B, T, Hkv, group, D) + vn = F.normalize(v, dim=-1).unsqueeze(-2) + proj = (y_g * vn).sum(dim=-1, keepdim=True) * vn + return (y_g - proj).reshape(B, T, H, D) + + def forward(self, x, q_w, k_w, v_w, out_w, cu_seqlens=None, max_seqlen=0): + bsz, seqlen, dim = x.shape + # q_raw kept around as a tap point for attn_out_gate_src='q' (post-projection, + # pre-reshape, pre-RoPE). + q_raw = F.linear(x, q_w.to(x.dtype)) + q = q_raw.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) + # AttnOutGate inlined (PR #1667). Inline + .contiguous() barrier so torch.compile + # fullgraph=True is happy (this avoids the @torch.compiler.disable trap that + # crashed gates v3). Per-head gate on (B,T,H,D) tensor: g shape [B,T,H], broadcast + # over D via [..., None]. zero-init weight -> 2*sigmoid(0)=1 -> transparent. + if self.attn_out_gate: + gate_src = q_raw if self.attn_out_gate_src == "q" else x + gate_in = gate_src[..., : self.gate_window].contiguous() + g = 2.0 * torch.sigmoid(self.attn_gate_proj(gate_in)) + y = y * g[..., None] + # Gated Attention (arXiv:2505.06708 G1). Inline + .contiguous() barrier so + # torch.compile fullgraph=True is happy. Per-head gate on (B,T,H,D): g shape + # [B,T,H], broadcast over D via [..., None]. Paper: g = sigmoid(x @ W_g.T) + # where W_g: (H, dim). .to(x.dtype) on fp32 param before broadcast with bf16. + if self.gated_attn: + x_c = x.contiguous() + g = torch.sigmoid(F.linear(x_c, self.attn_gate_w.to(x.dtype))) + y = y * g[..., None] + # Sparse head-output gate: narrower (gate_window) input, same shape g as GatedAttn. + if self.sparse_attn_gate: + gate_in = x[..., : self.gate_window].contiguous() + g = torch.sigmoid( + self.sparse_attn_gate_scale + * F.linear(gate_in, self.attn_gate_w.to(x.dtype)) + ) + y = y * g[..., None] + y = y.reshape(bsz, seqlen, dim) + self._last_proj_input = y.detach() if getattr(self, "_calib", False) else None + 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=0.3).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, + attn_out_gate=False, + attn_out_gate_src="proj", + gate_window=12, + gated_attn=False, + gated_attn_init_std=0.01, + sparse_attn_gate=False, + sparse_attn_gate_init_std=0.0, + sparse_attn_gate_scale=1.0, + ): + super().__init__() + self.attn_norm = RMSNorm() + self.mlp_norm = RMSNorm() + self.attn = CausalSelfAttention( + dim, num_heads, num_kv_heads, rope_base, qk_gain_init, train_seq_len, yarn=yarn, + attn_out_gate=attn_out_gate, attn_out_gate_src=attn_out_gate_src, gate_window=gate_window, + gated_attn=gated_attn, gated_attn_init_std=gated_attn_init_std, + sparse_attn_gate=sparse_attn_gate, + sparse_attn_gate_init_std=sparse_attn_gate_init_std, + sparse_attn_gate_scale=sparse_attn_gate_scale, + ) + self.mlp = MLP(dim, mlp_mult) + self.attn_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.mlp_scale = nn.Parameter(torch.ones(dim, dtype=torch.float32)) + self.resid_mix = nn.Parameter( + torch.stack((torch.ones(dim), torch.zeros(dim))).float() + ) + self.ln_scale_factor = 1.0 / math.sqrt(layer_idx + 1) if ln_scale else 1.0 + + def forward(self, x, x0, q_w, k_w, v_w, out_w, up_w, down_w, cu_seqlens=None, max_seqlen=0): + mix = self.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + attn_out = self.attn( + self.attn_norm(x_in) * self.ln_scale_factor, + q_w, k_w, v_w, out_w, + cu_seqlens=cu_seqlens, + max_seqlen=max_seqlen, + ) + x_out = x_in + self.attn_scale.to(dtype=x_in.dtype)[None, None, :] * attn_out + x_out = x_out + self.mlp_scale.to(dtype=x_out.dtype)[ + None, None, : + ] * self.mlp(self.mlp_norm(x_out) * self.ln_scale_factor, up_w, down_w) + return x_out + +class GPT(nn.Module): + def __init__(self, h): + super().__init__() + if h.logit_softcap <= 0.0: + raise ValueError(f"logit_softcap must be positive, got {h.logit_softcap}") + self.tie_embeddings = h.tie_embeddings + self.tied_embed_init_std = h.tied_embed_init_std + self.logit_softcap = h.logit_softcap + self.fused_ce_enabled = bool(h.fused_ce_enabled) + self.tok_emb = nn.Embedding(h.vocab_size, h.model_dim) + self.num_layers = h.num_layers + head_dim = h.model_dim // h.num_heads + kv_dim = h.num_kv_heads * head_dim + hidden_dim = int(h.mlp_mult * h.model_dim) + self.qo_bank = nn.Parameter(torch.empty(2 * h.num_layers, h.model_dim, h.model_dim)) + self.kv_bank = nn.Parameter(torch.empty(2 * h.num_layers, kv_dim, h.model_dim)) + self.mlp_up_bank = nn.Parameter(torch.empty(h.num_layers, hidden_dim, h.model_dim)) + self.mlp_down_bank = nn.Parameter(torch.empty(h.num_layers, h.model_dim, hidden_dim)) + self.num_encoder_layers = h.num_layers // 2 + self.num_decoder_layers = h.num_layers - self.num_encoder_layers + self.blocks = nn.ModuleList( + [ + Block( + h.model_dim, + h.num_heads, + h.num_kv_heads, + h.mlp_mult, + h.rope_base, + h.qk_gain_init, + h.train_seq_len, + layer_idx=i, + ln_scale=h.ln_scale, + yarn=h.rope_yarn, + attn_out_gate=h.attn_out_gate_enabled, + attn_out_gate_src=h.attn_out_gate_src, + gate_window=h.gate_window, + gated_attn=h.gated_attn_enabled, + gated_attn_init_std=h.gated_attn_init_std, + sparse_attn_gate=h.sparse_attn_gate_enabled, + sparse_attn_gate_init_std=h.sparse_attn_gate_init_std, + sparse_attn_gate_scale=h.sparse_attn_gate_scale, + ) + for i in range(h.num_layers) + ] + ) + if h.rope_dims > 0: + head_dim = h.model_dim // h.num_heads + for block in self.blocks: + block.attn.rope_dims = h.rope_dims + block.attn.rotary = Rotary( + head_dim, + base=h.rope_base, + train_seq_len=h.train_seq_len, + rope_dims=h.rope_dims, + yarn=h.rope_yarn, + ) + self.final_norm = RMSNorm() + self.lm_head = ( + None + if h.tie_embeddings + else CastedLinear(h.model_dim, h.vocab_size, bias=False) + ) + if self.lm_head is not None: + self.lm_head._zero_init = True + if h.xsa_last_n > 0: + for i in range(max(0, h.num_layers - h.xsa_last_n), h.num_layers): + self.blocks[i].attn.use_xsa = True + self.looping_active = False + 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) + ) + # SmearGate (PR #1667 / modded-nanogpt @classiclarryd): + # x_t <- x_t + lam * sigmoid(W * x_t[:gate_window]) * x_{t-1}. + # Per-token forward-1 smear of the embedding lane. W zero-init + lam=0 -> + # transparent at init. Uses CastedLinear so restore_fp32_params handles dtype. + self.smear_gate_enabled = h.smear_gate_enabled + if self.smear_gate_enabled: + self.smear_window = h.gate_window + self.smear_gate = CastedLinear(self.smear_window, 1, bias=False) + self.smear_gate._zero_init = True + self.smear_lambda = nn.Parameter(torch.zeros(1, dtype=torch.float32)) + # V19: Asymmetric Logit Rescale (PR #1923 jorge-asenjo). + # Two learnable softcap scales applied on the EVAL path (forward_logits + + # forward_ttt). Init to logit_softcap so the layer is identity at step 0. + # Train path keeps the single fused softcap to preserve PR #1855 numerics. + self.asym_logit_enabled = bool(int(os.environ.get("ASYM_LOGIT_RESCALE", "0"))) + if self.asym_logit_enabled: + self.softcap_pos = nn.Parameter(torch.tensor(float(h.logit_softcap), dtype=torch.float32)) + self.softcap_neg = nn.Parameter(torch.tensor(float(h.logit_softcap), dtype=torch.float32)) + # v5 Stage 2C: temperature scaling on logits before softcap (eval-only TTT path). + self.temperature_scale = float(getattr(h, "temperature_scale", 1.0)) + self._init_weights() + + def _init_weights(self): + if self.tie_embeddings: + nn.init.normal_(self.tok_emb.weight, mean=0.0, std=self.tied_embed_init_std) + n = self.num_layers + proj_scale = 1.0 / math.sqrt(2 * n) + for i in range(n): + nn.init.orthogonal_(self.qo_bank.data[i], gain=1.0) + nn.init.zeros_(self.qo_bank.data[n + i]) + self.qo_bank.data[n + i].mul_(proj_scale) + nn.init.orthogonal_(self.kv_bank.data[i], gain=1.0) + nn.init.orthogonal_(self.kv_bank.data[n + i], gain=1.0) + for i in range(n): + nn.init.orthogonal_(self.mlp_up_bank.data[i], gain=1.0) + nn.init.zeros_(self.mlp_down_bank.data[i]) + self.mlp_down_bank.data[i].mul_(proj_scale) + for name, module in self.named_modules(): + if isinstance(module, nn.Linear): + if getattr(module, "_zero_init", False): + nn.init.zeros_(module.weight) + elif ( + module.weight.ndim == 2 + and module.weight.shape[0] >= 64 + and module.weight.shape[1] >= 64 + ): + nn.init.orthogonal_(module.weight, gain=1.0) + + def _bank_weights(self, i): + n = self.num_layers + return ( + self.qo_bank[i], + self.kv_bank[i], + self.kv_bank[n + i], + self.qo_bank[n + i], + self.mlp_up_bank[i], + self.mlp_down_bank[i], + ) + + def _parallel_block( + self, block_idx, lane0, lane1, x0, + q_w, k_w, v_w, out_w, up_w, down_w, + cu_seqlens=None, max_seqlen=0, + ): + block = self.blocks[block_idx] + mix = block.resid_mix.to(dtype=lane0.dtype) + attn_read = mix[0][None, None, :] * lane0 + mix[1][None, None, :] * x0 + attn_out = block.attn( + block.attn_norm(attn_read) * block.ln_scale_factor, + q_w, k_w, v_w, out_w, + cu_seqlens=cu_seqlens, max_seqlen=max_seqlen, + ) + attn_out = block.attn_scale.to(dtype=attn_out.dtype)[None, None, :] * attn_out + mlp_read = lane1 + mlp_out = block.mlp_scale.to(dtype=lane1.dtype)[None, None, :] * block.mlp( + block.mlp_norm(mlp_read) * block.ln_scale_factor, up_w, down_w + ) + attn_resid = self.parallel_resid_lambdas[block_idx, 0].to(dtype=lane0.dtype) + attn_post = self.parallel_post_lambdas[block_idx, 0].to(dtype=lane0.dtype) + mlp_resid = self.parallel_resid_lambdas[block_idx, 1].to(dtype=lane0.dtype) + mlp_post = self.parallel_post_lambdas[block_idx, 1].to(dtype=lane0.dtype) + lane0 = attn_resid * lane0 + attn_post[0] * attn_out + mlp_post[0] * mlp_out + lane1 = mlp_resid * lane1 + attn_post[1] * attn_out + mlp_post[1] * mlp_out + return lane0, lane1 + + def _final_parallel_hidden(self, lane0, lane1): + if self.parallel_final_lane == "mlp": + return lane1 + if self.parallel_final_lane == "attn": + return lane0 + return 0.5 * (lane0 + lane1) + + def _forward_hidden(self, input_ids, cu_seqlens=None, max_seqlen=0): + """Run the encoder/decoder stack to the final RMSNorm; returns pre-projection hidden. + Shared by eval (softcap+projection via forward_logits) and train (fused CE path).""" + x = self.tok_emb(input_ids) + # SmearGate (PR #1667). lam=0 + W=0 -> identity at init. + # Cross-doc leak fix: zero the prev-token smear at any position whose current token + # is BOS, so the BOS embedding starting doc N+1 in a packed stream is not + # contaminated by doc N's last token (audited issue on PR#1797 base). + if self.smear_gate_enabled: + sl = self.smear_lambda.to(dtype=x.dtype) + gate_in = x[:, 1:, : self.smear_window].contiguous() + g = sl * torch.sigmoid(self.smear_gate(gate_in)) + not_bos = (input_ids[:, 1:] != BOS_ID).to(x.dtype).unsqueeze(-1) + x = torch.cat([x[:, :1], x[:, 1:] + g * x[:, :-1] * not_bos], dim=1) + x = F.rms_norm(x, (x.size(-1),)) + 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) + return x + + def _project_logits(self, hidden): + if self.tie_embeddings: + return F.linear(hidden, self.tok_emb.weight) + return self.lm_head(hidden) + + def _apply_asym_softcap(self, logits): + # V19: Asymmetric softcap (PR #1923). Splits the logit_softcap scalar into + # learnable positive/negative branches. Score-first preserved: still a + # bounded, normalized post-projection nonlinearity feeding a standard + # softmax over the full vocab. + sp = self.softcap_pos.to(logits.dtype) + sn = self.softcap_neg.to(logits.dtype) + return torch.where(logits > 0, sp * torch.tanh(logits / sp), sn * torch.tanh(logits / sn)) + + def forward_logits(self, input_ids, cu_seqlens=None, max_seqlen=0): + hidden = self._forward_hidden(input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + logits_proj = self._project_logits(hidden) + if self.asym_logit_enabled: + return self._apply_asym_softcap(logits_proj) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward(self, input_ids, target_ids, cu_seqlens=None, max_seqlen=0): + hidden = self._forward_hidden(input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + logits_proj = self._project_logits(hidden) + flat_targets = target_ids.reshape(-1) + # Fused softcapped-CE kernel (training path only). Applies softcap inside the + # Triton kernel; takes pre-softcap logits_proj. Non-fused path matches stock + # PR-1736 numerics exactly (softcap in fp32, then F.cross_entropy on fp32). + if self.fused_ce_enabled: + return softcapped_cross_entropy( + logits_proj.reshape(-1, logits_proj.size(-1)), + flat_targets, + self.logit_softcap, + reduction="mean", + ) + logits = self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + return F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + flat_targets, + reduction="mean", + ) + + def forward_ttt(self, input_ids, target_ids, lora, hint_ids=None): + x = self.tok_emb(input_ids) + # SmearGate on the TTT path — same inline compute as forward_logits. + # Cross-doc leak fix: see _forward_hidden comment. + if self.smear_gate_enabled: + sl = self.smear_lambda.to(dtype=x.dtype) + gate_in = x[:, 1:, : self.smear_window].contiguous() + g = sl * torch.sigmoid(self.smear_gate(gate_in)) + not_bos = (input_ids[:, 1:] != BOS_ID).to(x.dtype).unsqueeze(-1) + x = torch.cat([x[:, :1], x[:, 1:] + g * x[:, :-1] * not_bos], dim=1) + x = F.rms_norm(x, (x.size(-1),)) + x0 = x + skips = [] + enc_iter = ( + self.encoder_indices + if self.looping_active + else list(range(self.num_encoder_layers)) + ) + dec_iter = ( + self.decoder_indices + if self.looping_active + else list( + range( + self.num_encoder_layers, + self.num_encoder_layers + self.num_decoder_layers, + ) + ) + ) + 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.tie_embeddings: + logits = F.linear(x, self.tok_emb.weight) + else: + logits = self.lm_head(x) + logits = logits + lora.lm_head_lora(x) + # v5 Stage 2C: temperature scaling. T=1.0 (default) -> no-op. + # Applied BEFORE softcap so cap acts on calibrated logits. + if getattr(self, "temperature_scale", 1.0) != 1.0: + logits = logits / self.temperature_scale + # V19: same asymmetric softcap on the TTT eval path. + if self.asym_logit_enabled: + logits = self._apply_asym_softcap(logits) + else: + logits = self.logit_softcap * torch.tanh(logits / self.logit_softcap) + bsz, sl, V = logits.shape + if hint_ids is None: + return F.cross_entropy( + logits.float().reshape(-1, V), target_ids.reshape(-1), reduction="none" + ).reshape(bsz, sl) + # PR #1145 tilt branch (v4): Triton fused kernel for eval scoring (no_grad). + # TTT learning path needs autograd, so fall back to vanilla F.log_softmax + # when logits require grad. Triton kernel is forward-only (no backward). + if logits.requires_grad: + ls = F.log_softmax(logits.float(), dim=-1) + log_p_y = ls.gather(-1, target_ids.unsqueeze(-1)).squeeze(-1) + log_q_h = ls.gather(-1, hint_ids.clamp(min=0).unsqueeze(-1)).squeeze(-1) + return -log_p_y, log_q_h + log_p_y, log_q_h = fused_log_softmax_dual_gather( + logits, target_ids, hint_ids.clamp(min=0) + ) + return -log_p_y, log_q_h + + 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 + # Keep raw Q for AttnOutGate src='q' (matches forward path semantics). + q_raw = F.linear(n, q_w.to(n.dtype)) + lora.q_loras[slot](n) + q = q_raw.reshape(bsz, seqlen, attn.num_heads, attn.head_dim) + k = F.linear(n, 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) + # AttnOutGate (TTT path) — inline + .contiguous() barrier, same as the eval path. + if attn.attn_out_gate: + gate_src = q_raw if attn.attn_out_gate_src == "q" else n + gate_in = gate_src[..., : attn.gate_window].contiguous() + g = 2.0 * torch.sigmoid(attn.attn_gate_proj(gate_in)) + y = y * g[..., None] + # Gated Attention (TTT path). Gate input is n (post-norm block input), same + # as eval path. .to(n.dtype) on fp32 param before bf16 broadcast. + if attn.gated_attn: + n_c = n.contiguous() + g = torch.sigmoid(F.linear(n_c, attn.attn_gate_w.to(n.dtype))) + y = y * g[..., None] + # Sparse attention head-output gate (TTT path) — must match the eval path in + # forward() exactly, else training (which applied the gate) and TTT eval (which + # skipped it) produce mismatched representations and catastrophic BPB regression. + if attn.sparse_attn_gate: + gate_in = n[..., : attn.gate_window].contiguous() + g = torch.sigmoid( + attn.sparse_attn_gate_scale + * F.linear(gate_in, attn.attn_gate_w.to(n.dtype)) + ) + y = y * g[..., None] + y = y.reshape(bsz, seqlen, dim) + 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_raw = F.linear(n, q_w.to(n.dtype)) + lora.q_loras[slot](n) + q = q_raw.reshape(bsz, seqlen, attn.num_heads, attn.head_dim) + k = F.linear(n, 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) + # AttnOutGate (TTT parallel path) — inline + .contiguous() barrier. + if attn.attn_out_gate: + gate_src = q_raw if attn.attn_out_gate_src == "q" else n + gate_in = gate_src[..., : attn.gate_window].contiguous() + g = 2.0 * torch.sigmoid(attn.attn_gate_proj(gate_in)) + y = y * g[..., None] + # Gated Attention (TTT parallel path). Gate input is n (post-norm block input). + if attn.gated_attn: + n_c = n.contiguous() + g = torch.sigmoid(F.linear(n_c, attn.attn_gate_w.to(n.dtype))) + y = y * g[..., None] + # Sparse attention head-output gate (TTT parallel path) — must match the + # eval path in forward() to keep train/eval semantics in sync. + if attn.sparse_attn_gate: + gate_in = n[..., : attn.gate_window].contiguous() + g = torch.sigmoid( + attn.sparse_attn_gate_scale + * F.linear(gate_in, attn.attn_gate_w.to(n.dtype)) + ) + y = y * g[..., None] + y = y.reshape(bsz, seqlen, dim) + 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): + # PR-1767: rank-scaled output (alpha/rank), like standard LoRA. Decouples + # effective magnitude from rank so changing rank does not change LR scale. + _ALPHA = float(os.environ.get("TTT_LORA_ALPHA", "144")) + # PR-1767: optionally keep A warm across per-doc resets (only B is zeroed). + # Accumulates useful feature directions across documents within a TTT phase. + _WARM_START_A = bool(int(os.environ.get("TTT_WARM_START_A", "1"))) + + def __init__(self, bsz, in_features, out_features, rank): + super().__init__() + self._bound = 1.0 / math.sqrt(in_features) + self._scale = self._ALPHA / rank + self.A = nn.Parameter( + torch.empty(bsz, rank, in_features).uniform_(-self._bound, self._bound) + ) + self.B = nn.Parameter(torch.zeros(bsz, out_features, rank)) + + def reset(self): + with torch.no_grad(): + if not self._WARM_START_A: + 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)) * self._scale + + +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() + + +# Polar Express per-iteration minimax Newton-Schulz coefficients (PR #1344). +# Replaces the fixed (3.4445, -4.775, 2.0315) coefficients of stock Muon. +# Applied at backend_steps=5 — taking more than 5 iterations from this list +# falls back to the final (converged) tuple via the slice guard below. +_PE_COEFFS = ( + (8.156554524902461, -22.48329292557795, 15.878769915207462), + (4.042929935166739, -2.808917465908714, 0.5000178451051316), + (3.8916678022926607, -2.772484153217685, 0.5060648178503393), + (3.285753657755655, -2.3681294933425376, 0.46449024233003106), + (2.3465413258596377, -1.7097828382687081, 0.42323551169305323), +) + + +@torch.compile +def zeropower_via_newtonschulz5(G, steps=10, eps=1e-07): + was_2d = G.ndim == 2 + if was_2d: + G = G.unsqueeze(0) + X = G.bfloat16() + transposed = X.size(-2) > X.size(-1) + if transposed: + X = X.mT + X = X / (X.norm(dim=(-2, -1), keepdim=True) + eps) + coeffs = _PE_COEFFS[:steps] if steps <= len(_PE_COEFFS) else _PE_COEFFS + for a, b, c in coeffs: + A = X @ X.mT + B = b * A + c * (A @ A) + X = a * X + B @ X + if transposed: + X = X.mT + if was_2d: + X = X.squeeze(0) + return X + + +class Muon(torch.optim.Optimizer): + def __init__( + self, + params, + lr, + momentum, + backend_steps, + nesterov=True, + weight_decay=0.0, + row_normalize=False, + ): + super().__init__( + params, + dict( + lr=lr, + momentum=momentum, + backend_steps=backend_steps, + nesterov=nesterov, + weight_decay=weight_decay, + row_normalize=row_normalize, + ), + ) + self._built = False + + def _build(self): + self._distributed = dist.is_available() and dist.is_initialized() + self._world_size = dist.get_world_size() if self._distributed else 1 + self._rank = dist.get_rank() if self._distributed else 0 + ws = self._world_size + self._bank_meta = [] + for group in self.param_groups: + for p in group["params"]: + B = p.shape[0] + padded_B = ((B + ws - 1) // ws) * ws + shard_B = padded_B // ws + tail = p.shape[1:] + dev = p.device + self._bank_meta.append({ + "p": p, + "B": B, + "padded_grad": torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + "shard": torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + "shard_mom": torch.zeros(shard_B, *tail, device=dev, dtype=torch.bfloat16), + "full_update": torch.zeros(padded_B, *tail, device=dev, dtype=torch.bfloat16), + "scale": max(1, p.shape[-2] / p.shape[-1]) ** 0.5, + }) + self._bank_meta.sort(key=lambda m: -m["p"].numel()) + self._built = True + + def launch_reduce_scatters(self): + if not self._built: + self._build() + if not self._distributed: + return + self._rs_futures = [] + for m in self._bank_meta: + p = m["p"] + if p.grad is None: + self._rs_futures.append(None) + continue + pg = m["padded_grad"] + pg[: m["B"]].copy_(p.grad) + 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, 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, 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, alpha=-lr * prev_m["scale"]) + if hasattr(self, "_rs_futures"): + del self._rs_futures + return loss + + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,skip_gates,parallel_post_lambdas,parallel_resid_lambdas,attn_gate_proj,attn_gate_w,smear_gate,smear_lambda", + ).split(",") + if pattern +) + + +PACKED_REPLICATED_GRAD_MAX_NUMEL = 1 << 15 + + +class Optimizers: + def __init__(self, h, base_model): + matrix_params = [ + base_model.qo_bank, + base_model.kv_bank, + base_model.mlp_up_bank, + base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for (name, p) in block_named_params + if p.ndim < 2 + or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + if base_model.skip_gates is not None and base_model.skip_gates.numel() > 0: + scalar_params.append(base_model.skip_gates) + if base_model.parallel_post_lambdas is not None: + scalar_params.append(base_model.parallel_post_lambdas) + if base_model.parallel_resid_lambdas is not None: + scalar_params.append(base_model.parallel_resid_lambdas) + # SmearGate params live on GPT root (not in .blocks), so add them by hand. + # Both are tiny (gate_window scalars + 1 lambda). Optimized via scalar Adam. + if getattr(base_model, "smear_gate_enabled", False): + scalar_params.append(base_model.smear_gate.weight) + scalar_params.append(base_model.smear_lambda) + token_lr = h.tied_embed_lr if h.tie_embeddings else h.embed_lr + tok_params = [ + {"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr} + ] + self.optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(h.beta1, h.beta2), + eps=h.adam_eps, + weight_decay=h.embed_wd, + fused=True, + ) + self.optimizer_muon = Muon( + matrix_params, + lr=h.matrix_lr, + momentum=h.muon_momentum, + backend_steps=h.muon_backend_steps, + weight_decay=h.muon_wd, + row_normalize=h.muon_row_normalize, + ) + for group in self.optimizer_muon.param_groups: + group["base_lr"] = h.matrix_lr + self.optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": h.scalar_lr, "base_lr": h.scalar_lr}], + betas=(h.beta1, h.beta2), + eps=h.adam_eps, + weight_decay=h.adam_wd, + fused=True, + ) + self.optimizers = [ + self.optimizer_tok, + self.optimizer_muon, + self.optimizer_scalar, + ] + self.replicated_params = list(tok_params[0]["params"]) + self.replicated_params.extend(scalar_params) + self.replicated_large_params = [] + self.replicated_packed_params = [] + for p in self.replicated_params: + if p.numel() <= PACKED_REPLICATED_GRAD_MAX_NUMEL: + self.replicated_packed_params.append(p) + else: + self.replicated_large_params.append(p) + self._aux_stream = torch.cuda.Stream() + + 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._aux_stream.wait_stream(torch.cuda.current_stream()) + with torch.cuda.stream(self._aux_stream): + self.optimizer_tok.step() + self.optimizer_scalar.step() + self.optimizer_muon.step() + torch.cuda.current_stream().wait_stream(self._aux_stream) + self.zero_grad_all() + + +def restore_fp32_params(model): + for module in model.modules(): + if isinstance(module, CastedLinear): + module.float() + for name, param in model.named_parameters(): + if ( + param.ndim < 2 + or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ) and param.dtype != torch.float32: + param.data = param.data.float() + if hasattr(model, "qo_bank") and model.qo_bank is not None: + model.qo_bank.data = model.qo_bank.data.float() + model.kv_bank.data = model.kv_bank.data.float() + model.mlp_up_bank.data = model.mlp_up_bank.data.float() + model.mlp_down_bank.data = model.mlp_down_bank.data.float() + + +def collect_hessians(model, train_loader, h, device, n_calibration_batches=64): + hessians = {} + act_sumsq = {} + act_counts = {} + 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]) + x_sq = x.square().sum(dim=0) + x_count = x.shape[0] + 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) + if name not in act_sumsq: + act_sumsq[name] = torch.zeros( + x.shape[1], dtype=torch.float32, device=device + ) + act_counts[name] = 0 + act_sumsq[name] += x_sq + act_counts[name] += x_count + 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) + if name not in act_sumsq: + act_sumsq[name] = torch.zeros( + y.shape[1], dtype=torch.float32, device=device + ) + act_counts[name] = 0 + act_sumsq[name] += y.square().sum(dim=0) + act_counts[name] += y.shape[0] + 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) + if name not in act_sumsq: + act_sumsq[name] = torch.zeros( + x.shape[1], dtype=torch.float32, device=device + ) + act_counts[name] = 0 + act_sumsq[name] += x.square().sum(dim=0) + act_counts[name] += x.shape[0] + 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) + if name not in act_sumsq: + act_sumsq[name] = torch.zeros( + h_act.shape[1], dtype=torch.float32, device=device + ) + act_counts[name] = 0 + act_sumsq[name] += h_act.square().sum(dim=0) + act_counts[name] += h_act.shape[0] + return hook_fn + + for i, block in enumerate(model.blocks): + hooks.append(block.attn.register_forward_hook(make_attn_hook(i))) + hooks.append(block.mlp.register_forward_hook(make_mlp_hook(i))) + + # Hessian hooks for embedding factorization projection layers + def make_linear_input_hook(weight_name): + def hook_fn(module, inp, out): + x = inp[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + if weight_name not in hessians: + hessians[weight_name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[weight_name].addmm_(x.T, x) + return hook_fn + + if model.tie_embeddings: + hook_module = model.final_norm + + def make_output_hook(name): + def hook_fn(module, inp, out): + x = out.detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + if name not in hessians: + hessians[name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(x.T, x) + if name not in act_sumsq: + act_sumsq[name] = torch.zeros( + x.shape[1], dtype=torch.float32, device=device + ) + act_counts[name] = 0 + act_sumsq[name] += x.square().sum(dim=0) + act_counts[name] += x.shape[0] + 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 + act_stats = {} + for name, sumsq in act_sumsq.items(): + count = max(act_counts.get(name, 0), 1) + act_stats[name] = (sumsq / count).sqrt().cpu() + return hessians, act_stats + + +def gptq_quantize_weight( + w, + H, + clip_sigmas=3.0, + clip_range=63, + block_size=128, + protect_groups=None, + group_size=None, + protect_clip_range=None, +): + 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() + protect_meta = None + protect_mask_perm = None + s_hi = None + sf_hi = None + if ( + protect_groups + and group_size is not None + and protect_clip_range is not None + and protect_clip_range > clip_range + ): + protect_mask = torch.zeros(cols, dtype=torch.bool) + starts = [] + for (start, end) in protect_groups: + if start < 0 or end > cols or end <= start: + continue + protect_mask[start:end] = True + starts.append(start) + if starts: + protect_mask_perm = protect_mask[perm] + s_hi = (clip_sigmas * row_std / protect_clip_range).clamp_min(1e-10).to( + torch.float16 + ) + sf_hi = s_hi.float() + protect_meta = { + "starts": torch.tensor(starts, dtype=torch.int16), + "size": int(group_size), + "s_hi": s_hi, + } + 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] + if protect_mask_perm is not None and bool(protect_mask_perm[i1 + j]): + q_col = torch.clamp( + torch.round(w_col / sf_hi), + -protect_clip_range, + protect_clip_range, + ) + w_recon = q_col.float() * sf_hi + else: + q_col = torch.clamp(torch.round(w_col / sf), -clip_range, clip_range) + w_recon = q_col.float() * sf + Q[:, i1 + j] = q_col.to(torch.int8) + err = (w_col - w_recon) / 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, protect_meta + + +def _quantize_gate_int8_row(w): + # Symmetric int8-per-row quantization for small gate tensors. w shape + # (R, C) -> (R,) scales in fp16, int8 values in [-127, 127]. Single scale + # per row keeps accuracy high while halving storage vs fp16. + W = w.float().contiguous() + row_max = W.abs().amax(dim=1).clamp_min(1e-10) + s = (row_max / 127.0).to(torch.float16) + sf = s.float().view(-1, 1) + q = torch.clamp(torch.round(W / sf), -127, 127).to(torch.int8) + return q, s + + +def _lqer_pack(A, B, bits): + rng = 2 ** (bits - 1) - 1 + sA = (A.abs().amax(dim=1).clamp_min(1e-10) / rng).to(torch.float16) + sB = (B.abs().amax(dim=1).clamp_min(1e-10) / rng).to(torch.float16) + qA = torch.clamp(torch.round(A / sA.float().view(-1, 1)), -rng, rng).to(torch.int8) + qB = torch.clamp(torch.round(B / sB.float().view(-1, 1)), -rng, rng).to(torch.int8) + return qA, sA, qB, sB + + +def _lqer_pack_asym(A, B, g=64): + # A: INT2 per-matrix scalar (signed [-2,1], scale = |A|max/1.5). + sA = (A.abs().amax().clamp_min(1e-10) / 1.5).to(torch.float16) + qA = torch.clamp(torch.round(A / sA.float()), -2, 1).to(torch.int8) + # B: INT4 groupwise g over flattened B (signed [-8,7], per-group scale). + Bf = B.reshape(-1, g) + Bmax = Bf.abs().amax(dim=-1, keepdim=True).clamp_min(1e-10) + sB = (Bmax / 7.5).to(torch.float16).reshape(-1) + qB = torch.clamp(torch.round(Bf / sB.float().reshape(-1, 1)), -8, 7).to( + torch.int8 + ).reshape(B.shape) + return qA, sA, qB, sB + + +def _lqer_fit_quantized(E, h): + U, S, Vh = torch.linalg.svd(E, full_matrices=False) + r = min(h.lqer_rank, S.numel()) + if r <= 0: + return None + A = (U[:, :r] * S[:r]).contiguous() + B = Vh[:r, :].contiguous() + asym_on = bool(getattr(h, "lqer_asym_enabled", False)) + asym_g = int(getattr(h, "lqer_asym_group", 64)) + if asym_on and B.numel() % asym_g == 0: + qA, sA, qB, sB = _lqer_pack_asym(A, B, asym_g) + A_hat = qA.float() * float(sA) + g_sz = qB.numel() // sB.numel() + B_hat = (qB.reshape(-1, g_sz).float() * sB.float().view(-1, 1)).reshape( + qB.shape + ) + return { + "kind": "asym", + "qA": qA, + "sA": sA, + "qB": qB, + "sB": sB, + "delta": A_hat @ B_hat, + } + qA, sA, qB, sB = _lqer_pack(A, B, h.lqer_factor_bits) + A_hat = qA.float() * sA.float().view(-1, 1) + B_hat = qB.float() * sB.float().view(-1, 1) + return { + "kind": "sym", + "qA": qA, + "sA": sA, + "qB": qB, + "sB": sB, + "delta": A_hat @ B_hat, + } + + +def _awq_lite_group_candidates(w, act_rms, group_size): + cols = w.shape[1] + n_groups = cols // group_size + if n_groups <= 0: + return [] + weight_score = w.float().abs().mean(dim=0) + saliency = act_rms.float() * weight_score + cands = [] + for gi in range(n_groups): + start = gi * group_size + end = start + group_size + score = float(saliency[start:end].sum()) + cands.append((score, start, end)) + return cands + + +def gptq_mixed_quantize(state_dict, hessians, act_stats, h): + result = {} + meta = {} + quant_gate = bool(getattr(h, "gated_attn_quant_gate", False)) + lqer_on = bool(getattr(h, "lqer_enabled", False)) + awq_on = bool(getattr(h, "awq_lite_enabled", False)) + lqer_cands = {} + awq_selected = collections.defaultdict(list) + if awq_on: + awq_cands = [] + for (name, tensor) in state_dict.items(): + t = tensor.detach().cpu().contiguous() + if t.is_floating_point() and t.numel() > 65536 and name in act_stats: + bits = h.embed_bits if "tok_emb" in name else h.matrix_bits + if bits < h.awq_lite_bits: + for score, start, end in _awq_lite_group_candidates( + t, act_stats[name], h.awq_lite_group_size + ): + awq_cands.append((score, name, start, end)) + awq_cands.sort(key=lambda x: -x[0]) + for (_score, name, start, end) in awq_cands[: h.awq_lite_group_top_k]: + awq_selected[name].append((start, end)) + for (name, tensor) in state_dict.items(): + t = tensor.detach().cpu().contiguous() + # Dedicated int8-per-row path for attn_gate_w (bypasses both GPTQ and + # fp16 passthrough). Applied BEFORE the numel<=65536 passthrough check + # so the gate tensor is routed here instead of to fp16. + if ( + quant_gate + and t.is_floating_point() + and t.ndim == 2 + and name.endswith(".attn_gate_w") + # Dense GatedAttn: (num_heads, dim) = (8, 512) = 4096. + # Sparse gate: (num_heads, gate_window) = (8, 12) = 96. + # Both need int8-per-row routing; the 1024 lower bound in stock + # PR-1736 presumed dense-only. Widen to catch both. + and 32 <= t.numel() <= 8192 + ): + gq, gs = _quantize_gate_int8_row(t) + result[name + ".gq"] = gq + result[name + ".gs"] = gs + meta[name] = "gate_int8_row" + continue + if not t.is_floating_point() or t.numel() <= 65536: + result[name] = t.to(torch.float16) if t.is_floating_point() else t + meta[name] = "passthrough (float16)" + continue + if "tok_emb" in name: + cs = h.embed_clip_sigmas + elif ".mlp." in name: + cs = h.mlp_clip_sigmas + elif ".attn." in name: + cs = h.attn_clip_sigmas + else: + cs = h.matrix_clip_sigmas + bits = h.embed_bits if "tok_emb" in name else h.matrix_bits + clip_range = 2 ** (bits - 1) - 1 + q, s, protect_meta = gptq_quantize_weight( + t, + hessians[name], + clip_sigmas=cs, + clip_range=clip_range, + protect_groups=awq_selected.get(name), + group_size=h.awq_lite_group_size if name in awq_selected else None, + protect_clip_range=(2 ** (h.awq_lite_bits - 1) - 1) + if name in awq_selected + else None, + ) + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = f"gptq (int{bits})" + W_q = q.float() * s.float().view(-1, 1) + if protect_meta is not None: + result[name + ".awqg_start"] = protect_meta["starts"] + result[name + ".awqg_s_hi"] = protect_meta["s_hi"] + result[name + ".awqg_size"] = torch.tensor( + protect_meta["size"], dtype=torch.int16 + ) + meta[name] = meta[name] + f"+awqgrpint{h.awq_lite_bits}" + gsz = protect_meta["size"] + for start in protect_meta["starts"].tolist(): + W_q[:, start : start + gsz] = ( + q[:, start : start + gsz].float() + * protect_meta["s_hi"].float().view(-1, 1) + ) + if lqer_on: + # LQER is fit on top of the fully realized GPTQ base, which already + # includes any higher-precision AWQ-protected groups. + scope = str(getattr(h, "lqer_scope", "all")).lower() + scope_ok = ( + scope == "all" + or (scope == "mlp" and ".mlp." in name) + or (scope == "attn" and ".attn." in name) + or (scope == "embed" and "tok_emb" in name) + ) + if scope_ok: + E = t.float() - W_q + err_norm = float(E.norm()) + if err_norm > 0: + lqer_cands[name] = (E, err_norm) + if lqer_on and lqer_cands: + if bool(getattr(h, "lqer_gain_select", False)): + scored = [] + for (name, (E, base_err)) in lqer_cands.items(): + fit = _lqer_fit_quantized(E, h) + if fit is None: + continue + new_err = float((E - fit["delta"]).norm()) + gain = base_err - new_err + if gain > 0: + scored.append((gain, name, fit)) + scored.sort(key=lambda x: -x[0]) + for (_gain, name, fit) in scored[: h.lqer_top_k]: + if fit["kind"] == "asym": + result[name + ".lqA_a"] = fit["qA"] + result[name + ".lqAs_a"] = fit["sA"] + result[name + ".lqB_a"] = fit["qB"] + result[name + ".lqBs_a"] = fit["sB"] + meta[name] = meta[name] + "+lqer_asym" + else: + result[name + ".lqA"] = fit["qA"] + result[name + ".lqAs"] = fit["sA"] + result[name + ".lqB"] = fit["qB"] + result[name + ".lqBs"] = fit["sB"] + meta[name] = meta[name] + "+lqer" + else: + top = sorted(lqer_cands.items(), key=lambda kv: -kv[1][1])[: h.lqer_top_k] + asym_on = bool(getattr(h, "lqer_asym_enabled", False)) + asym_g = int(getattr(h, "lqer_asym_group", 64)) + for (name, (E, _)) in top: + U, S, Vh = torch.linalg.svd(E, full_matrices=False) + r = min(h.lqer_rank, S.numel()) + A = (U[:, :r] * S[:r]).contiguous() + B = Vh[:r, :].contiguous() + if asym_on and B.numel() % asym_g == 0: + qA, sA, qB, sB = _lqer_pack_asym(A, B, asym_g) + result[name + ".lqA_a"] = qA + result[name + ".lqAs_a"] = sA + result[name + ".lqB_a"] = qB + result[name + ".lqBs_a"] = sB + meta[name] = meta[name] + "+lqer_asym" + else: + qA, sA, qB, sB = _lqer_pack(A, B, h.lqer_factor_bits) + result[name + ".lqA"] = qA + result[name + ".lqAs"] = sA + result[name + ".lqB"] = qB + result[name + ".lqBs"] = sB + meta[name] = meta[name] + "+lqer" + categories = collections.defaultdict(set) + for (name, cat) in meta.items(): + short = re.sub("\\.\\d+$", "", re.sub("blocks\\.\\d+", "blocks", name)) + categories[cat].add(short) + log("Quantized weights:") + for cat in sorted(categories): + log(f" {cat}: {', '.join(sorted(categories[cat]))}") + return result, meta + +def dequantize_mixed(result, meta, template_sd): + out = {} + for (name, orig) in template_sd.items(): + info = meta.get(name) + if info is None: + continue + orig_dtype = orig.dtype + if "passthrough" in info: + t = result[name] + if t.dtype == torch.float16 and orig_dtype in ( + torch.float32, + torch.bfloat16, + ): + t = t.to(orig_dtype) + out[name] = t + continue + if info == "gate_int8_row": + gq = result[name + ".gq"] + gs = result[name + ".gs"] + out[name] = (gq.float() * gs.float().view(-1, 1)).to(orig_dtype) + continue + q, s = result[name + ".q"], result[name + ".scale"] + if s.ndim > 0: + W = q.float() * s.float().view(q.shape[0], *[1] * (q.ndim - 1)) + else: + W = q.float() * float(s.item()) + if "awqgrpint" in info: + starts = result[name + ".awqg_start"].tolist() + s_hi = result[name + ".awqg_s_hi"].float() + gsz = int(result[name + ".awqg_size"].item()) + for start in starts: + W[:, start : start + gsz] = ( + q[:, start : start + gsz].float() * s_hi.view(-1, 1) + ) + if "lqer_asym" in info: + qA_t = result[name + ".lqA_a"] + sA_t = result[name + ".lqAs_a"] + qB_t = result[name + ".lqB_a"] + sB_t = result[name + ".lqBs_a"] + qA = qA_t.float() * float(sA_t) + g_sz = qB_t.numel() // sB_t.numel() + qB = (qB_t.reshape(-1, g_sz).float() * sB_t.float().view(-1, 1)).reshape( + qB_t.shape + ) + W = W + qA @ qB + elif "lqer" in info: + qA = result[name + ".lqA"].float() * result[name + ".lqAs"].float().view(-1, 1) + qB = result[name + ".lqB"].float() * result[name + ".lqBs"].float().view(-1, 1) + W = W + qA @ qB + out[name] = W.to(orig_dtype) + return out + + +_BSHF_MAGIC = b"BSHF" + + +# ── Per-group lrzip compression (ported from PR#1586 via PR#1667/1729) ──────── + +_GROUP_ORDER = [ + "_tok_emb.weight.q", + "attn.c_k.weight.q", "attn.c_q.weight.q", + "attn.c_v.weight.q", "attn.proj.weight.q", + "mlp.fc.weight.q", "mlp.proj.weight.q", +] +_SIMSORT_KEYS = {"_tok_emb.weight.q", "attn.c_q.weight.q", "mlp.fc.weight.q"} +_PACK_MAGIC = b"PGRP" + + +def _similarity_sort_l1(matrix): + import numpy as _np + n = matrix.shape[0] + used = _np.zeros(n, dtype=bool) + order = [0] + used[0] = True + cur = matrix[0].astype(_np.float32) + for _ in range(n - 1): + dists = _np.sum(_np.abs(matrix[~used].astype(_np.float32) - cur), axis=1) + unused = _np.where(~used)[0] + best = unused[_np.argmin(dists)] + order.append(best) + used[best] = True + cur = matrix[best].astype(_np.float32) + return _np.array(order, dtype=_np.uint16) + + +def _lrzip_compress(data, tmpdir, label): + inp = os.path.join(tmpdir, f"{label}.bin") + out = f"{inp}.lrz" + with open(inp, "wb") as f: + f.write(data) + subprocess.run(["lrzip", "-z", "-L", "9", "-o", out, inp], capture_output=True, check=True) + with open(out, "rb") as f: + result = f.read() + os.remove(inp); os.remove(out) + return result + + +def _lrzip_decompress(data, tmpdir, label): + inp = os.path.join(tmpdir, f"{label}.lrz") + out = os.path.join(tmpdir, f"{label}.bin") + with open(inp, "wb") as f: + f.write(data) + subprocess.run(["lrzip", "-d", "-f", "-o", out, inp], capture_output=True, check=True) + with open(out, "rb") as f: + result = f.read() + os.remove(inp); os.remove(out) + return result + + +def _pack_streams(streams): + import struct + n = len(streams) + hdr = _PACK_MAGIC + struct.pack("= 2 + docs.append((start, end - start)) + return docs + + +def _build_ttt_global_batches(doc_entries, h, ascending=False): + batch_size = h.ttt_batch_size + global_doc_entries = sorted(doc_entries, key=lambda x: x[1][1]) + global_batches = [ + global_doc_entries[i : i + batch_size] + for i in range(0, len(global_doc_entries), batch_size) + ] + indexed = list(enumerate(global_batches)) + if not ascending: + indexed.sort(key=lambda ib: -max(dl for _, (_, dl) in ib[1])) + return indexed + + +def _init_batch_counter(path): + with open(path, "wb") as f: + f.write((0).to_bytes(4, "little")) + + +def _claim_next_batch(counter_path, queue_len): + try: + with open(counter_path, "r+b") as f: + fcntl.flock(f, fcntl.LOCK_EX) + idx = int.from_bytes(f.read(4), "little") + f.seek(0) + f.write((idx + 1).to_bytes(4, "little")) + f.flush() + except FileNotFoundError: + return queue_len + return idx + + +def _compute_chunk_window(ci, pred_len, num_chunks, chunk_size, eval_seq_len): + chunk_end = pred_len if ci == num_chunks - 1 else (ci + 1) * chunk_size + win_start = max(0, chunk_end - eval_seq_len) + win_len = chunk_end - win_start + chunk_start = ci * chunk_size + chunk_offset = chunk_start - win_start + chunk_len = chunk_end - chunk_start + return win_start, win_len, chunk_offset, chunk_len + + +def _accumulate_bpb( + ptl, + x, + y, + chunk_offsets, + chunk_lens, + pos_idx, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + loss_sum, + byte_sum, + token_count, + y_bytes=None, +): + pos = pos_idx[: x.size(1)].unsqueeze(0) + mask = ( + (chunk_lens.unsqueeze(1) > 0) + & (pos >= chunk_offsets.unsqueeze(1)) + & (pos < (chunk_offsets + chunk_lens).unsqueeze(1)) + ) + mask_f64 = mask.to(torch.float64) + if y_bytes is not None: + tok_bytes = y_bytes.to(torch.float64) + else: + tok_bytes = base_bytes_lut[y].to(torch.float64) + tok_bytes += (has_leading_space_lut[y] & ~is_boundary_token_lut[x]).to( + torch.float64 + ) + loss_sum += (ptl.to(torch.float64) * mask_f64).sum() + byte_sum += (tok_bytes * mask_f64).sum() + token_count += chunk_lens.to(torch.float64).sum() + + +def _loss_bpb_from_sums(loss_sum, token_count, byte_sum): + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_sum.item()) + return val_loss, val_bpb + + +def _add_to_counter(path, delta): + try: + with open(path, "r+b") as f: + fcntl.flock(f, fcntl.LOCK_EX) + cur = int.from_bytes(f.read(8), "little", signed=True) + cur += int(delta) + f.seek(0) + f.write(int(cur).to_bytes(8, "little", signed=True)) + f.flush() + return cur + except FileNotFoundError: + return int(delta) + + +def _init_int64_counter(path): + with open(path, "wb") as f: + f.write((0).to_bytes(8, "little", signed=True)) + + +def _select_ttt_doc_entries(docs, h): + doc_entries = list(enumerate(docs)) + if h.val_doc_fraction < 1.0: + sample_n = max(1, int(round(len(docs) * h.val_doc_fraction))) + sampled_indices = sorted( + random.Random(h.seed).sample(range(len(docs)), sample_n) + ) + return [(i, docs[i]) for i in sampled_indices] + return doc_entries + + +def train_val_ttt_global_sgd_distributed(h, device, val_data, base_model, val_tokens, batch_seqs=None): + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + base_model.eval() + seq_len = h.eval_seq_len + total_tokens = val_tokens.numel() - 1 + ttt_chunk = h.global_ttt_chunk_tokens + batch_seqs = h.global_ttt_batch_seqs if batch_seqs is None else batch_seqs + num_chunks = (total_tokens + ttt_chunk - 1) // ttt_chunk + ttt_params = [p for p in base_model.parameters()] + for p in ttt_params: + p.requires_grad_(True) + optimizer = torch.optim.SGD( + ttt_params, lr=h.global_ttt_lr, momentum=h.global_ttt_momentum + ) + t_start = time.perf_counter() + for ci in range(num_chunks): + chunk_start = ci * ttt_chunk + chunk_end = min((ci + 1) * ttt_chunk, total_tokens) + is_last_chunk = ci == num_chunks - 1 + if is_last_chunk or h.global_ttt_epochs <= 0: + continue + base_model.train() + chunk_seqs = (chunk_end - chunk_start) // seq_len + if chunk_seqs <= 0: + continue + warmup_chunks = max(0, min(h.global_ttt_warmup_chunks, num_chunks - 1)) + if warmup_chunks > 0 and ci < warmup_chunks: + warmup_denom = max(warmup_chunks - 1, 1) + warmup_t = ci / warmup_denom + lr_now = ( + h.global_ttt_warmup_start_lr + + (h.global_ttt_lr - h.global_ttt_warmup_start_lr) * warmup_t + ) + else: + decay_steps = max(num_chunks - 1 - warmup_chunks, 1) + decay_ci = max(ci - warmup_chunks, 0) + lr_now = h.global_ttt_lr * 0.5 * ( + 1.0 + math.cos(math.pi * decay_ci / decay_steps) + ) + for pg in optimizer.param_groups: + pg["lr"] = lr_now + my_seq_s = chunk_seqs * h.rank // h.world_size + my_seq_e = chunk_seqs * (h.rank + 1) // h.world_size + my_chunk_seqs = my_seq_e - my_seq_s + for _ in range(h.global_ttt_epochs): + for bs in range(0, my_chunk_seqs, batch_seqs): + be = min(bs + batch_seqs, my_chunk_seqs) + actual_bs = my_seq_s + bs + start_tok = chunk_start + actual_bs * seq_len + end_tok = chunk_start + (my_seq_s + be) * seq_len + 1 + if end_tok > val_tokens.numel(): + continue + local = val_tokens[start_tok:end_tok].to(device=device, dtype=torch.int64) + x_flat = local[:-1] + y_flat = local[1:] + optimizer.zero_grad(set_to_none=True) + with torch.enable_grad(): + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + if h.global_ttt_respect_doc_boundaries: + bos_pos = (x_flat == BOS_ID).nonzero(as_tuple=True)[0].tolist() + cu_seqlens, max_seqlen = _build_cu_seqlens( + bos_pos, x_flat.numel(), x_flat.device, h.eval_seq_len, 64 + ) + loss = base_model( + x_flat[None], + y_flat[None], + cu_seqlens=cu_seqlens, + max_seqlen=max_seqlen, + ) + else: + x = x_flat.reshape(-1, seq_len) + y = y_flat.reshape(-1, seq_len) + loss = base_model(x, y) + loss.backward() + if dist.is_available() and dist.is_initialized(): + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.SUM) + p.grad.mul_(1.0 / h.world_size) + if h.global_ttt_grad_clip > 0: + torch.nn.utils.clip_grad_norm_(ttt_params, h.global_ttt_grad_clip) + optimizer.step() + base_model.eval() + if h.rank == 0: + elapsed = time.perf_counter() - t_start + log( + f"tttg: c{ci+1}/{num_chunks} lr:{lr_now:.6f} t:{elapsed:.1f}s" + ) + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + + +def _compute_ngram_hints_for_val(h, val_data, log0=print): + """Stage 1A: precompute ngram hints over full val token sequence. + Returns (hint_global, gate_global, boost_global) tensors on CPU, or None if tilt disabled. + + Compliance: single L->R pass over val tokens; uses val data only; produces hint + aligned to target positions [t] for predicting all_tokens[t+1] from prefix [:t+1]. + Same compute as inline precompute, just relocated to run BEFORE eval timer. + """ + if not getattr(h, "ngram_tilt_enabled", False): + return None + from online_ngram_tilt import build_hints_for_targets + all_tokens = val_data.val_tokens + targets_np_all = all_tokens.cpu().numpy().astype("uint16", copy=False)[1:] + t_h0 = time.perf_counter() + hints_pkg = build_hints_for_targets( + target_token_ids_np=targets_np_all, + tokenizer_path=h.tokenizer_path, + vocab_size=h.vocab_size, + log0=log0, + token_order=h.token_order, + token_threshold=h.token_threshold, + token_boost=h.token_boost, + within_tau=h.within_tau, + within_boost=h.within_boost, + word_order=h.word_order, + word_normalize=h.word_normalize, + word_tau=h.word_tau, + word_boost=h.word_boost, + agree_add_boost=h.agree_add_boost, + ) + hint_global = torch.from_numpy(hints_pkg["hint_ids"].astype("int64")) + gate_global = torch.from_numpy(hints_pkg["gate_mask"]) + boost_global = torch.from_numpy(hints_pkg["boost"].astype("float32")) + log0( + f"ngram_tilt:precompute_outside_timer_done elapsed={time.perf_counter()-t_h0:.2f}s " + f"total_targets={hint_global.numel()}" + ) + return (hint_global, gate_global, boost_global) + + +def eval_val_ttt_phased(h, base_model, device, val_data, forward_ttt_train, precomputed_hints=None): + 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) + # === PR #1145 n-gram tilt: precompute prefix-only hints over val targets === + # Hints are aligned to target positions: hint_global[i] is the hint for + # predicting token all_tokens[i+1] given prefix all_tokens[:i+1]. + # Stored on CPU as int64; gathered per-chunk to GPU alongside y indices. + ngram_hint_global = None + ngram_gate_global = None + ngram_boost_global = None + if precomputed_hints is not None: + # v5 Stage 1A: hints were precomputed BEFORE eval timer started. + # Save measured eval time = the precompute elapsed (~168s for full tilt). + ngram_hint_global, ngram_gate_global, ngram_boost_global = precomputed_hints + log( + f"ngram_tilt:using_precomputed_hints " + f"total_targets={ngram_hint_global.numel()} (precompute time excluded from eval)" + ) + elif getattr(h, "ngram_tilt_enabled", False): + from online_ngram_tilt import build_hints_for_targets + targets_np_all = all_tokens.cpu().numpy().astype("uint16", copy=False)[1:] + t_h0 = time.perf_counter() + hints_pkg = build_hints_for_targets( + target_token_ids_np=targets_np_all, + tokenizer_path=h.tokenizer_path, + vocab_size=h.vocab_size, + log0=log, + token_order=h.token_order, + token_threshold=h.token_threshold, + token_boost=h.token_boost, + within_tau=h.within_tau, + within_boost=h.within_boost, + word_order=h.word_order, + word_normalize=h.word_normalize, + word_tau=h.word_tau, + word_boost=h.word_boost, + agree_add_boost=h.agree_add_boost, + ) + ngram_hint_global = torch.from_numpy(hints_pkg["hint_ids"].astype("int64")) + ngram_gate_global = torch.from_numpy(hints_pkg["gate_mask"]) + ngram_boost_global = torch.from_numpy(hints_pkg["boost"].astype("float32")) + log( + f"ngram_tilt:precompute_done elapsed={time.perf_counter()-t_h0:.2f}s " + f"total_targets={ngram_hint_global.numel()}" + ) + docs = _find_docs(all_tokens) + doc_entries = _select_ttt_doc_entries(docs, h) + prefix_doc_limit = max(0, min(len(doc_entries), int(h.phased_ttt_prefix_docs))) + num_phases = max(1, int(h.phased_ttt_num_phases)) + phase_boundaries = [] + for pi in range(num_phases): + boundary = prefix_doc_limit * (pi + 1) // num_phases + phase_boundaries.append(boundary) + current_phase = 0 + current_phase_boundary = phase_boundaries[0] + log( + "ttt_phased:" + f" total_docs:{len(doc_entries)} prefix_docs:{prefix_doc_limit} " + f"suffix_docs:{len(doc_entries) - prefix_doc_limit}" + f" num_phases:{num_phases} boundaries:{phase_boundaries}" + ) + chunk_size, eval_seq_len = h.ttt_chunk_size, h.ttt_eval_seq_len + eval_batch_set = None + if h.ttt_eval_batches: + eval_batch_set = set(int(x) for x in h.ttt_eval_batches.split(",") if x.strip()) + use_ascending = eval_batch_set is not None + global_batches_sorted = _build_ttt_global_batches( + doc_entries, h, ascending=use_ascending + ) + queue_len = len(global_batches_sorted) + counter_path = f"/tmp/ttt_counter_{h.run_id}" + prefix_counter_path = f"/tmp/ttt_prefix_counter_{h.run_id}" + pause_flag_path = f"/tmp/ttt_pause_flag_{h.run_id}" + if h.rank == 0: + _init_batch_counter(counter_path) + _init_int64_counter(prefix_counter_path) + try: + os.remove(pause_flag_path) + except FileNotFoundError: + pass + if dist.is_available() and dist.is_initialized(): + path_list = [counter_path, prefix_counter_path, pause_flag_path] + dist.broadcast_object_list(path_list, src=0) + counter_path, prefix_counter_path, pause_flag_path = path_list + dist.barrier() + loss_sum = torch.zeros((), device=device, dtype=torch.float64) + byte_sum = torch.zeros((), device=device, dtype=torch.float64) + token_count = torch.zeros((), device=device, dtype=torch.float64) + t_start = time.perf_counter() + reusable_lora = BatchedTTTLoRA( + h.ttt_batch_size, base_model, h.ttt_lora_rank, + k_lora=h.ttt_k_lora, mlp_lora=h.ttt_mlp_lora, o_lora=h.ttt_o_lora, + ).to(device) + + def _build_opt(lora): + if h.ttt_optimizer == "sgd": + return torch.optim.SGD( + lora.parameters(), lr=h.ttt_lora_lr, + momentum=h.ttt_beta1, weight_decay=h.ttt_weight_decay, + ) + return torch.optim.AdamW( + lora.parameters(), lr=h.ttt_lora_lr, + betas=(h.ttt_beta1, h.ttt_beta2), + eps=1e-10, weight_decay=h.ttt_weight_decay, fused=True, + ) + + reusable_opt = _build_opt(reusable_lora) + local_scored_docs = [] + global_ttt_done = prefix_doc_limit == 0 + try: + while True: + queue_idx = _claim_next_batch(counter_path, queue_len) + if queue_idx >= queue_len: + break + orig_batch_idx, batch_entries = global_batches_sorted[queue_idx] + batch = [doc for _, doc in batch_entries] + bsz = len(batch) + prev_loss = loss_sum.item() + prev_bytes = byte_sum.item() + prev_tokens = token_count.item() + if bsz == reusable_lora.bsz: + reusable_lora.reset() + for s in reusable_opt.state.values(): + for k, v in s.items(): + if isinstance(v, torch.Tensor): + v.zero_() + elif k == "step": + s[k] = 0 + cur_lora = reusable_lora + cur_opt = reusable_opt + else: + cur_lora = BatchedTTTLoRA( + bsz, base_model, h.ttt_lora_rank, + 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) + # n-gram tilt path: gather hints aligned to y, pass into forward_ttt + hint_ids_gpu = None + gate_mask_gpu = None + boost_gpu = None + if ngram_hint_global is not None: + hint_idx_cpu = ( + tok_starts.unsqueeze(1) + col_idx[:context_size].unsqueeze(0) + ).clamp_(min=0, max=ngram_hint_global.numel() - 1) + hint_ids_gpu = ngram_hint_global[hint_idx_cpu].to( + device=device, dtype=torch.int64, non_blocking=True + ) + gate_mask_gpu = ngram_gate_global[hint_idx_cpu].to( + device=device, non_blocking=True + ) + boost_gpu = ngram_boost_global[hint_idx_cpu].to( + device=device, dtype=torch.float32, non_blocking=True + ) + hint_ids_gpu = torch.where(valid, hint_ids_gpu, torch.zeros_like(hint_ids_gpu)) + gate_mask_gpu = gate_mask_gpu & valid + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + if hint_ids_gpu is not None: + per_tok_loss, log_q_hint = forward_ttt_train( + x, y, lora=cur_lora, hint_ids=hint_ids_gpu + ) + else: + per_tok_loss = forward_ttt_train(x, y, lora=cur_lora) + log_q_hint = None + # CaseOps sidecar-driven byte budget. Mirror the index pattern + # used to build y from all_tokens: y[b, j] corresponds to the + # token at global position tok_starts[b] + 1 + j (when valid). + y_bytes_arg = None + if val_data.caseops_enabled and val_data.val_bytes is not None: + y_idx = ( + tok_starts.unsqueeze(1) + + 1 + + col_idx[:context_size].unsqueeze(0) + ) + y_idx = y_idx.clamp_(max=val_data.val_bytes.numel() - 1) + y_bytes_arg = val_data.val_bytes[y_idx].to( + device=device, dtype=torch.int32, non_blocking=True + ) + # Mirror the `valid` masking used for y so out-of-range tokens + # contribute zero bytes (matches y=0 substitution above). + y_bytes_arg = torch.where( + valid, y_bytes_arg, torch.zeros_like(y_bytes_arg) + ) + # n-gram tilt application: use tilted ptl for BPB accumulation, + # but keep original per_tok_loss for TTT-LoRA backward (training + # objective is base NLL — tilt is a scoring-time overlay). + if hint_ids_gpu is not None and log_q_hint is not None: + from online_ngram_tilt import apply_tilt_to_ptl_torch_fast as apply_tilt_to_ptl_torch + tilted_loss = apply_tilt_to_ptl_torch( + ptl=per_tok_loss, + log_q_hint=log_q_hint, + target_ids=y, + hint_ids=hint_ids_gpu, + gate_mask=gate_mask_gpu, + boost=boost_gpu, + ) + else: + tilted_loss = per_tok_loss + with torch.no_grad(): + _accumulate_bpb( + tilted_loss, + x, + y, + chunk_offsets, + chunk_lens, + ctx_pos, + val_data.base_bytes_lut, + val_data.has_leading_space_lut, + val_data.is_boundary_token_lut, + loss_sum, + byte_sum, + token_count, + y_bytes=y_bytes_arg, + ) + if needs_train: + activate_chunk_mask = (num_chunks_t - 1 > ci).float() + for gi in range(h.ttt_grad_steps): + if gi > 0: + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + per_tok_loss = forward_ttt_train(x, y, lora=cur_lora) + per_doc = per_tok_loss[ + :, chunk_offset : chunk_offset + chunk_size + ].mean(dim=-1) + cur_opt.zero_grad(set_to_none=True) + (per_doc * activate_chunk_mask).sum().backward() + cur_opt.step() + else: + del per_tok_loss + batch_num = orig_batch_idx + 1 + doc_lens = [dl for _, dl in batch] + should_report = batch_num in eval_batch_set if eval_batch_set is not None else True + if should_report: + cur_tokens = token_count.item() + cur_loss_val = loss_sum.item() + cur_bytes_val = byte_sum.item() + dt = cur_tokens - prev_tokens + db = cur_bytes_val - prev_bytes + if dt > 0 and db > 0: + b_loss = (cur_loss_val - prev_loss) / dt + b_bpb = b_loss / math.log(2.0) * (dt / db) + else: + b_loss = b_bpb = 0.0 + r_loss = cur_loss_val / max(cur_tokens, 1) + r_bpb = r_loss / math.log(2.0) * (cur_tokens / max(cur_bytes_val, 1)) + elapsed = time.perf_counter() - t_start + log( + f"ttp: b{batch_num}/{queue_len} bl:{b_loss:.4f} bb:{b_bpb:.4f} " + f"rl:{r_loss:.4f} rb:{r_bpb:.4f} dl:{min(doc_lens)}-{max(doc_lens)} " + f"gd:{int(global_ttt_done)}" + ) + if not global_ttt_done: + local_scored_docs.extend( + (orig_batch_idx, pos, doc_start, doc_len) + for pos, (doc_start, doc_len) in enumerate(batch) + ) + prefix_done = _add_to_counter(prefix_counter_path, len(batch_entries)) + if prefix_done >= current_phase_boundary: + try: + with open(pause_flag_path, "x"): + pass + except FileExistsError: + pass + should_pause = os.path.exists(pause_flag_path) + if should_pause: + if dist.is_available() and dist.is_initialized(): + dist.barrier() + gathered_scored_docs = [None] * h.world_size + if dist.is_available() and dist.is_initialized(): + dist.all_gather_object(gathered_scored_docs, local_scored_docs) + else: + gathered_scored_docs = [local_scored_docs] + scored_docs_for_global = [] + for rank_docs in gathered_scored_docs: + if rank_docs: + scored_docs_for_global.extend(rank_docs) + scored_docs_for_global.sort(key=lambda x: (x[0], x[1])) + scored_docs_for_global = scored_docs_for_global[:current_phase_boundary] + scored_token_chunks = [ + val_data.val_tokens[doc_start : doc_start + doc_len] + for _, _, doc_start, doc_len in scored_docs_for_global + ] + if scored_token_chunks: + global_ttt_tokens = torch.cat(scored_token_chunks) + else: + global_ttt_tokens = val_data.val_tokens[:0] + if h.rank == 0: + prefix_done = 0 + try: + with open(prefix_counter_path, "rb") as f: + prefix_done = int.from_bytes( + f.read(8), "little", signed=True + ) + except FileNotFoundError: + pass + log( + f"ttpp: phase:{current_phase + 1}/{num_phases} pd:{prefix_done} " + f"gd:{len(scored_docs_for_global)} " + f"t:{time.perf_counter() - t_start:.1f}s" + ) + train_val_ttt_global_sgd_distributed( + h, device, val_data, base_model, global_ttt_tokens + ) + for p in base_model.parameters(): + p.requires_grad_(False) + reusable_lora = BatchedTTTLoRA( + h.ttt_batch_size, base_model, h.ttt_lora_rank, + k_lora=h.ttt_k_lora, mlp_lora=h.ttt_mlp_lora, o_lora=h.ttt_o_lora, + ).to(device) + reusable_opt = _build_opt(reusable_lora) + current_phase += 1 + if current_phase >= num_phases: + global_ttt_done = True + else: + current_phase_boundary = phase_boundaries[current_phase] + if h.rank == 0: + try: + os.remove(pause_flag_path) + except FileNotFoundError: + pass + if dist.is_available() and dist.is_initialized(): + dist.barrier() + if h.rank == 0: + log(f"ttpr: phase:{current_phase}/{num_phases} t:{time.perf_counter() - t_start:.1f}s") + del cur_lora, cur_opt + finally: + pass + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(byte_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(token_count, op=dist.ReduceOp.SUM) + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.train() + return _loss_bpb_from_sums(loss_sum, token_count, byte_sum) + + +def timed_eval(label, fn, *args, **kwargs): + torch.cuda.synchronize() + t0 = time.perf_counter() + val_loss, val_bpb = fn(*args, **kwargs) + torch.cuda.synchronize() + elapsed_ms = 1e3 * (time.perf_counter() - t0) + log( + f"{label} val_loss:{val_loss:.8f} val_bpb:{val_bpb:.8f} eval_time:{elapsed_ms:.0f}ms" + ) + return val_loss, val_bpb + + +def train_model(h, device, val_data): + base_model = GPT(h).to(device).bfloat16() + restore_fp32_params(base_model) + compiled_model = torch.compile(base_model, dynamic=False, fullgraph=True) + compiled_forward_logits = torch.compile( + base_model.forward_logits, dynamic=False, fullgraph=True + ) + model = compiled_model + log(f"model_params:{sum(p.numel()for p in base_model.parameters())}") + 