diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/README.md b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/README.md new file mode 100644 index 0000000000..33f370c142 --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/README.md @@ -0,0 +1,179 @@ +# Record: PR #1854 neural stack, budget-compliant 3-seed reproduction — val_bpb 1.06777 (3-seed mean) + +**val_bpb: 1.06777** (3-seed mean, std 0.00106) | **15,951,074 bytes** (mean) | 8×H100 SXM, ≤600s train / ≤600s eval + +This submission is a **3-seed validated, eval-budget-compliant reproduction** of the neural-stack portion of [PR #1854](https://github.com/openai/parameter-golf/pull/1854) (ndokutovich), with `PHASED_TTT_PREFIX_DOCS` reduced from 2000 → 1500 to fit the 600s evaluation budget cleanly. **Beats merged-leaderboard SOTA PR #1493 (bigbag, 1.0810) by 0.01323 BPB** at ~13σ statistical significance. + +The reported `val_bpb` is the **standard token-level NLL → byte conversion** (`val_loss / log(2) · tokens/bytes`) for the post-quantization, post-Phased-TTT model. **No byte-level mixture is claimed** — see "Note on byte-PPM mixture" below. + +## Result (3 seeds, 8×H100 80GB SXM) + +| Seed | val_bpb | Total bytes | Eval time | +|------|------:|------:|------:| +| 42 | 1.06686 | 15,952,086 | 374.6s | +| 1337 | 1.06893 | 15,949,941 | 371.0s | +| 314 | 1.06752 | 15,951,195 | 327.7s | +| **Mean** | **1.06777** | **15,951,074** | **357.8s** | +| **Std** | **0.00106** | | | + +- **vs merged leaderboard PR #1493 @bigbag (1.0810)**: **−0.01323 BPB** (12.5× std clearance over the 0.005-nat threshold; p ≪ 0.0001) +- All 3 artifacts under 16,000,000 bytes (max 15,952,086, margin 47,914) +- All 3 eval times under 600s wallclock (max 374.6s, margin 225.4s) + +## What's new vs PR #1854 + +PR #1854's reported eval wallclock is **~700s** (per its own log breakdown: ttt_phased 516s + ppm_mix 116s + diagnostics 67s), which is over the 600s evaluation budget. This submission demonstrates that the same neural stack achieves **identical post-TTT val_bpb (~1.067)** while fitting cleanly under the 600s eval budget by reducing `PHASED_TTT_PREFIX_DOCS` from 2000 to 1500. Statistical evidence: 3-seed std of 0.00106 confirms the val_bpb is stable at this trim. + +This decoupling matters because the 600s eval budget is an explicit contest constraint (closed PRs in Issue #677 cite eval over-budget as grounds for rejection — see PR #503 closure). A budget-compliant 1.067 record is a more defensible record candidate than a slightly lower over-budget one. + +## Note on byte-PPM mixture (not claimed) + +`train_gpt.py` includes our exploratory multibin-λ refinement of the byte-level PPM-D mixer (a 4-tier graduated gate: `[(0.95, 0.02), (0.85, 0.10), (0.75, 0.40), (0.0, 1.0)]`). When run with `PPM_ENABLED=1`, it produces `mix_bpb ≈ 0.861` on the val byte stream — a 0.04 BPB delta below the `quantized_ttt_phased val_bpb`. + +We **do not claim `mix_bpb`** in this submission. The byte-PPM mixture relies on a per-byte spreading approximation `per_byte_logp = token_logp / n_bytes` whose normalization properties over the 256-byte alphabet are an open community question. Specifically: under the assumption that the NN's byte-level distribution is the geometric mean of the token's per-byte spread, the convex combination with PPM-D's normalized byte distribution is not unambiguously a Kraft-compliant prefix code. Until that interpretation is settled by the maintainers, we report only the standard `val_bpb` derived from token-level NLL, which has no such ambiguity. + +The multibin mixer code is left in `train_gpt.py` so this submission is a single self-contained reproducible artifact. Setting `PPM_ENABLED=0` in the reproduction command produces only the standard `val_bpb` line and skips the mixer entirely. + +## Issue #1017 four-condition compliance (for the standard val_bpb path) + +| Condition | How this submission satisfies it | +|---|---| +| **C1 Causality** | Standard sliding-window eval; each token scored from prefix only. Phased TTT is score-first per PR #1413's protocol. | +| **C2 Normalized** | The model's softmax over the SP8192 token vocabulary is a proper normalized distribution. The reported `val_bpb` is `(total token NLL) / log(2) / total bytes`, the standard token-level codelength normalized by byte count. | +| **C3 Score-before-update** | Phased TTT scores each chunk under `torch.no_grad()` before any optimizer step. | +| **C4 Single pass** | One left-to-right pass; no rescoring or oracle selection. | + +Additional compliance: +- **No SLOT, no token-level n-gram cache, no logit bias.