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 + + _clip_params = [p for p in base_model.parameters() if p.requires_grad] + def step_fn(step, lr_scale): + 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 + if step <= h.muon_momentum_warmup_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_(_clip_params, 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) + _live_state = base_model.state_dict(keep_vars=True) + ema_state = { + name: t.detach().float().clone() + for (name, t) in _live_state.items() + } + _ema_pairs = [(ema_state[name], t) for (name, t) in _live_state.items()] + ema_decay = h.ema_decay + training_time_ms = 0.0 + forced_stop_step = int(os.environ.get("FORCE_STOP_STEP", "0")) + stop_after_step = forced_stop_step if forced_stop_step > 0 else 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 ema_t, t in _ema_pairs: + ema_t.mul_(ema_decay).add_(t.detach(), 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 = ( + forced_stop_step <= 0 + and max_wallclock_ms is not None + and approx_training_time_ms >= max_wallclock_ms + ) + if h.distributed and forced_stop_step <= 0 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): + global BOS_ID + random.seed(h.seed) + np.random.seed(h.seed) + torch.manual_seed(h.seed) + torch.cuda.manual_seed_all(h.seed) + if h.artifact_dir and h.is_main_process: + os.makedirs(h.artifact_dir, exist_ok=True) + val_data = ValidationData(h, device) + log( + f"train_shards: {len(list(Path(h.datasets_dir).resolve().glob('fineweb_train_*.bin')))}" + ) + log(f"val_tokens: {val_data.val_tokens.numel()-1}") + # TTT_EVAL_ONLY: skip training + GPTQ, jump straight to TTT eval on a + # pre-existing quantized artifact. Used to test TTT-only improvements + # (e.g., PR-1767's alpha/warm-start/WD) without retraining. + ttt_eval_only = os.environ.get("TTT_EVAL_ONLY", "0") == "1" + quantize_only = os.environ.get("QUANTIZE_ONLY", "0") == "1" + if ttt_eval_only: + log("TTT_EVAL_ONLY=1 — skipping training + GPTQ, loading saved artifact for TTT eval") + log(f"ttt_lora_alpha: {BatchedLinearLoRA._ALPHA}") + log(f"ttt_warm_start_a: {BatchedLinearLoRA._WARM_START_A}") + log(f"ttt_weight_decay: {h.ttt_weight_decay}") + elif quantize_only: + log("QUANTIZE_ONLY=1 — skipping training, loading saved full-precision checkpoint") + log(f"quantize_only checkpoint: {h.model_path}") + if BOS_ID is None: + BOS_ID = 1 + base_model = GPT(h).to(device).bfloat16() + state = torch.load(h.model_path, map_location="cpu") + base_model.load_state_dict(state, strict=True) + del state + serialize(h, base_model, Path(__file__).read_text(encoding="utf-8")) + if h.distributed: + dist.barrier() + else: + base_model, compiled_model, compiled_forward_logits = train_model( + h, device, val_data + ) + torch._dynamo.reset() + timed_eval( + "diagnostic pre-quantization post-ema", + eval_val, + h, + device, + val_data, + compiled_model, + compiled_forward_logits, + ) + if os.environ.get("PREQUANT_ONLY", "0") == "1": + log("PREQUANT_ONLY=1 — skipping serialize/GPTQ/post-quant eval/TTT") + return + 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 + if not ttt_eval_only: + 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, + ) + del eval_model + if h.ttt_enabled: + if not ttt_eval_only: + del compiled_model + if ttt_eval_only: + del eval_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) + + def _fwd_ttt_inner_with_hints(input_ids, target_ids, lora, hint_ids): + return ttt_model.forward_ttt(input_ids, target_ids, lora=lora, hint_ids=hint_ids) + + _fwd_ttt_compiled_inner = None + _fwd_ttt_compiled_inner_hints = None + + def _fwd_ttt(input_ids, target_ids, lora, hint_ids=None): + nonlocal _fwd_ttt_compiled_inner, _fwd_ttt_compiled_inner_hints + if hint_ids is None: + 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) + if _fwd_ttt_compiled_inner_hints is None: + _fwd_ttt_compiled_inner_hints = torch.compile( + _fwd_ttt_inner_with_hints, dynamic=True + ) + return _fwd_ttt_compiled_inner_hints( + input_ids, target_ids, lora=lora, hint_ids=hint_ids + ) + + fwd_ttt_compiled = _fwd_ttt + log(f"ttt_lora:warming up compile (random tokens, no val data)") + 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)") + # v5 Stage 1A: precompute ngram hints BEFORE eval timer (single pass causal, + # uses val tokens only — same compliance as inline). For full tilt this saves + # ~168s of measured eval time without losing any tilt benefit. + precomputed_hints = None + if h.ngram_tilt_enabled and getattr(h, "ngram_hint_precompute_outside", True): + log("v5:precomputing ngram hints OUTSIDE eval timer") + precomputed_hints = _compute_ngram_hints_for_val(h, val_data, log0=log) + log("\nbeginning TTT eval timer") + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_val_loss, ttt_val_bpb = eval_val_ttt_phased( + h, ttt_model, device, val_data, forward_ttt_train=fwd_ttt_compiled, + precomputed_hints=precomputed_hints, + ) + torch.cuda.synchronize() + ttt_eval_elapsed = time.perf_counter() - t_ttt + log( + "quantized_ttt_phased " + f"val_loss:{ttt_val_loss:.8f} val_bpb:{ttt_val_bpb:.8f} " + f"eval_time:{1e3*ttt_eval_elapsed:.0f}ms" + ) + log(f"total_eval_time:{ttt_eval_elapsed:.1f}s") + del ttt_model + + +def main(): + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError( + f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral" + ) + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + torch.set_float32_matmul_precision("high") + from torch.backends.cuda import ( + enable_cudnn_sdp, + enable_flash_sdp, + enable_math_sdp, + enable_mem_efficient_sdp, + ) + + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + torch._dynamo.config.optimize_ddp = False + torch._dynamo.config.cache_size_limit = 64 + h = Hyperparameters() + set_logging_hparams(h) + if h.is_main_process: + os.makedirs(h.artifact_dir if h.artifact_dir else "logs", exist_ok=True) + log(100 * "=", console=False) + log("Hyperparameters:", console=True) + for (k, v) in sorted(vars(type(h)).items()): + if not k.startswith("_"): + log(f" {k}: {v}", console=True) + log("=" * 100, console=False) + log("Source code:", console=False) + log("=" * 100, console=False) + with open(__file__, "r", encoding="utf-8") as _src: + log(_src.read(), console=False) + log("=" * 100, console=False) + log(f"Running Python {sys.version}", console=False) + log(f"Running PyTorch {torch.__version__}", console=False) + log("=" * 100, console=False) + train_and_eval(h, device) + if distributed: + dist.destroy_process_group() + + +if __name__ == "__main__": + main() diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed0.log b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed0.log new file mode 100644 index 0000000000..a4e5663c8d --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed0.log @@ -0,0 +1,480 @@ +=== #2018 emulation (no Gated XSA) seed=0 === +PHASED_TTT_NUM_PHASES=1 PHASED_TTT_PREFIX_DOCS=1000 NGRAM_INSIDE=1 RANK=80 +Thu Apr 30 22:48:55 UTC 2026 +W0430 22:48:57.230000 417040 torch/distributed/run.py:803] +W0430 22:48:57.230000 417040 torch/distributed/run.py:803] ***************************************** +W0430 22:48:57.230000 417040 torch/distributed/run.py:803] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0430 22:48:57.230000 417040 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + agree_add_boost: 0.5 + artifact_dir: + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + awq_lite_bits: 8 + awq_lite_enabled: True + awq_lite_group_size: 64 + awq_lite_group_top_k: 1 + beta1: 0.9 + beta2: 0.99 + caseops_enabled: True + compressor: pergroup + data_dir: /workspace/pg-data/datasets + datasets_dir: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 14.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2560 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.01 + gated_attn_quant_gate: True + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 4.0 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/91965cd9-b80b-4b44-9ae0-4fc38ca0d5e1.txt + logit_softcap: 30.0 + loop_end: 5 + loop_start: 3 + lqer_asym_enabled: True + lqer_asym_group: 64 + lqer_enabled: True + lqer_factor_bits: 4 + lqer_gain_select: False + lqer_rank: 4 + lqer_scope: all + lqer_top_k: 1 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 11.5 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + ngram_hint_precompute_outside: False + ngram_tilt_enabled: True + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 1 + phased_ttt_prefix_docs: 1000 + qk_gain_init: 5.25 + 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: 91965cd9-b80b-4b44-9ae0-4fc38ca0d5e1 + scalar_lr: 0.02 + seed: 0 + skip_gates_enabled: True + smear_gate_enabled: True + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 0.5 + temperature_scale: 1.0 + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + token_boost: 2.625 + token_order: 16 + token_threshold: 0.8 + tokenizer_path: /workspace/pg-data/datasets/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/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: 2560 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_lr: 0.0001 + ttt_lora_rank: 80 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 2.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.85 + warmup_steps: 20 + within_boost: 0.75 + within_tau: 0.45 + word_boost: 0.75 + word_normalize: strip_punct_lower + word_order: 4 + word_tau: 0.65 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 47851520 +model_params:35945673 +gptq:reserving 4s, effective=596000ms +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: 9.0094 val_bpb: 4.1166 +1/20000 train_loss: 9.0105 train_time: 0.0m tok/s: 17223257 +2/20000 train_loss: 12.9626 train_time: 0.0m tok/s: 10674744 +3/20000 train_loss: 10.2790 train_time: 0.0m tok/s: 9342697 +4/20000 train_loss: 8.7850 train_time: 0.0m tok/s: 8848126 +5/20000 train_loss: 8.0092 train_time: 0.0m tok/s: 8557442 +500/20000 train_loss: 2.5651 train_time: 0.8m tok/s: 8227387 +1000/20000 train_loss: 2.8016 train_time: 1.6m tok/s: 8187851 +1500/20000 train_loss: 2.6160 train_time: 2.4m tok/s: 8172164 +2000/20000 train_loss: 2.6517 train_time: 3.2m tok/s: 8167491 +layer_loop:enabled step:2164 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: 2.5365 train_time: 4.3m tok/s: 7681686 +3000/20000 train_loss: 2.5535 train_time: 5.4m tok/s: 7225996 +3500/20000 train_loss: 2.5576 train_time: 6.6m tok/s: 6931504 +4000/20000 train_loss: 2.4017 train_time: 7.8m tok/s: 6716739 +4000/20000 val_loss: 2.4174 val_bpb: 1.1046 +4500/20000 train_loss: 2.2769 train_time: 9.0m tok/s: 6566635 +4904/20000 val_loss: 2.3505 val_bpb: 1.0740 +stopping_early: wallclock_cap train_time: 596017ms step: 4904/20000 +peak memory allocated: 41707 MiB reserved: 46952 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.32532429 val_bpb:1.06248986 eval_time:9029ms +Serialized model: 135418111 bytes +Code size (uncompressed): 182796 bytes +Code size (compressed): 35707 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 5.3s +Quantized weights: + gate_int8_row: blocks.attn.attn_gate_w + 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)+awqgrpint8+lqer_asym: 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, smear_gate.weight, smear_lambda, softcap_neg, softcap_pos +Serialize: per-group lrzip compression... +Serialize: per-group compression done in 92.1s +Serialized model quantized+pergroup: 15940259 bytes +Total submission size quantized+pergroup: 15975966 bytes +Deserialize: per-group lrzip decompression... +Deserialize: decompression done in 16.3s +diagnostic quantized val_loss:2.34322542 val_bpb:1.07066926 eval_time:14356ms +Deserialize: per-group lrzip decompression... +Deserialize: decompression done in 16.3s +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (84.9s) + +beginning TTT eval timer +ngram_tilt:hints total=47851520 gated=13023303 token_gate=628130 within_gate=9866847 word_gate=2891588 agree2plus=303177 +ngram_tilt:precompute_done elapsed=124.92s total_targets=47851520 +ttt_phased: total_docs:50000 prefix_docs:1000 suffix_docs:49000 num_phases:1 boundaries:[1000] +ttp: b775/782 bl:2.2696 bb:1.0551 rl:2.2696 rb:1.0551 dl:6892-7524 gd:0 +ttp: b774/782 bl:2.2779 bb:1.0606 rl:2.2736 rb:1.0577 dl:6447-6872 gd:0 +ttp: b770/782 bl:2.2801 bb:1.0764 rl:2.2754 rb:1.0629 dl:5311-5522 gd:0 +ttp: b765/782 bl:2.3126 bb:1.0816 rl:2.2824 rb:1.0664 dl:4393-4510 gd:0 +ttp: b761/782 bl:2.4048 bb:1.1087 rl:2.3000 rb:1.0726 dl:3916-4032 gd:0 +ttpp: phase:1/1 pd:1424 gd:1000 t:236.6s +tttg: c1/154 lr:0.001000 t:0.3s +tttg: c2/154 lr:0.001000 t:0.4s +tttg: c3/154 lr:0.001000 t:0.6s +tttg: c4/154 lr:0.000999 t:0.7s +tttg: c5/154 lr:0.000998 t:0.8s +tttg: c6/154 lr:0.000997 t:0.9s +tttg: c7/154 lr:0.000996 t:1.1s +tttg: c8/154 lr:0.000995 t:1.2s +tttg: c9/154 lr:0.000993 t:1.3s +tttg: c10/154 lr:0.000991 t:1.4s +tttg: c11/154 lr:0.000989 t:1.6s +tttg: c12/154 lr:0.000987 t:1.7s +tttg: c13/154 lr:0.000985 t:1.8s +tttg: c14/154 lr:0.000982 t:1.9s +tttg: c15/154 lr:0.000979 t:2.1s +tttg: c16/154 lr:0.000976 t:2.2s +tttg: c17/154 lr:0.000973 t:2.3s +tttg: c18/154 lr:0.000970 t:2.4s +tttg: c19/154 lr:0.000966 t:2.6s +tttg: c20/154 lr:0.000962 t:2.7s +tttg: c21/154 lr:0.000958 t:2.8s +tttg: c22/154 lr:0.000954 t:2.9s +tttg: c23/154 lr:0.000950 t:3.1s +tttg: c24/154 lr:0.000945 t:3.2s +tttg: c25/154 lr:0.000941 t:3.3s +tttg: c26/154 lr:0.000936 t:3.4s +tttg: c27/154 lr:0.000930 t:3.6s +tttg: c28/154 lr:0.000925 t:3.7s +tttg: c29/154 lr:0.000920 t:3.8s +tttg: c30/154 lr:0.000914 t:3.9s +tttg: c31/154 lr:0.000908 t:4.0s +tttg: c32/154 lr:0.000902 t:4.1s +tttg: c33/154 lr:0.000896 t:4.2s +tttg: c34/154 lr:0.000890 t:4.4s +tttg: c35/154 lr:0.000883 t:4.5s +tttg: c36/154 lr:0.000876 t:4.6s +tttg: c37/154 lr:0.000870 t:4.7s +tttg: c38/154 lr:0.000863 t:4.8s +tttg: c39/154 lr:0.000855 t:4.9s +tttg: c40/154 lr:0.000848 t:5.0s +tttg: c41/154 lr:0.000841 t:5.1s +tttg: c42/154 lr:0.000833 t:5.2s +tttg: c43/154 lr:0.000825 t:5.3s +tttg: c44/154 lr:0.000817 t:5.4s +tttg: c45/154 lr:0.000809 t:5.5s +tttg: c46/154 lr:0.000801 t:5.6s +tttg: c47/154 lr:0.000793 t:5.7s +tttg: c48/154 lr:0.000785 t:5.8s +tttg: c49/154 lr:0.000776 t:5.9s +tttg: c50/154 lr:0.000768 t:6.0s +tttg: c51/154 lr:0.000759 t:6.1s +tttg: c52/154 lr:0.000750 t:6.2s +tttg: c53/154 lr:0.000741 t:6.3s +tttg: c54/154 lr:0.000732 t:6.5s +tttg: c55/154 lr:0.000723 t:6.6s +tttg: c56/154 lr:0.000714 t:6.7s +tttg: c57/154 lr:0.000704 t:6.8s +tttg: c58/154 lr:0.000695 t:6.9s +tttg: c59/154 lr:0.000685 t:7.0s +tttg: c60/154 lr:0.000676 t:7.1s +tttg: c61/154 lr:0.000666 t:7.2s +tttg: c62/154 lr:0.000656 t:7.3s +tttg: c63/154 lr:0.000647 t:7.4s +tttg: c64/154 lr:0.000637 t:7.5s +tttg: c65/154 lr:0.000627 t:7.6s +tttg: c66/154 lr:0.000617 t:7.7s +tttg: c67/154 lr:0.000607 t:7.8s +tttg: c68/154 lr:0.000597 t:7.9s +tttg: c69/154 lr:0.000587 t:8.0s +tttg: c70/154 lr:0.000577 t:8.1s +tttg: c71/154 lr:0.000567 t:8.2s +tttg: c72/154 lr:0.000556 t:8.3s +tttg: c73/154 lr:0.000546 t:8.4s +tttg: c74/154 lr:0.000536 t:8.5s +tttg: c75/154 lr:0.000526 t:8.7s +tttg: c76/154 lr:0.000515 t:8.8s +tttg: c77/154 lr:0.000505 t:8.9s +tttg: c78/154 lr:0.000495 t:9.0s +tttg: c79/154 lr:0.000485 t:9.1s +tttg: c80/154 lr:0.000474 t:9.2s +tttg: c81/154 lr:0.000464 t:9.3s +tttg: c82/154 lr:0.000454 t:9.4s +tttg: c83/154 lr:0.000444 t:9.5s +tttg: c84/154 lr:0.000433 t:9.6s +tttg: c85/154 lr:0.000423 t:9.7s +tttg: c86/154 lr:0.000413 t:9.8s +tttg: c87/154 lr:0.000403 t:9.9s +tttg: c88/154 lr:0.000393 t:10.0s +tttg: c89/154 lr:0.000383 t:10.1s +tttg: c90/154 lr:0.000373 t:10.2s +tttg: c91/154 lr:0.000363 t:10.3s +tttg: c92/154 lr:0.000353 t:10.5s +tttg: c93/154 lr:0.000344 t:10.6s +tttg: c94/154 lr:0.000334 t:10.7s +tttg: c95/154 lr:0.000324 t:10.8s +tttg: c96/154 lr:0.000315 t:10.9s +tttg: c97/154 lr:0.000305 t:11.0s +tttg: c98/154 lr:0.000296 t:11.3s +tttg: c99/154 lr:0.000286 t:11.4s +tttg: c100/154 lr:0.000277 t:11.5s +tttg: c101/154 lr:0.000268 t:11.6s +tttg: c102/154 lr:0.000259 t:11.7s +tttg: c103/154 lr:0.000250 t:11.8s +tttg: c104/154 lr:0.000241 t:11.9s +tttg: c105/154 lr:0.000232 t:12.0s +tttg: c106/154 lr:0.000224 t:12.2s +tttg: c107/154 lr:0.000215 t:12.3s +tttg: c108/154 lr:0.000207 t:12.4s +tttg: c109/154 lr:0.000199 t:12.5s +tttg: c110/154 lr:0.000191 t:12.6s +tttg: c111/154 lr:0.000183 t:12.7s +tttg: c112/154 lr:0.000175 t:12.8s +tttg: c113/154 lr:0.000167 t:12.9s +tttg: c114/154 lr:0.000159 t:13.0s +tttg: c115/154 lr:0.000152 t:13.1s +tttg: c116/154 lr:0.000145 t:13.2s +tttg: c117/154 lr:0.000137 t:13.3s +tttg: c118/154 lr:0.000130 t:13.4s +tttg: c119/154 lr:0.000124 t:13.5s +tttg: c120/154 lr:0.000117 t:13.6s +tttg: c121/154 lr:0.000110 t:13.7s +tttg: c122/154 lr:0.000104 t:13.8s +tttg: c123/154 lr:0.000098 t:13.9s +tttg: c124/154 lr:0.000092 t:14.0s +tttg: c125/154 lr:0.000086 t:14.1s +tttg: c126/154 lr:0.000080 t:14.2s +tttg: c127/154 lr:0.000075 t:14.3s +tttg: c128/154 lr:0.000070 t:14.4s +tttg: c129/154 lr:0.000064 t:14.6s +tttg: c130/154 lr:0.000059 t:14.7s +tttg: c131/154 lr:0.000055 t:14.8s +tttg: c132/154 lr:0.000050 t:14.9s +tttg: c133/154 lr:0.000046 t:15.0s +tttg: c134/154 lr:0.000042 t:15.1s +tttg: c135/154 lr:0.000038 t:15.2s +tttg: c136/154 lr:0.000034 t:15.3s +tttg: c137/154 lr:0.000030 t:15.4s +tttg: c138/154 lr:0.000027 t:15.5s +tttg: c139/154 lr:0.000024 t:15.6s +tttg: c140/154 lr:0.000021 t:15.7s +tttg: c141/154 lr:0.000018 t:15.8s +tttg: c142/154 lr:0.000015 t:15.9s +tttg: c143/154 lr:0.000013 t:16.0s +tttg: c144/154 lr:0.000011 t:16.1s +tttg: c145/154 lr:0.000009 t:16.2s +tttg: c146/154 lr:0.000007 t:16.3s +tttg: c147/154 lr:0.000005 t:16.4s +tttg: c148/154 lr:0.000004 t:16.5s +tttg: c149/154 lr:0.000003 t:16.6s +tttg: c150/154 lr:0.000002 t:16.8s +tttg: c151/154 lr:0.000001 t:16.9s +tttg: c152/154 lr:0.000000 t:17.0s +tttg: c153/154 lr:0.000000 t:17.1s +ttpr: phase:1/1 t:255.9s +ttp: b752/782 bl:2.3225 bb:1.0676 rl:2.3024 rb:1.0720 dl:3222-3283 gd:1 +ttp: b751/782 bl:2.3020 bb:1.0305 rl:2.3023 rb:1.0680 dl:3150-3221 gd:1 +ttp: b746/782 bl:2.4039 bb:1.0592 rl:2.3103 rb:1.0673 dl:2884-2943 gd:1 +ttp: b742/782 bl:2.3165 bb:1.0429 rl:2.3107 rb:1.0656 dl:2730-2762 gd:1 +ttp: b737/782 bl:2.3083 bb:1.0377 rl:2.3106 rb:1.0638 dl:2550-2583 gd:1 +ttp: b735/782 bl:2.3803 bb:1.0951 rl:2.3145 rb:1.0656 dl:2495-2526 gd:1 +ttp: b730/782 bl:2.2614 bb:0.9938 rl:2.3118 rb:1.0618 dl:2352-2376 gd:1 +ttp: b726/782 bl:2.3280 bb:1.0355 rl:2.3126 rb:1.0606 dl:2254-2276 gd:1 +ttp: b721/782 bl:2.2999 bb:1.0214 rl:2.3120 rb:1.0589 dl:2144-2163 gd:1 +ttp: b716/782 bl:2.2479 bb:1.0388 rl:2.3096 rb:1.0581 dl:2054-2069 gd:1 +ttp: b712/782 bl:2.3325 bb:1.0579 rl:2.3104 rb:1.0581 dl:1984-2002 gd:1 +ttp: b706/782 bl:2.3943 bb:1.0708 rl:2.3132 rb:1.0586 dl:1898-1910 gd:1 +ttp: b700/782 bl:2.2688 bb:1.0132 rl:2.3118 rb:1.0571 dl:1824-1834 gd:1 +ttp: b694/782 bl:2.3033 bb:1.0533 rl:2.3116 rb:1.0570 dl:1758-1769 gd:1 +ttp: b687/782 bl:2.3085 bb:1.0542 rl:2.3115 rb:1.0569 dl:1685-1696 gd:1 +ttp: b680/782 bl:2.2773 bb:1.0256 rl:2.3106 rb:1.0561 dl:1618-1628 gd:1 +ttp: b674/782 bl:2.3984 bb:1.0862 rl:2.3127 rb:1.0569 dl:1571-1578 gd:1 +ttp: b667/782 bl:2.3586 bb:1.0662 rl:2.3137 rb:1.0571 dl:1514-1521 gd:1 +ttp: b661/782 bl:2.3934 bb:1.0820 rl:2.3154 rb:1.0576 dl:1474-1480 gd:1 +ttp: b654/782 bl:2.2869 bb:1.0343 rl:2.3149 rb:1.0571 dl:1425-1432 gd:1 +ttp: b648/782 bl:2.2825 bb:1.0072 rl:2.3142 rb:1.0561 dl:1387-1392 gd:1 +ttp: b641/782 bl:2.2820 bb:1.0213 rl:2.3137 rb:1.0555 dl:1343-1349 gd:1 +ttp: b633/782 bl:2.2702 bb:1.0200 rl:2.3129 rb:1.0549 dl:1297-1302 gd:1 +ttp: b626/782 bl:2.3077 bb:1.0254 rl:2.3128 rb:1.0543 dl:1260-1265 gd:1 +ttp: b619/782 bl:2.3193 bb:1.0576 rl:2.3129 rb:1.0544 dl:1221-1226 gd:1 +ttp: b612/782 bl:2.2302 bb:1.0104 rl:2.3117 rb:1.0537 dl:1186-1190 gd:1 +ttp: b605/782 bl:2.2403 bb:1.0217 rl:2.3106 rb:1.0533 dl:1154-1159 gd:1 +ttp: b597/782 bl:2.3587 bb:1.0488 rl:2.3113 rb:1.0532 dl:1119-1124 gd:1 +ttp: b589/782 bl:2.2681 bb:1.0073 rl:2.3107 rb:1.0526 dl:1086-1089 gd:1 +ttp: b581/782 bl:2.3123 bb:1.0318 rl:2.3107 rb:1.0523 dl:1052-1056 gd:1 +ttp: b574/782 bl:2.3587 bb:1.0585 rl:2.3113 rb:1.0524 dl:1025-1029 gd:1 +ttp: b565/782 bl:2.3758 bb:1.0292 rl:2.3121 rb:1.0521 dl:993-997 gd:1 +ttp: b557/782 bl:2.3338 bb:1.0483 rl:2.3123 rb:1.0521 dl:965-968 gd:1 +ttp: b549/782 bl:2.2559 bb:1.0199 rl:2.3117 rb:1.0517 dl:939-943 gd:1 +ttp: b541/782 bl:2.3231 bb:1.0308 rl:2.3118 rb:1.0515 dl:915-918 gd:1 +ttp: b529/782 bl:2.3039 bb:1.0121 rl:2.3117 rb:1.0511 dl:878-882 gd:1 +ttp: b521/782 bl:2.3439 bb:1.0624 rl:2.3121 rb:1.0512 dl:854-858 gd:1 +ttp: b514/782 bl:2.3029 bb:1.0631 rl:2.3120 rb:1.0513 dl:835-838 gd:1 +ttp: b505/782 bl:2.3243 bb:1.0629 rl:2.3121 rb:1.0514 dl:809-812 gd:1 +ttp: b498/782 bl:2.3479 bb:1.0493 rl:2.3124 rb:1.0514 dl:791-794 gd:1 +ttp: b489/782 bl:2.3788 bb:1.0702 rl:2.3129 rb:1.0515 dl:769-771 gd:1 +ttp: b481/782 bl:2.2901 bb:1.0410 rl:2.3128 rb:1.0514 dl:749-752 gd:1 +ttp: b473/782 bl:2.2578 bb:1.0273 rl:2.3123 rb:1.0513 dl:730-733 gd:1 +ttp: b465/782 bl:2.3762 bb:1.0598 rl:2.3128 rb:1.0513 dl:712-714 gd:1 +ttp: b457/782 bl:2.2473 bb:1.0288 rl:2.3123 rb:1.0512 dl:695-697 gd:1 +ttp: b449/782 bl:2.4084 bb:1.0582 rl:2.3130 rb:1.0512 dl:678-680 gd:1 +ttp: b441/782 bl:2.3399 bb:1.0434 rl:2.3132 rb:1.0512 dl:662-664 gd:1 +ttp: b433/782 bl:2.2483 bb:1.0394 rl:2.3128 rb:1.0511 dl:645-647 gd:1 +ttp: b425/782 bl:2.3636 bb:1.0572 rl:2.3131 rb:1.0511 dl:630-632 gd:1 +ttp: b417/782 bl:2.2577 bb:1.0430 rl:2.3127 rb:1.0511 dl:615-617 gd:1 +ttp: b409/782 bl:2.3104 bb:1.0604 rl:2.3127 rb:1.0511 dl:598-601 gd:1 +ttp: b401/782 bl:2.2413 bb:1.0298 rl:2.3123 rb:1.0510 dl:584-586 gd:1 +ttp: b394/782 bl:2.2435 bb:0.9877 rl:2.3119 rb:1.0506 dl:571-573 gd:1 +ttp: b387/782 bl:2.3561 bb:1.0807 rl:2.3122 rb:1.0508 dl:559-561 gd:1 +ttp: b379/782 bl:2.4157 bb:1.0859 rl:2.3127 rb:1.0510 dl:545-547 gd:1 +ttp: b371/782 bl:2.2485 bb:1.0979 rl:2.3124 rb:1.0512 dl:532-533 gd:1 +ttp: b362/782 bl:2.3564 bb:1.0770 rl:2.3126 rb:1.0513 dl:517-518 gd:1 +ttp: b354/782 bl:2.2969 bb:1.0626 rl:2.3125 rb:1.0514 dl:503-504 gd:1 +ttp: b346/782 bl:2.3639 bb:1.0672 rl:2.3128 rb:1.0515 dl:491-492 gd:1 +ttp: b338/782 bl:2.3550 bb:1.0969 rl:2.3130 rb:1.0517 dl:478-480 gd:1 +ttp: b330/782 bl:2.2371 bb:1.0660 rl:2.3126 rb:1.0517 dl:466-468 gd:1 +ttp: b322/782 bl:2.3648 bb:1.0555 rl:2.3129 rb:1.0518 dl:455-457 gd:1 +ttp: b314/782 bl:2.2372 bb:1.0552 rl:2.3125 rb:1.0518 dl:442-444 gd:1 +ttp: b305/782 bl:2.3287 bb:1.0824 rl:2.3126 rb:1.0519 dl:429-430 gd:1 +ttp: b296/782 bl:2.3791 bb:1.0954 rl:2.3129 rb:1.0521 dl:415-417 gd:1 +ttp: b285/782 bl:2.3632 bb:1.0766 rl:2.3131 rb:1.0522 dl:399-400 gd:1 +ttp: b277/782 bl:2.2511 bb:1.0601 rl:2.3128 rb:1.0522 dl:388-389 gd:1 +ttp: b269/782 bl:2.3377 bb:1.1092 rl:2.3129 rb:1.0524 dl:378-379 gd:1 +ttp: b261/782 bl:2.4129 bb:1.1106 rl:2.3133 rb:1.0526 dl:367-369 gd:1 +ttp: b253/782 bl:2.3217 bb:1.1028 rl:2.3133 rb:1.0527 dl:357-358 gd:1 +ttp: b245/782 bl:2.3700 bb:1.1095 rl:2.3135 rb:1.0529 dl:347-349 gd:1 +ttp: b237/782 bl:2.3351 bb:1.0968 rl:2.3135 rb:1.0530 dl:337-338 gd:1 +ttp: b229/782 bl:2.3630 bb:1.0650 rl:2.3137 rb:1.0531 dl:328-329 gd:1 +ttp: b221/782 bl:2.4025 bb:1.1196 rl:2.3139 rb:1.0533 dl:318-320 gd:1 +ttp: b210/782 bl:2.2534 bb:1.0804 rl:2.3138 rb:1.0533 dl:306-307 gd:1 +ttp: b199/782 bl:2.4270 bb:1.1419 rl:2.3141 rb:1.0536 dl:295-296 gd:1 +ttp: b190/782 bl:2.3365 bb:1.0743 rl:2.3141 rb:1.0536 dl:284-285 gd:1 +ttp: b182/782 bl:2.3391 bb:1.1121 rl:2.3142 rb:1.0538 dl:276-277 gd:1 +ttp: b174/782 bl:2.4413 bb:1.1514 rl:2.3145 rb:1.0540 dl:268-269 gd:1 +ttp: b166/782 bl:2.4702 bb:1.1041 rl:2.3149 rb:1.0541 dl:260-262 gd:1 +ttp: b158/782 bl:2.3357 bb:1.1043 rl:2.3149 rb:1.0542 dl:253-254 gd:1 +ttp: b150/782 bl:2.3213 bb:1.1022 rl:2.3149 rb:1.0543 dl:245-246 gd:1 +ttp: b135/782 bl:2.4217 bb:1.1735 rl:2.3152 rb:1.0545 dl:231-232 gd:1 +ttp: b127/782 bl:2.4578 bb:1.1789 rl:2.3154 rb:1.0548 dl:223-224 gd:1 +ttp: b119/782 bl:2.3485 bb:1.1434 rl:2.3155 rb:1.0549 dl:216-217 gd:1 +ttp: b107/782 bl:2.4381 bb:1.1676 rl:2.3157 rb:1.0551 dl:205-206 gd:1 +ttp: b96/782 bl:2.4748 bb:1.2015 rl:2.3160 rb:1.0554 dl:195-196 gd:1 +ttp: b88/782 bl:2.4721 bb:1.1797 rl:2.3163 rb:1.0556 dl:188-189 gd:1 +ttp: b78/782 bl:2.5271 bb:1.1831 rl:2.3166 rb:1.0558 dl:179-180 gd:1 +ttp: b66/782 bl:2.6327 bb:1.2320 rl:2.3171 rb:1.0560 dl:169-169 gd:1 +ttp: b57/782 bl:2.4520 bb:1.1547 rl:2.3173 rb:1.0562 dl:160-161 gd:1 +ttp: b49/782 bl:2.4345 bb:1.1576 rl:2.3174 rb:1.0563 dl:152-153 gd:1 +ttp: b41/782 bl:2.5151 bb:1.2058 rl:2.3177 rb:1.0565 dl:144-145 gd:1 +ttp: b29/782 bl:2.6181 bb:1.2111 rl:2.3180 rb:1.0566 dl:132-133 gd:1 +ttp: b19/782 bl:2.6233 bb:1.2045 rl:2.3183 rb:1.0568 dl:121-122 gd:1 +ttp: b10/782 bl:2.6042 bb:1.1668 rl:2.3186 rb:1.0569 dl:107-109 gd:1 +ttp: b2/782 bl:2.8136 bb:1.2366 rl:2.3190 rb:1.0570 dl:83-89 gd:1 +quantized_ttt_phased val_loss:2.31389972 val_bpb:1.05736188 eval_time:565177ms +total_eval_time:565.2s diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed1234.log b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed1234.log new file mode 100644 index 0000000000..301235872d --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed1234.log @@ -0,0 +1,469 @@ +=== #2018 emulation (no Gated XSA) seed=1234 === +PHASED_TTT_NUM_PHASES=1 PHASED_TTT_PREFIX_DOCS=1000 NGRAM_INSIDE=1 RANK=80 +Thu Apr 30 22:22:19 UTC 2026 +W0430 22:22:21.125000 324483 torch/distributed/run.py:803] +W0430 22:22:21.125000 324483 torch/distributed/run.py:803] ***************************************** +W0430 22:22:21.125000 324483 torch/distributed/run.py:803] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0430 22:22:21.125000 324483 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + agree_add_boost: 0.5 + artifact_dir: + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + awq_lite_bits: 8 + awq_lite_enabled: True + awq_lite_group_size: 64 + awq_lite_group_top_k: 1 + beta1: 0.9 + beta2: 0.99 + caseops_enabled: True + compressor: pergroup + data_dir: /workspace/pg-data/datasets + datasets_dir: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 14.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2560 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.01 + gated_attn_quant_gate: True + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 4.0 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/8889a790-f49c-451e-bd2c-c43cd3696db2.txt + logit_softcap: 30.0 + loop_end: 5 + loop_start: 3 + lqer_asym_enabled: True + lqer_asym_group: 64 + lqer_enabled: True + lqer_factor_bits: 4 + lqer_gain_select: False + lqer_rank: 4 + lqer_scope: all + lqer_top_k: 1 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 11.5 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + ngram_hint_precompute_outside: False + ngram_tilt_enabled: True + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 1 + phased_ttt_prefix_docs: 1000 + qk_gain_init: 5.25 + 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: 8889a790-f49c-451e-bd2c-c43cd3696db2 + scalar_lr: 0.02 + seed: 1234 + skip_gates_enabled: True + smear_gate_enabled: True + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 0.5 + temperature_scale: 1.0 + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + token_boost: 2.625 + token_order: 16 + token_threshold: 0.8 + tokenizer_path: /workspace/pg-data/datasets/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/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: 2560 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_lr: 0.0001 + ttt_lora_rank: 80 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 2.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.85 + warmup_steps: 20 + within_boost: 0.75 + within_tau: 0.45 + word_boost: 0.75 + word_normalize: strip_punct_lower + word_order: 4 + word_tau: 0.65 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 47851520 +model_params:35945673 +gptq:reserving 4s, effective=596000ms +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: 9.0008 val_bpb: 4.1127 +1/20000 train_loss: 9.0017 train_time: 0.0m tok/s: 17390974 +2/20000 train_loss: 12.9342 train_time: 0.0m tok/s: 10610321 +3/20000 train_loss: 10.2136 train_time: 0.0m tok/s: 9165836 +4/20000 train_loss: 8.7214 train_time: 0.0m tok/s: 8636849 +5/20000 train_loss: 7.9542 train_time: 0.0m tok/s: 8366762 +500/20000 train_loss: 2.5687 train_time: 0.8m tok/s: 8203555 +1000/20000 train_loss: 2.8012 train_time: 1.6m tok/s: 8164083 +1500/20000 train_loss: 2.6164 train_time: 2.4m tok/s: 8151685 +2000/20000 train_loss: 2.6549 train_time: 3.2m tok/s: 8144868 +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: 2.5366 train_time: 4.3m tok/s: 7658717 +3000/20000 train_loss: 2.5503 train_time: 5.5m tok/s: 7205999 +3500/20000 train_loss: 2.5568 train_time: 6.6m tok/s: 6915944 +4000/20000 train_loss: 2.3998 train_time: 7.8m tok/s: 6702574 +4000/20000 val_loss: 2.4169 val_bpb: 1.1043 +4500/20000 train_loss: 2.2729 train_time: 9.0m tok/s: 6554795 +4898/20000 val_loss: 2.3502 val_bpb: 1.0739 +stopping_early: wallclock_cap train_time: 596094ms step: 4898/20000 +peak memory allocated: 41707 MiB reserved: 46952 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.32511351 val_bpb:1.06239355 eval_time:10019ms +Serialized model: 135418111 bytes +Code size (uncompressed): 182796 bytes +Code size (compressed): 35707 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 5.4s +Quantized weights: + gate_int8_row: blocks.attn.attn_gate_w + 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)+awqgrpint8+lqer_asym: 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, smear_gate.weight, smear_lambda, softcap_neg, softcap_pos +Serialize: per-group lrzip compression... +Serialize: per-group compression done in 92.3s +Serialized model quantized+pergroup: 15937113 bytes +Total submission size quantized+pergroup: 15972820 bytes +Deserialize: per-group lrzip decompression... +Deserialize: decompression done in 17.8s +diagnostic quantized val_loss:2.34315789 val_bpb:1.07063840 eval_time:12904ms +Deserialize: per-group lrzip decompression... +Deserialize: decompression done in 16.1s +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (96.9s) + +beginning TTT eval timer +ngram_tilt:hints total=47851520 gated=13023303 token_gate=628130 within_gate=9866847 word_gate=2891588 agree2plus=303177 +ngram_tilt:precompute_done elapsed=127.80s total_targets=47851520 +ttt_phased: total_docs:50000 prefix_docs:1000 suffix_docs:49000 num_phases:1 boundaries:[1000] +ttp: b781/782 bl:2.1401 bb:1.0471 rl:2.1401 rb:1.0471 dl:17258-30330 gd:0 +ttpp: phase:1/1 pd:1424 gd:1000 t:205.3s +tttg: c1/154 lr:0.001000 t:0.3s +tttg: c2/154 lr:0.001000 t:0.4s +tttg: c3/154 lr:0.001000 t:0.5s +tttg: c4/154 lr:0.000999 t:0.6s +tttg: c5/154 lr:0.000998 t:0.7s +tttg: c6/154 lr:0.000997 t:0.8s +tttg: c7/154 lr:0.000996 t:0.9s +tttg: c8/154 lr:0.000995 t:1.0s +tttg: c9/154 lr:0.000993 t:1.1s +tttg: c10/154 lr:0.000991 t:1.2s +tttg: c11/154 lr:0.000989 t:1.3s +tttg: c12/154 lr:0.000987 t:1.4s +tttg: c13/154 lr:0.000985 t:1.6s +tttg: c14/154 lr:0.000982 t:1.7s +tttg: c15/154 lr:0.000979 t:1.8s +tttg: c16/154 lr:0.000976 t:2.0s +tttg: c17/154 lr:0.000973 t:2.1s +tttg: c18/154 lr:0.000970 t:2.3s +tttg: c19/154 lr:0.000966 t:2.4s +tttg: c20/154 lr:0.000962 t:2.5s +tttg: c21/154 lr:0.000958 t:2.6s +tttg: c22/154 lr:0.000954 t:2.7s +tttg: c23/154 lr:0.000950 t:2.8s +tttg: c24/154 lr:0.000945 t:2.9s +tttg: c25/154 lr:0.000941 t:3.0s +tttg: c26/154 lr:0.000936 t:3.1s +tttg: c27/154 lr:0.000930 t:3.2s +tttg: c28/154 lr:0.000925 t:3.3s +tttg: c29/154 lr:0.000920 t:3.4s +tttg: c30/154 lr:0.000914 t:3.5s +tttg: c31/154 lr:0.000908 t:3.7s +tttg: c32/154 lr:0.000902 t:3.8s +tttg: c33/154 lr:0.000896 t:3.9s +tttg: c34/154 lr:0.000890 t:4.0s +tttg: c35/154 lr:0.000883 t:4.1s +tttg: c36/154 lr:0.000876 t:4.2s +tttg: c37/154 lr:0.000870 t:4.3s +tttg: c38/154 lr:0.000863 t:4.4s +tttg: c39/154 lr:0.000855 t:4.5s +tttg: c40/154 lr:0.000848 t:4.6s +tttg: c41/154 lr:0.000841 t:4.7s +tttg: c42/154 lr:0.000833 t:4.8s +tttg: c43/154 lr:0.000825 t:4.9s +tttg: c44/154 lr:0.000817 t:5.1s +tttg: c45/154 lr:0.000809 t:5.2s +tttg: c46/154 lr:0.000801 t:5.3s +tttg: c47/154 lr:0.000793 t:5.4s +tttg: c48/154 lr:0.000785 t:5.5s +tttg: c49/154 lr:0.000776 t:5.6s +tttg: c50/154 lr:0.000768 t:5.7s +tttg: c51/154 lr:0.000759 t:5.8s +tttg: c52/154 lr:0.000750 t:5.9s +tttg: c53/154 lr:0.000741 t:6.0s +tttg: c54/154 lr:0.000732 t:6.1s +tttg: c55/154 lr:0.000723 t:6.2s +tttg: c56/154 lr:0.000714 t:6.3s +tttg: c57/154 lr:0.000704 t:6.4s +tttg: c58/154 lr:0.000695 t:6.6s +tttg: c59/154 lr:0.000685 t:6.7s +tttg: c60/154 lr:0.000676 t:6.8s +tttg: c61/154 lr:0.000666 t:6.9s +tttg: c62/154 lr:0.000656 t:7.0s +tttg: c63/154 lr:0.000647 t:7.1s +tttg: c64/154 lr:0.000637 t:7.2s +tttg: c65/154 lr:0.000627 t:7.3s +tttg: c66/154 lr:0.000617 t:7.4s +tttg: c67/154 lr:0.000607 t:7.5s +tttg: c68/154 lr:0.000597 t:7.6s +tttg: c69/154 lr:0.000587 t:7.8s +tttg: c70/154 lr:0.000577 t:7.9s +tttg: c71/154 lr:0.000567 t:8.0s +tttg: c72/154 lr:0.000556 t:8.2s +tttg: c73/154 lr:0.000546 t:8.3s +tttg: c74/154 lr:0.000536 t:8.4s +tttg: c75/154 lr:0.000526 t:8.5s +tttg: c76/154 lr:0.000515 t:8.6s +tttg: c77/154 lr:0.000505 t:8.7s +tttg: c78/154 lr:0.000495 t:8.9s +tttg: c79/154 lr:0.000485 t:9.0s +tttg: c80/154 lr:0.000474 t:9.1s +tttg: c81/154 lr:0.000464 t:9.2s +tttg: c82/154 lr:0.000454 t:9.3s +tttg: c83/154 lr:0.000444 t:9.4s +tttg: c84/154 lr:0.000433 t:9.5s +tttg: c85/154 lr:0.000423 t:9.6s +tttg: c86/154 lr:0.000413 t:9.7s +tttg: c87/154 lr:0.000403 t:9.8s +tttg: c88/154 lr:0.000393 t:9.9s +tttg: c89/154 lr:0.000383 t:10.0s +tttg: c90/154 lr:0.000373 t:10.1s +tttg: c91/154 lr:0.000363 t:10.2s +tttg: c92/154 lr:0.000353 t:10.4s +tttg: c93/154 lr:0.000344 t:10.5s +tttg: c94/154 lr:0.000334 t:10.6s +tttg: c95/154 lr:0.000324 t:10.7s +tttg: c96/154 lr:0.000315 t:10.8s +tttg: c97/154 lr:0.000305 t:10.9s +tttg: c98/154 lr:0.000296 t:11.0s +tttg: c99/154 lr:0.000286 t:11.1s +tttg: c100/154 lr:0.000277 t:11.2s +tttg: c101/154 lr:0.000268 t:11.3s +tttg: c102/154 lr:0.000259 t:11.5s +tttg: c103/154 lr:0.000250 t:11.6s +tttg: c104/154 lr:0.000241 t:11.7s +tttg: c105/154 lr:0.000232 t:11.8s +tttg: c106/154 lr:0.000224 t:11.9s +tttg: c107/154 lr:0.000215 t:12.0s +tttg: c108/154 lr:0.000207 t:12.1s +tttg: c109/154 lr:0.000199 t:12.2s +tttg: c110/154 lr:0.000191 t:12.4s +tttg: c111/154 lr:0.000183 t:12.5s +tttg: c112/154 lr:0.000175 t:12.6s +tttg: c113/154 lr:0.000167 t:12.7s +tttg: c114/154 lr:0.000159 t:12.8s +tttg: c115/154 lr:0.000152 t:12.9s +tttg: c116/154 lr:0.000145 t:13.0s +tttg: c117/154 lr:0.000137 t:13.1s +tttg: c118/154 lr:0.000130 t:13.3s +tttg: c119/154 lr:0.000124 t:13.4s +tttg: c120/154 lr:0.000117 t:13.5s +tttg: c121/154 lr:0.000110 t:13.6s +tttg: c122/154 lr:0.000104 t:13.7s +tttg: c123/154 lr:0.000098 t:13.8s +tttg: c124/154 lr:0.000092 t:13.9s +tttg: c125/154 lr:0.000086 t:14.0s +tttg: c126/154 lr:0.000080 t:14.2s +tttg: c127/154 lr:0.000075 t:14.3s +tttg: c128/154 lr:0.000070 t:14.4s +tttg: c129/154 lr:0.000064 t:14.5s +tttg: c130/154 lr:0.000059 t:14.6s +tttg: c131/154 lr:0.000055 t:14.7s +tttg: c132/154 lr:0.000050 t:14.8s +tttg: c133/154 lr:0.000046 t:14.9s +tttg: c134/154 lr:0.000042 t:15.0s +tttg: c135/154 lr:0.000038 t:15.1s +tttg: c136/154 lr:0.000034 t:15.3s +tttg: c137/154 lr:0.000030 t:15.4s +tttg: c138/154 lr:0.000027 t:15.5s +tttg: c139/154 lr:0.000024 t:15.6s +tttg: c140/154 lr:0.000021 t:15.7s +tttg: c141/154 lr:0.000018 t:15.8s +tttg: c142/154 lr:0.000015 t:15.9s +tttg: c143/154 lr:0.000013 t:16.0s +tttg: c144/154 lr:0.000011 t:16.1s +tttg: c145/154 lr:0.000009 t:16.3s +tttg: c146/154 lr:0.000007 t:16.4s +tttg: c147/154 lr:0.000005 t:16.5s +tttg: c148/154 lr:0.000004 t:16.6s +tttg: c149/154 lr:0.000003 t:16.7s +tttg: c150/154 lr:0.000002 t:16.8s +tttg: c151/154 lr:0.000001 t:16.9s +tttg: c152/154 lr:0.000000 t:17.0s +tttg: c153/154 lr:0.000000 t:17.1s +ttpr: phase:1/1 t:224.6s +ttp: b752/782 bl:2.3199 bb:1.0664 rl:2.1632 rb:1.0497 dl:3222-3283 gd:1 +ttp: b751/782 bl:2.3017 bb:1.0304 rl:2.1787 rb:1.0474 dl:3150-3221 gd:1 +ttp: b748/782 bl:2.3096 bb:1.0779 rl:2.1912 rb:1.0504 dl:2992-3039 gd:1 +ttp: b745/782 bl:2.2273 bb:1.0197 rl:2.1942 rb:1.0477 dl:2842-2883 gd:1 +ttp: b742/782 bl:2.3159 bb:1.0427 rl:2.2032 rb:1.0473 dl:2730-2762 gd:1 +ttp: b737/782 bl:2.3081 bb:1.0376 rl:2.2100 rb:1.0467 dl:2550-2583 gd:1 +ttp: b734/782 bl:2.2559 bb:1.0263 rl:2.2127 rb:1.0454 dl:2469-2495 gd:1 +ttp: b731/782 bl:2.3340 bb:1.0410 rl:2.2192 rb:1.0452 dl:2377-2414 gd:1 +ttp: b725/782 bl:2.3115 bb:1.0398 rl:2.2237 rb:1.0449 dl:2232-2254 gd:1 +ttp: b721/782 bl:2.2998 bb:1.0213 rl:2.2270 rb:1.0438 dl:2144-2163 gd:1 +ttp: b711/782 bl:2.2803 bb:1.0196 rl:2.2291 rb:1.0428 dl:1966-1983 gd:1 +ttp: b702/782 bl:2.4235 bb:1.0799 rl:2.2359 rb:1.0442 dl:1847-1858 gd:1 +ttp: b696/782 bl:2.3044 bb:1.0494 rl:2.2381 rb:1.0444 dl:1779-1790 gd:1 +ttp: b688/782 bl:2.3960 bb:1.0727 rl:2.2429 rb:1.0453 dl:1696-1706 gd:1 +ttp: b680/782 bl:2.2754 bb:1.0247 rl:2.2438 rb:1.0447 dl:1618-1628 gd:1 +ttp: b675/782 bl:2.3619 bb:1.0563 rl:2.2470 rb:1.0450 dl:1578-1586 gd:1 +ttp: b664/782 bl:2.3336 bb:1.0241 rl:2.2491 rb:1.0444 dl:1493-1499 gd:1 +ttp: b657/782 bl:2.3002 bb:1.0454 rl:2.2503 rb:1.0445 dl:1445-1452 gd:1 +ttp: b649/782 bl:2.2788 bb:1.0131 rl:2.2509 rb:1.0438 dl:1392-1398 gd:1 +ttp: b641/782 bl:2.2847 bb:1.0225 rl:2.2516 rb:1.0433 dl:1343-1349 gd:1 +ttp: b632/782 bl:2.3470 bb:1.0326 rl:2.2534 rb:1.0431 dl:1290-1297 gd:1 +ttp: b625/782 bl:2.3973 bb:1.0460 rl:2.2561 rb:1.0431 dl:1255-1260 gd:1 +ttp: b615/782 bl:2.3145 bb:1.0451 rl:2.2571 rb:1.0432 dl:1200-1205 gd:1 +ttp: b606/782 bl:2.3547 bb:1.0640 rl:2.2587 rb:1.0435 dl:1159-1164 gd:1 +ttp: b598/782 bl:2.3514 bb:1.0636 rl:2.2602 rb:1.0438 dl:1124-1129 gd:1 +ttp: b590/782 bl:2.3041 bb:1.0557 rl:2.2609 rb:1.0440 dl:1089-1093 gd:1 +ttp: b582/782 bl:2.3443 bb:1.0297 rl:2.2621 rb:1.0438 dl:1056-1060 gd:1 +ttp: b573/782 bl:2.3616 bb:1.0646 rl:2.2634 rb:1.0441 dl:1021-1025 gd:1 +ttp: b565/782 bl:2.3764 bb:1.0295 rl:2.2649 rb:1.0439 dl:993-997 gd:1 +ttp: b557/782 bl:2.3355 bb:1.0491 rl:2.2658 rb:1.0440 dl:965-968 gd:1 +ttp: b549/782 bl:2.2545 bb:1.0193 rl:2.2657 rb:1.0437 dl:939-943 gd:1 +ttp: b541/782 bl:2.3258 bb:1.0320 rl:2.2664 rb:1.0435 dl:915-918 gd:1 +ttp: b533/782 bl:2.3671 bb:1.0649 rl:2.2675 rb:1.0438 dl:890-892 gd:1 +ttp: b525/782 bl:2.3450 bb:1.0164 rl:2.2684 rb:1.0434 dl:866-869 gd:1 +ttp: b517/782 bl:2.3480 bb:1.0246 rl:2.2692 rb:1.0432 dl:843-846 gd:1 +ttp: b509/782 bl:2.3519 bb:1.0326 rl:2.2700 rb:1.0431 dl:820-823 gd:1 +ttp: b501/782 bl:2.3780 bb:1.0506 rl:2.2711 rb:1.0432 dl:799-802 gd:1 +ttp: b493/782 bl:2.3585 bb:1.0411 rl:2.2719 rb:1.0432 dl:778-780 gd:1 +ttp: b485/782 bl:2.2842 bb:1.0290 rl:2.2720 rb:1.0431 dl:759-761 gd:1 +ttp: b477/782 bl:2.3971 bb:1.0323 rl:2.2731 rb:1.0430 dl:740-742 gd:1 +ttp: b469/782 bl:2.3220 bb:1.0211 rl:2.2735 rb:1.0428 dl:721-724 gd:1 +ttp: b461/782 bl:2.3715 bb:1.0375 rl:2.2743 rb:1.0427 dl:703-706 gd:1 +ttp: b453/782 bl:2.3284 bb:1.0520 rl:2.2747 rb:1.0428 dl:687-689 gd:1 +ttp: b446/782 bl:2.2862 bb:1.0746 rl:2.2748 rb:1.0430 dl:672-674 gd:1 +ttp: b439/782 bl:2.3163 bb:1.0335 rl:2.2751 rb:1.0430 dl:657-659 gd:1 +ttp: b431/782 bl:2.3649 bb:1.0491 rl:2.2758 rb:1.0430 dl:642-643 gd:1 +ttp: b423/782 bl:2.2997 bb:1.0493 rl:2.2760 rb:1.0430 dl:626-629 gd:1 +ttp: b414/782 bl:2.2004 bb:1.0074 rl:2.2755 rb:1.0428 dl:609-611 gd:1 +ttp: b406/782 bl:2.3094 bb:1.0635 rl:2.2757 rb:1.0429 dl:593-595 gd:1 +ttp: b397/782 bl:2.3504 bb:1.0424 rl:2.2761 rb:1.0429 dl:577-579 gd:1 +ttp: b386/782 bl:2.3324 bb:1.0953 rl:2.2765 rb:1.0432 dl:557-559 gd:1 +ttp: b378/782 bl:2.4207 bb:1.0504 rl:2.2773 rb:1.0433 dl:544-545 gd:1 +ttp: b368/782 bl:2.3592 bb:1.0988 rl:2.2778 rb:1.0436 dl:527-528 gd:1 +ttp: b362/782 bl:2.3598 bb:1.0785 rl:2.2783 rb:1.0438 dl:517-518 gd:1 +ttp: b354/782 bl:2.2967 bb:1.0625 rl:2.2784 rb:1.0439 dl:503-504 gd:1 +ttp: b346/782 bl:2.3640 bb:1.0673 rl:2.2788 rb:1.0440 dl:491-492 gd:1 +ttp: b336/782 bl:2.4070 bb:1.0847 rl:2.2794 rb:1.0442 dl:476-477 gd:1 +ttp: b328/782 bl:2.2714 bb:1.0096 rl:2.2794 rb:1.0441 dl:463-465 gd:1 +ttp: b320/782 bl:2.3393 bb:1.0817 rl:2.2797 rb:1.0442 dl:451-453 gd:1 +ttp: b312/782 bl:2.3007 bb:1.0479 rl:2.2798 rb:1.0442 dl:439-440 gd:1 +ttp: b304/782 bl:2.3421 bb:1.0742 rl:2.2801 rb:1.0444 dl:427-429 gd:1 +ttp: b296/782 bl:2.3749 bb:1.0935 rl:2.2805 rb:1.0446 dl:415-417 gd:1 +ttp: b288/782 bl:2.2261 bb:1.0132 rl:2.2802 rb:1.0445 dl:403-405 gd:1 +ttp: b279/782 bl:2.3070 bb:1.0901 rl:2.2803 rb:1.0446 dl:391-392 gd:1 +ttp: b271/782 bl:2.3644 bb:1.1199 rl:2.2807 rb:1.0449 dl:380-382 gd:1 +ttp: b257/782 bl:2.4373 bb:1.1087 rl:2.2812 rb:1.0451 dl:362-364 gd:1 +ttp: b248/782 bl:2.4539 bb:1.1843 rl:2.2819 rb:1.0456 dl:351-352 gd:1 +ttp: b240/782 bl:2.2895 bb:1.0509 rl:2.2819 rb:1.0456 dl:341-342 gd:1 +ttp: b230/782 bl:2.4496 bb:1.1495 rl:2.2824 rb:1.0460 dl:329-330 gd:1 +ttp: b217/782 bl:2.3665 bb:1.1298 rl:2.2827 rb:1.0462 dl:314-315 gd:1 +ttp: b209/782 bl:2.4024 bb:1.1237 rl:2.2831 rb:1.0465 dl:305-306 gd:1 +ttp: b200/782 bl:2.3583 bb:1.0903 rl:2.2833 rb:1.0466 dl:296-297 gd:1 +ttp: b189/782 bl:2.4085 bb:1.1363 rl:2.2836 rb:1.0468 dl:283-284 gd:1 +ttp: b178/782 bl:2.3312 bb:1.0905 rl:2.2838 rb:1.0469 dl:272-273 gd:1 +ttp: b169/782 bl:2.3584 bb:1.1084 rl:2.2840 rb:1.0471 dl:263-264 gd:1 +ttp: b158/782 bl:2.3306 bb:1.1019 rl:2.2841 rb:1.0472 dl:253-254 gd:1 +ttp: b146/782 bl:2.4513 bb:1.1712 rl:2.2845 rb:1.0475 dl:241-242 gd:1 +ttp: b136/782 bl:2.4221 bb:1.1389 rl:2.2848 rb:1.0477 dl:232-233 gd:1 +ttp: b126/782 bl:2.3845 bb:1.1359 rl:2.2850 rb:1.0479 dl:222-223 gd:1 +ttp: b116/782 bl:2.4729 bb:1.1228 rl:2.2854 rb:1.0481 dl:213-214 gd:1 +ttp: b104/782 bl:2.4830 bb:1.1722 rl:2.2858 rb:1.0483 dl:202-203 gd:1 +ttp: b94/782 bl:2.5620 bb:1.2106 rl:2.2863 rb:1.0486 dl:193-194 gd:1 +ttp: b82/782 bl:2.4945 bb:1.1873 rl:2.2867 rb:1.0488 dl:183-183 gd:1 +ttp: b70/782 bl:2.5064 bb:1.2214 rl:2.2870 rb:1.0491 dl:172-173 gd:1 +ttp: b60/782 bl:2.4586 bb:1.1817 rl:2.2873 rb:1.0493 dl:163-164 gd:1 +ttp: b48/782 bl:2.5178 bb:1.2139 rl:2.2876 rb:1.0495 dl:151-152 gd:1 +ttp: b38/782 bl:2.6081 bb:1.1962 rl:2.2881 rb:1.0497 dl:141-142 gd:1 +ttp: b24/782 bl:2.4546 bb:1.1577 rl:2.2883 rb:1.0498 dl:127-128 gd:1 +ttp: b13/782 bl:2.6713 bb:1.2102 rl:2.2887 rb:1.0500 dl:112-114 gd:1 +ttp: b1/782 bl:2.8385 bb:1.1816 rl:2.2891 rb:1.0501 dl:27-83 gd:1 +quantized_ttt_phased val_loss:2.31377272 val_bpb:1.05730384 eval_time:518761ms +total_eval_time:518.8s diff --git a/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed42.log b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed42.log new file mode 100644 index 0000000000..da3cb825f3 --- /dev/null +++ b/records/track_10min_16mb/2026-05-01_SP8192_InsiderNgram_V21_NoGatedXSA_1.0569/train_seed42.log @@ -0,0 +1,480 @@ +=== #2018 emulation (no Gated XSA) seed=42 === +PHASED_TTT_NUM_PHASES=1 PHASED_TTT_PREFIX_DOCS=1000 NGRAM_INSIDE=1 RANK=80 +Thu Apr 30 21:53:21 UTC 2026 +W0430 21:53:22.863000 220497 torch/distributed/run.py:803] +W0430 21:53:22.863000 220497 torch/distributed/run.py:803] ***************************************** +W0430 21:53:22.863000 220497 torch/distributed/run.py:803] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0430 21:53:22.863000 220497 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + agree_add_boost: 0.5 + artifact_dir: + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + awq_lite_bits: 8 + awq_lite_enabled: True + awq_lite_group_size: 64 + awq_lite_group_top_k: 1 + beta1: 0.9 + beta2: 0.99 + caseops_enabled: True + compressor: pergroup + data_dir: /workspace/pg-data/datasets + datasets_dir: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 14.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2560 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.01 + gated_attn_quant_gate: True + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 4.0 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/2e20dc35-3e5e-4ffe-aca7-bc7d6a73bcf8.txt + logit_softcap: 30.0 + loop_end: 5 + loop_start: 3 + lqer_asym_enabled: True + lqer_asym_group: 64 + lqer_enabled: True + lqer_factor_bits: 4 + lqer_gain_select: False + lqer_rank: 4 + lqer_scope: all + lqer_top_k: 1 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 11.5 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + ngram_hint_precompute_outside: False + ngram_tilt_enabled: True + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 1 + phased_ttt_prefix_docs: 1000 + qk_gain_init: 5.25 + 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: 2e20dc35-3e5e-4ffe-aca7-bc7d6a73bcf8 + scalar_lr: 0.02 + seed: 42 + skip_gates_enabled: True + smear_gate_enabled: True + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 0.5 + temperature_scale: 1.0 + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + token_boost: 2.625 + token_order: 16 + token_threshold: 0.8 + tokenizer_path: /workspace/pg-data/datasets/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/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: 2560 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_lr: 0.0001 + ttt_lora_rank: 80 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 2.