** Inherited from PR #1854's stack. +- **No pre-quant TTT on val data.** Score-first phased TTT only (PR #1413 lineage). +- **No external network access** at eval time. Tokenizer unchanged from PR #1854's CaseOps SP8192. + +## Normalization proof (Eppie / mhuen / abaybektursun litmus) + +Issue #677 surfaced a class of invalid submissions where (a) the predictive distribution did not sum to 1 over the full token vocabulary Σ, or (b) the byte denominator was inflated by tokenizer encoding artifacts. The headline `val_bpb` reported here is clean on both axes. The explicit checks: + +**(1) Full-vocab normalization over Σ.** The headline number is produced by the standard token-level path in `eval_val()` and the score-first phased TTT path. Per token position `t`: + +```python +per_token_loss = F.cross_entropy(logits.reshape(-1, V), y.reshape(-1), reduction="none") +val_loss_sum += per_token_loss.to(torch.float64).sum() +``` + +`F.cross_entropy` is `−log softmax(logits)[y]` over the full V=8192 vocabulary. The softmax distribution sums to 1 by construction. There is no "score the correct token only" shortcut, no hash-bucket-only normalization, no eval-built n-gram cache contributing to the headline metric, and no `np.where(match, blend, p_model)` of the kind Eppie flagged in #913 / #933 / #944. The `_ppm_mixture_bpb` function is in the file but its output (`mix_bpb`) is **not** the reported number — see "Note on byte-PPM mixture" above. + +**(2) Byte denominator equals original UTF-8 bytes.** The CaseOps tokenizer transforms text by inserting private-use sentinels `..` (each 3 UTF-8 bytes) that mark capitalization. Naïve `len(piece.encode("utf-8"))` would charge those sentinel bytes and inflate the denominator, lowering `val_bpb` artificially — the same class of bug that closed PR #1184 (Scylla, "byte accounting bug in candidate.meta.npz", per author's own closing comment). + +This submission **bypasses** the inflated path. With `CASEOPS_ENABLED=1`, `eval_val` reads per-token original-byte counts from a precomputed sidecar (`fineweb_val_bytes_*.bin`, identical shard layout to `fineweb_val_*.bin`): + +```python +# train_gpt.py:2618-2626 +if val_data.caseops_enabled and val_data.val_bytes is not None: + sidecar_slice = val_data.val_bytes[raw_start + 1 : raw_end].to(...) + val_byte_count += sidecar_slice.to(torch.float64).sum() +``` + +The sidecar is built in `prepare_caseops_data.py:_byte_counts`. That function: + +1. iterates the transformed string char by char, +2. **skips** the four case-op sentinels, +3. accumulates UTF-8 byte length only of non-sentinel characters into a prefix sum `prefix[]`, +4. attributes `prefix[end] − prefix[start]` original bytes to each token piece's surface span. + +By telescoping, `Σ_i sidecar[i] = prefix[len(transformed)] = Σ_(non-sentinel char) utf8_bytes(char) = len(original_text.encode("utf-8"))`. BOS tokens are assigned 0 bytes (correct: they correspond to no original characters). The reported `val_bpb` is therefore `total_NLL_nats / (Σ_i sidecar[i] · ln 2)`, where the denominator is the actual UTF-8 byte length of the original validation text — not the inflated transformed-text byte length. + +**(3) Eppie's reconstruction litmus.** The 16 MB artifact (compressed model + `train_gpt.py` + `lossless_caps.py` + `prepare_caseops_data.py` + tokenizer model) plus the published validation token stream can reconstruct the original validation text byte-for-byte: `decode_lossless_caps_v2(sp.decode(val_tokens))` is the exact inverse of the `prepare_caseops_data.py` pipeline. The reported BPB therefore corresponds to a real arithmetic-coding rate over the original UTF-8 stream, not over a transformed/inflated stream and not over an under-normalized auxiliary distribution. + +**(4) One-line reviewer reproduction.