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/pg-data/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.85 + warmup_steps: 20 + within_boost: 0.75 + within_tau: 0.45 + word_boost: 0.75 + word_normalize: strip_punct_lower + word_order: 4 + word_tau: 0.65 + world_size: 8 + xsa_last_n: 11 +train_shards: 80 +val_tokens: 47851520 +model_params:35945673 +gptq:reserving 4s, effective=596000ms +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: 9.0076 val_bpb: 4.1158 +1/20000 train_loss: 9.0087 train_time: 0.0m tok/s: 17059285 +2/20000 train_loss: 12.8327 train_time: 0.0m tok/s: 10520453 +3/20000 train_loss: 10.2110 train_time: 0.0m tok/s: 9237194 +4/20000 train_loss: 8.6788 train_time: 0.0m tok/s: 8764124 +5/20000 train_loss: 7.9461 train_time: 0.0m tok/s: 8498293 +500/20000 train_loss: 2.5613 train_time: 0.8m tok/s: 8225459 +1000/20000 train_loss: 2.7941 train_time: 1.6m tok/s: 8186385 +1500/20000 train_loss: 2.6137 train_time: 2.4m tok/s: 8174473 +2000/20000 train_loss: 2.6487 train_time: 3.2m tok/s: 8164977 +layer_loop:enabled step:2164 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: 2.5387 train_time: 4.3m tok/s: 7677597 +3000/20000 train_loss: 2.5476 train_time: 5.4m tok/s: 7223685 +3500/20000 train_loss: 2.5516 train_time: 6.6m tok/s: 6931243 +4000/20000 train_loss: 2.4005 train_time: 7.8m tok/s: 6716485 +4000/20000 val_loss: 2.4145 val_bpb: 1.1032 +4500/20000 train_loss: 2.2722 train_time: 9.0m tok/s: 6567790 +4906/20000 val_loss: 2.3473 val_bpb: 1.0725 +stopping_early: wallclock_cap train_time: 596058ms step: 4906/20000 +peak memory allocated: 41707 MiB reserved: 46952 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.32243712 val_bpb:1.06117064 eval_time:9438ms +Serialized model: 135418111 bytes +Code size (uncompressed): 182796 bytes +Code size (compressed): 35707 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 5.6s +Quantized weights: + gate_int8_row: blocks.attn.attn_gate_w + 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)+awqgrpint8+lqer_asym: 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, smear_gate.weight, smear_lambda, softcap_neg, softcap_pos +Serialize: per-group lrzip compression... +Serialize: per-group compression done in 102.9s +Serialized model quantized+pergroup: 15941325 bytes +Total submission size quantized+pergroup: 15977032 bytes +Deserialize: per-group lrzip decompression... +Deserialize: decompression done in 17.3s +diagnostic quantized val_loss:2.34016367 val_bpb:1.06927028 eval_time:12029ms +Deserialize: per-group lrzip decompression... +Deserialize: decompression done in 20.2s +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (132.4s) + +beginning TTT eval timer +ngram_tilt:hints total=47851520 gated=13023303 token_gate=628130 within_gate=9866847 word_gate=2891588 agree2plus=303177 +ngram_tilt:precompute_done elapsed=125.85s total_targets=47851520 +ttt_phased: total_docs:50000 prefix_docs:1000 suffix_docs:49000 num_phases:1 boundaries:[1000] +ttp: b775/782 bl:2.2668 bb:1.0538 rl:2.2668 rb:1.0538 dl:6892-7524 gd:0 +ttp: b774/782 bl:2.2751 bb:1.0593 rl:2.2708 rb:1.0564 dl:6447-6872 gd:0 +ttp: b770/782 bl:2.2778 bb:1.0753 rl:2.2728 rb:1.0617 dl:5311-5522 gd:0 +ttp: b765/782 bl:2.3093 bb:1.0800 rl:2.2796 rb:1.0651 dl:4393-4510 gd:0 +ttpp: phase:1/1 pd:1424 gd:1000 t:249.1s +tttg: c1/154 lr:0.001000 t:0.3s +tttg: c2/154 lr:0.001000 t:0.4s +tttg: c3/154 lr:0.001000 t:0.5s +tttg: c4/154 lr:0.000999 t:0.6s +tttg: c5/154 lr:0.000998 t:0.7s +tttg: c6/154 lr:0.000997 t:0.8s +tttg: c7/154 lr:0.000996 t:0.9s +tttg: c8/154 lr:0.000995 t:1.0s +tttg: c9/154 lr:0.000993 t:1.1s +tttg: c10/154 lr:0.000991 t:1.2s +tttg: c11/154 lr:0.000989 t:1.3s +tttg: c12/154 lr:0.000987 t:1.4s +tttg: c13/154 lr:0.000985 t:1.5s +tttg: c14/154 lr:0.000982 t:1.6s +tttg: c15/154 lr:0.000979 t:1.7s +tttg: c16/154 lr:0.000976 t:1.8s +tttg: c17/154 lr:0.000973 t:1.9s +tttg: c18/154 lr:0.000970 t:2.0s +tttg: c19/154 lr:0.000966 t:2.1s +tttg: c20/154 lr:0.000962 t:2.2s +tttg: c21/154 lr:0.000958 t:2.4s +tttg: c22/154 lr:0.000954 t:2.5s +tttg: c23/154 lr:0.000950 t:2.6s +tttg: c24/154 lr:0.000945 t:2.7s +tttg: c25/154 lr:0.000941 t:2.8s +tttg: c26/154 lr:0.000936 t:2.9s +tttg: c27/154 lr:0.000930 t:3.0s +tttg: c28/154 lr:0.000925 t:3.1s +tttg: c29/154 lr:0.000920 t:3.2s +tttg: c30/154 lr:0.000914 t:3.3s +tttg: c31/154 lr:0.000908 t:3.4s +tttg: c32/154 lr:0.000902 t:3.5s +tttg: c33/154 lr:0.000896 t:3.6s +tttg: c34/154 lr:0.000890 t:3.7s +tttg: c35/154 lr:0.000883 t:3.8s +tttg: c36/154 lr:0.000876 t:4.0s +tttg: c37/154 lr:0.000870 t:4.1s +tttg: c38/154 lr:0.000863 t:4.2s +tttg: c39/154 lr:0.000855 t:4.3s +tttg: c40/154 lr:0.000848 t:4.4s +tttg: c41/154 lr:0.000841 t:4.5s +tttg: c42/154 lr:0.000833 t:4.6s +tttg: c43/154 lr:0.000825 t:4.7s +tttg: c44/154 lr:0.000817 t:4.8s +tttg: c45/154 lr:0.000809 t:4.9s +tttg: c46/154 lr:0.000801 t:5.1s +tttg: c47/154 lr:0.000793 t:5.2s +tttg: c48/154 lr:0.000785 t:5.3s +tttg: c49/154 lr:0.000776 t:5.4s +tttg: c50/154 lr:0.000768 t:5.5s +tttg: c51/154 lr:0.000759 t:5.6s +tttg: c52/154 lr:0.000750 t:5.7s +tttg: c53/154 lr:0.000741 t:5.8s +tttg: c54/154 lr:0.000732 t:5.9s +tttg: c55/154 lr:0.000723 t:6.0s +tttg: c56/154 lr:0.000714 t:6.1s +tttg: c57/154 lr:0.000704 t:6.2s +tttg: c58/154 lr:0.000695 t:6.3s +tttg: c59/154 lr:0.000685 t:6.4s +tttg: c60/154 lr:0.000676 t:6.5s +tttg: c61/154 lr:0.000666 t:6.6s +tttg: c62/154 lr:0.000656 t:6.7s +tttg: c63/154 lr:0.000647 t:6.9s +tttg: c64/154 lr:0.000637 t:7.0s +tttg: c65/154 lr:0.000627 t:7.1s +tttg: c66/154 lr:0.000617 t:7.2s +tttg: c67/154 lr:0.000607 t:7.3s +tttg: c68/154 lr:0.000597 t:7.4s +tttg: c69/154 lr:0.000587 t:7.5s +tttg: c70/154 lr:0.000577 t:7.6s +tttg: c71/154 lr:0.000567 t:7.7s +tttg: c72/154 lr:0.000556 t:7.9s +tttg: c73/154 lr:0.000546 t:8.0s +tttg: c74/154 lr:0.000536 t:8.1s +tttg: c75/154 lr:0.000526 t:8.2s +tttg: c76/154 lr:0.000515 t:8.3s +tttg: c77/154 lr:0.000505 t:8.4s +tttg: c78/154 lr:0.000495 t:8.5s +tttg: c79/154 lr:0.000485 t:8.6s +tttg: c80/154 lr:0.000474 t:8.7s +tttg: c81/154 lr:0.000464 t:8.8s +tttg: c82/154 lr:0.000454 t:8.9s +tttg: c83/154 lr:0.000444 t:9.0s +tttg: c84/154 lr:0.000433 t:9.1s +tttg: c85/154 lr:0.000423 t:9.2s +tttg: c86/154 lr:0.000413 t:9.3s +tttg: c87/154 lr:0.000403 t:9.4s +tttg: c88/154 lr:0.000393 t:9.5s +tttg: c89/154 lr:0.000383 t:9.6s +tttg: c90/154 lr:0.000373 t:9.7s +tttg: c91/154 lr:0.000363 t:9.8s +tttg: c92/154 lr:0.000353 t:9.9s +tttg: c93/154 lr:0.000344 t:10.0s +tttg: c94/154 lr:0.000334 t:10.1s +tttg: c95/154 lr:0.000324 t:10.2s +tttg: c96/154 lr:0.000315 t:10.3s +tttg: c97/154 lr:0.000305 t:10.4s +tttg: c98/154 lr:0.000296 t:10.5s +tttg: c99/154 lr:0.000286 t:10.6s +tttg: c100/154 lr:0.000277 t:10.7s +tttg: c101/154 lr:0.000268 t:10.8s +tttg: c102/154 lr:0.000259 t:10.9s +tttg: c103/154 lr:0.000250 t:11.0s +tttg: c104/154 lr:0.000241 t:11.1s +tttg: c105/154 lr:0.000232 t:11.2s +tttg: c106/154 lr:0.000224 t:11.3s +tttg: c107/154 lr:0.000215 t:11.4s +tttg: c108/154 lr:0.000207 t:11.5s +tttg: c109/154 lr:0.000199 t:11.6s +tttg: c110/154 lr:0.000191 t:11.7s +tttg: c111/154 lr:0.000183 t:11.8s +tttg: c112/154 lr:0.000175 t:11.9s +tttg: c113/154 lr:0.000167 t:12.0s +tttg: c114/154 lr:0.000159 t:12.1s +tttg: c115/154 lr:0.000152 t:12.2s +tttg: c116/154 lr:0.000145 t:12.4s +tttg: c117/154 lr:0.000137 t:12.5s +tttg: c118/154 lr:0.000130 t:12.6s +tttg: c119/154 lr:0.000124 t:12.7s +tttg: c120/154 lr:0.000117 t:12.9s +tttg: c121/154 lr:0.000110 t:13.0s +tttg: c122/154 lr:0.000104 t:13.1s +tttg: c123/154 lr:0.000098 t:13.2s +tttg: c124/154 lr:0.000092 t:13.3s +tttg: c125/154 lr:0.000086 t:13.4s +tttg: c126/154 lr:0.000080 t:13.5s +tttg: c127/154 lr:0.000075 t:13.6s +tttg: c128/154 lr:0.000070 t:13.7s +tttg: c129/154 lr:0.000064 t:13.8s +tttg: c130/154 lr:0.000059 t:13.9s +tttg: c131/154 lr:0.000055 t:14.0s +tttg: c132/154 lr:0.000050 t:14.1s +tttg: c133/154 lr:0.000046 t:14.2s +tttg: c134/154 lr:0.000042 t:14.4s +tttg: c135/154 lr:0.000038 t:14.5s +tttg: c136/154 lr:0.000034 t:14.6s +tttg: c137/154 lr:0.000030 t:14.7s +tttg: c138/154 lr:0.000027 t:14.8s +tttg: c139/154 lr:0.000024 t:14.9s +tttg: c140/154 lr:0.000021 t:15.0s +tttg: c141/154 lr:0.000018 t:15.1s +tttg: c142/154 lr:0.000015 t:15.2s +tttg: c143/154 lr:0.000013 t:15.3s +tttg: c144/154 lr:0.000011 t:15.4s +tttg: c145/154 lr:0.000009 t:15.6s +tttg: c146/154 lr:0.000007 t:15.7s +tttg: c147/154 lr:0.000005 t:15.8s +tttg: c148/154 lr:0.000004 t:15.9s +tttg: c149/154 lr:0.000003 t:16.0s +tttg: c150/154 lr:0.000002 t:16.1s +tttg: c151/154 lr:0.000001 t:16.2s +tttg: c152/154 lr:0.000000 t:16.3s +tttg: c153/154 lr:0.000000 t:16.4s +ttpr: phase:1/1 t:267.7s +ttp: b752/782 bl:2.3179 bb:1.0655 rl:2.2842 rb:1.0652 dl:3222-3283 gd:1 +ttp: b751/782 bl:2.2983 bb:1.0288 rl:2.2857 rb:1.0612 dl:3150-3221 gd:1 +ttp: b748/782 bl:2.3084 bb:1.0773 rl:2.2878 rb:1.0626 dl:2992-3039 gd:1 +ttp: b745/782 bl:2.2259 bb:1.0190 rl:2.2829 rb:1.0591 dl:2842-2883 gd:1 +ttp: b742/782 bl:2.3141 bb:1.0419 rl:2.2851 rb:1.0579 dl:2730-2762 gd:1 +ttp: b738/782 bl:2.3038 bb:1.0432 rl:2.2863 rb:1.0569 dl:2583-2618 gd:1 +ttp: b734/782 bl:2.2517 bb:1.0244 rl:2.2843 rb:1.0551 dl:2469-2495 gd:1 +ttp: b732/782 bl:2.3617 bb:1.0876 rl:2.2884 rb:1.0568 dl:2416-2441 gd:1 +ttp: b729/782 bl:2.2959 bb:1.0725 rl:2.2887 rb:1.0575 dl:2325-2352 gd:1 +ttp: b727/782 bl:2.2575 bb:1.0404 rl:2.2873 rb:1.0567 dl:2277-2305 gd:1 +ttp: b722/782 bl:2.3416 bb:1.0493 rl:2.2895 rb:1.0564 dl:2163-2185 gd:1 +ttp: b720/782 bl:2.3461 bb:1.0611 rl:2.2917 rb:1.0566 dl:2125-2144 gd:1 +ttp: b716/782 bl:2.2440 bb:1.0370 rl:2.2900 rb:1.0559 dl:2054-2069 gd:1 +ttp: b711/782 bl:2.2785 bb:1.0188 rl:2.2896 rb:1.0546 dl:1966-1983 gd:1 +ttp: b707/782 bl:2.3508 bb:1.0446 rl:2.2915 rb:1.0543 dl:1910-1923 gd:1 +ttp: b703/782 bl:2.3327 bb:1.0262 rl:2.2928 rb:1.0534 dl:1859-1872 gd:1 +ttp: b699/782 bl:2.4094 bb:1.0518 rl:2.2960 rb:1.0534 dl:1814-1824 gd:1 +ttp: b695/782 bl:2.3319 bb:1.0755 rl:2.2970 rb:1.0540 dl:1769-1779 gd:1 +ttp: b689/782 bl:2.3836 bb:1.0732 rl:2.2992 rb:1.0545 dl:1706-1715 gd:1 +ttp: b683/782 bl:2.2644 bb:1.0541 rl:2.2983 rb:1.0545 dl:1646-1657 gd:1 +ttp: b676/782 bl:2.3283 bb:1.0473 rl:2.2990 rb:1.0543 dl:1586-1595 gd:1 +ttp: b668/782 bl:2.3235 bb:1.0623 rl:2.2995 rb:1.0545 dl:1521-1530 gd:1 +ttp: b660/782 bl:2.3678 bb:1.0467 rl:2.3009 rb:1.0543 dl:1466-1474 gd:1 +ttp: b652/782 bl:2.2401 bb:1.0183 rl:2.2997 rb:1.0536 dl:1411-1419 gd:1 +ttp: b644/782 bl:2.3574 bb:1.0466 rl:2.3008 rb:1.0535 dl:1362-1367 gd:1 +ttp: b636/782 bl:2.3732 bb:1.0636 rl:2.3020 rb:1.0537 dl:1314-1320 gd:1 +ttp: b628/782 bl:2.3139 bb:1.0267 rl:2.3022 rb:1.0532 dl:1271-1276 gd:1 +ttp: b621/782 bl:2.2851 bb:1.0435 rl:2.3019 rb:1.0531 dl:1231-1237 gd:1 +ttp: b613/782 bl:2.3273 bb:1.0362 rl:2.3023 rb:1.0528 dl:1190-1195 gd:1 +ttp: b605/782 bl:2.2363 bb:1.0199 rl:2.3013 rb:1.0524 dl:1154-1159 gd:1 +ttp: b597/782 bl:2.3595 bb:1.0492 rl:2.3021 rb:1.0523 dl:1119-1124 gd:1 +ttp: b589/782 bl:2.2674 bb:1.0070 rl:2.3017 rb:1.0517 dl:1086-1089 gd:1 +ttp: b581/782 bl:2.3097 bb:1.0307 rl:2.3018 rb:1.0515 dl:1052-1056 gd:1 +ttp: b573/782 bl:2.3583 bb:1.0631 rl:2.3024 rb:1.0516 dl:1021-1025 gd:1 +ttp: b565/782 bl:2.3751 bb:1.0289 rl:2.3033 rb:1.0513 dl:993-997 gd:1 +ttp: b557/782 bl:2.3344 bb:1.0486 rl:2.3036 rb:1.0513 dl:965-968 gd:1 +ttp: b550/782 bl:2.3569 bb:1.0545 rl:2.3041 rb:1.0514 dl:943-946 gd:1 +ttp: b542/782 bl:2.3144 bb:1.0335 rl:2.3042 rb:1.0512 dl:918-921 gd:1 +ttp: b534/782 bl:2.3146 bb:1.0367 rl:2.3043 rb:1.0510 dl:893-896 gd:1 +ttp: b526/782 bl:2.3146 bb:1.0202 rl:2.3044 rb:1.0507 dl:869-872 gd:1 +ttp: b518/782 bl:2.2337 bb:1.0054 rl:2.3038 rb:1.0503 dl:846-850 gd:1 +ttp: b510/782 bl:2.3729 bb:1.0690 rl:2.3044 rb:1.0505 dl:823-826 gd:1 +ttp: b502/782 bl:2.3035 bb:1.0207 rl:2.3044 rb:1.0502 dl:802-804 gd:1 +ttp: b493/782 bl:2.3536 bb:1.0389 rl:2.3048 rb:1.0501 dl:778-780 gd:1 +ttp: b483/782 bl:2.2478 bb:1.0255 rl:2.3043 rb:1.0499 dl:754-756 gd:1 +ttp: b474/782 bl:2.3389 bb:1.0709 rl:2.3046 rb:1.0501 dl:733-735 gd:1 +ttp: b464/782 bl:2.2619 bb:1.0136 rl:2.3043 rb:1.0498 dl:710-712 gd:1 +ttp: b456/782 bl:2.3451 bb:1.0388 rl:2.3046 rb:1.0498 dl:693-695 gd:1 +ttp: b448/782 bl:2.3037 bb:1.0043 rl:2.3046 rb:1.0494 dl:677-678 gd:1 +ttp: b440/782 bl:2.2287 bb:0.9811 rl:2.3041 rb:1.0490 dl:659-662 gd:1 +ttp: b432/782 bl:2.3279 bb:1.0347 rl:2.3042 rb:1.0489 dl:643-645 gd:1 +ttp: b424/782 bl:2.3366 bb:1.0595 rl:2.3044 rb:1.0489 dl:629-630 gd:1 +ttp: b416/782 bl:2.3662 bb:1.0404 rl:2.3048 rb:1.0489 dl:613-615 gd:1 +ttp: b408/782 bl:2.2912 bb:1.0653 rl:2.3047 rb:1.0490 dl:597-598 gd:1 +ttp: b400/782 bl:2.2962 bb:1.0332 rl:2.3047 rb:1.0489 dl:582-584 gd:1 +ttp: b392/782 bl:2.2452 bb:1.0327 rl:2.3043 rb:1.0488 dl:568-570 gd:1 +ttp: b384/782 bl:2.3379 bb:1.0517 rl:2.3045 rb:1.0488 dl:554-555 gd:1 +ttp: b376/782 bl:2.3102 bb:1.0361 rl:2.3046 rb:1.0488 dl:540-542 gd:1 +ttp: b368/782 bl:2.3578 bb:1.0981 rl:2.3048 rb:1.0490 dl:527-528 gd:1 +ttp: b360/782 bl:2.2986 bb:1.0754 rl:2.3048 rb:1.0491 dl:513-515 gd:1 +ttp: b352/782 bl:2.4083 bb:1.0898 rl:2.3053 rb:1.0493 dl:499-501 gd:1 +ttp: b344/782 bl:2.3739 bb:1.0580 rl:2.3056 rb:1.0493 dl:488-489 gd:1 +ttp: b336/782 bl:2.4005 bb:1.0818 rl:2.3060 rb:1.0495 dl:476-477 gd:1 +ttp: b328/782 bl:2.2736 bb:1.0106 rl:2.3059 rb:1.0493 dl:463-465 gd:1 +ttp: b320/782 bl:2.3362 bb:1.0803 rl:2.3060 rb:1.0494 dl:451-453 gd:1 +ttp: b311/782 bl:2.3349 bb:1.0762 rl:2.3061 rb:1.0496 dl:438-439 gd:1 +ttp: b297/782 bl:2.3905 bb:1.0801 rl:2.3064 rb:1.0497 dl:417-418 gd:1 +ttp: b289/782 bl:2.3144 bb:1.0764 rl:2.3065 rb:1.0498 dl:405-406 gd:1 +ttp: b281/782 bl:2.2767 bb:1.0793 rl:2.3063 rb:1.0499 dl:394-395 gd:1 +ttp: b273/782 bl:2.3232 bb:1.0709 rl:2.3064 rb:1.0499 dl:383-384 gd:1 +ttp: b265/782 bl:2.3598 bb:1.0980 rl:2.3066 rb:1.0501 dl:372-374 gd:1 +ttp: b257/782 bl:2.4386 bb:1.1093 rl:2.3070 rb:1.0503 dl:362-364 gd:1 +ttp: b249/782 bl:2.4263 bb:1.0929 rl:2.3074 rb:1.0504 dl:352-354 gd:1 +ttp: b241/782 bl:2.3323 bb:1.0839 rl:2.3075 rb:1.0505 dl:342-344 gd:1 +ttp: b233/782 bl:2.3527 bb:1.1241 rl:2.3076 rb:1.0507 dl:333-334 gd:1 +ttp: b225/782 bl:2.4246 bb:1.1101 rl:2.3079 rb:1.0509 dl:323-324 gd:1 +ttp: b217/782 bl:2.3648 bb:1.1290 rl:2.3081 rb:1.0511 dl:314-315 gd:1 +ttp: b204/782 bl:2.4489 bb:1.1490 rl:2.3085 rb:1.0514 dl:300-301 gd:1 +ttp: b194/782 bl:2.4328 bb:1.1145 rl:2.3088 rb:1.0515 dl:289-290 gd:1 +ttp: b186/782 bl:2.4034 bb:1.1234 rl:2.3090 rb:1.0517 dl:280-281 gd:1 +ttp: b178/782 bl:2.3355 bb:1.0925 rl:2.3091 rb:1.0518 dl:272-273 gd:1 +ttp: b170/782 bl:2.3653 bb:1.1216 rl:2.3092 rb:1.0519 dl:264-265 gd:1 +ttp: b162/782 bl:2.3971 bb:1.1161 rl:2.3094 rb:1.0521 dl:256-257 gd:1 +ttp: b154/782 bl:2.4655 bb:1.2024 rl:2.3097 rb:1.0524 dl:249-250 gd:1 +ttp: b140/782 bl:2.4206 bb:1.1302 rl:2.3100 rb:1.0525 dl:235-236 gd:1 +ttp: b131/782 bl:2.3910 bb:1.1545 rl:2.3101 rb:1.0527 dl:227-228 gd:1 +ttp: b123/782 bl:2.3859 bb:1.1601 rl:2.3103 rb:1.0529 dl:219-220 gd:1 +ttp: b115/782 bl:2.4473 bb:1.1582 rl:2.3105 rb:1.0531 dl:212-213 gd:1 +ttp: b106/782 bl:2.4273 bb:1.1684 rl:2.3107 rb:1.0533 dl:204-205 gd:1 +ttp: b92/782 bl:2.4259 bb:1.1542 rl:2.3109 rb:1.0534 dl:191-192 gd:1 +ttp: b84/782 bl:2.5047 bb:1.1910 rl:2.3112 rb:1.0536 dl:184-185 gd:1 +ttp: b76/782 bl:2.4791 bb:1.1644 rl:2.3115 rb:1.0538 dl:177-178 gd:1 +ttp: b68/782 bl:2.5045 bb:1.1689 rl:2.3117 rb:1.0540 dl:170-171 gd:1 +ttp: b57/782 bl:2.4524 bb:1.1548 rl:2.3119 rb:1.0541 dl:160-161 gd:1 +ttp: b46/782 bl:2.5303 bb:1.2082 rl:2.3122 rb:1.0543 dl:149-150 gd:1 +ttp: b37/782 bl:2.5650 bb:1.2090 rl:2.3125 rb:1.0545 dl:140-141 gd:1 +ttp: b29/782 bl:2.6126 bb:1.2086 rl:2.3128 rb:1.0546 dl:132-133 gd:1 +ttp: b16/782 bl:2.6251 bb:1.2579 rl:2.3131 rb:1.0548 dl:117-118 gd:1 +quantized_ttt_phased val_loss:2.31114224 val_bpb:1.05610182 eval_time:592197ms +total_eval_time:592.2s