** To verify the byte denominator independently, after running data prep: + +```python +import glob, numpy as np +def _shard(f): + n = int(np.fromfile(f, dtype=np.int32, count=256)[2]) + return np.fromfile(f, dtype=np.uint16, count=n, offset=256*4) +total_sidecar = sum(int(_shard(f).sum()) for f in sorted(glob.glob("data/.../fineweb_val_bytes_*.bin"))) +# total_sidecar should equal len(original_validation_text.encode("utf-8")) +``` + +## Compute & artifact compliance + +| Item | Value | Limit | Margin | +|---|--:|--:|--:| +| Training wallclock | 600s (cap-bound, all 3 seeds) | 600s | 0 (by cap) | +| Evaluation wallclock (max seed) | 374.6s | 600s | 225.4s | +| Artifact total bytes (max seed) | 15,952,086 | 16,000,000 | 47,914 | +| Code (uncompressed) | 161,565 | — | — | +| Code (pyminify + lzma) | 33,305 | — | — | +| Quantized model (int6 + brotli) | 15,918,781 | — | — | + +## Lineage and credits + +- **PR #549** (@abaybektursun) — score-first TTT framework +- **PR #1394** (@clarkkev) — SP8192 + multi-phase score-first TTT + GPTQ embeddings + SDClip +- **PR #1413** (@dexhunter) — legal score-first TTT with QK-Gain +- **PR #1493** (@bigbag) — 3-layer recurrence + parallel residuals + QK-Gain 5.25 (current merged leaderboard) +- **PR #1729** (@romeerp) — CaseOps bijective case transform +- **PR #1787** (@nprime06) — SparseAttnGate + PolarNS + MIN_LR + FusedCE +- **PR #1797** (@dexhunter) — Smear gate + LQER asymmetric (this submission's neural base) +- **PR #1854** (@ndokutovich) — PR #1797 base + PR #1835 PPM-D port (this submission's direct predecessor's neural stack) +- **PR #1835** (@anmarhindi) — original byte-level PPM-D mixture (we include a multibin-gate refinement in code, but do not claim its score) + +This submission's contribution is twofold: +1. **Eval-budget-compliant 3-seed reproduction** of PR #1854's neural stack (val_bpb 1.06777 mean, std 0.00106) with `PHASED_TTT_PREFIX_DOCS=1500`, fitting cleanly under the 600s eval cap. +2. **Multibin-λ refinement** of the PR #1835 PPM-D mixer (included in code, runs at `PPM_ENABLED=1`). We document its measured `mix_bpb` of ~0.861 on the val byte stream but do not claim it as the headline `val_bpb` due to the byte-spread normalization question. + +## Reproduction + +### Data prep (run once, ~30 min on CPU pod) + +```bash +python3 -c "from huggingface_hub import hf_hub_download; \ + hf_hub_download(repo_id='willdepueoai/parameter-golf', \ + filename='datasets/docs_selected.jsonl', \ + repo_type='dataset', local_dir='./hf_cache')" + +python3 prepare_caseops_data.py \ + --docs ./hf_cache/datasets/docs_selected.jsonl \ + --out ./data/datasets/fineweb10B_sp8192_caseops/datasets \ + --sp ./tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model \ + --workers 16 --max-docs 5000000 +``` + +### 3-seed training + eval (~$54 on RunPod 8×H100 SXM) + +```bash +DATA_PATH=./data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved +TOKENIZER_PATH=./tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + +for SEED in 42 1337 314; do + CASEOPS_ENABLED=1 \ + PHASED_TTT_PREFIX_DOCS=1500 PHASED_TTT_NUM_PHASES=3 \ + MATRIX_CLIP_SIGMAS=12.85 ATTN_CLIP_SIGMAS=13.0 MLP_CLIP_SIGMAS=12.0 \ + EMBED_BITS=7 EMBED_CLIP_SIGMAS=15.0 \ + MATRIX_LR=0.026 MIN_LR=0.1 \ + FUSED_CE_ENABLED=1 SPARSE_ATTN_GATE_ENABLED=1 \ + SMEAR_GATE_ENABLED=1 GATE_WINDOW=12 \ + LQER_ENABLED=1 LQER_RANK=4 LQER_TOP_K=3 LQER_FACTOR_BITS=4 \ + LQER_ASYM_ENABLED=1 LQER_ASYM_GROUP=64 \ + TTT_WARM_START_A=1 \ + GPTQ_RESERVE_SECONDS=0.5 GPTQ_CALIBRATION_BATCHES=16 \ + PPM_ENABLED=1 PPM_ORDER=5 PPM_SUBSET_TOKENS=4000000 \ + DATA_PATH="$DATA_PATH" TOKENIZER_PATH="$TOKENIZER_PATH" \ + SEED=$SEED \ + torchrun --standalone --nproc_per_node=8 train_gpt.py > train_seed${SEED}.log 2>&1 +done +``` + +The headline `val_bpb` for each run is logged as the `quantized_ttt_phased val_bpb:` field. The same logs also include the exploratory `mix_bpb` from the multibin mixer; that is not claimed. + +To reproduce **only** the headline `val_bpb` (skipping the mixer entirely), set `PPM_ENABLED=0` in the env block above. + +## Files + +- `README.md` — this file +- `submission.json` — metadata +- `train_gpt.py` — PR #1854's neural stack with the multibin mixer addition (claimed `val_bpb` is the standard, mixer-independent path) +- `lossless_caps.py` — verbatim from PR #1854 +- `prepare_caseops_data.py` — verbatim from PR #1854 +- `tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model` — verbatim +- `train_seed{42,1337,314}.log` — per-seed train+eval logs +- `final_model.int6.ptz` — quantized model artifact (best seed) diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/final_model.int6.ptz b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/final_model.int6.ptz new file mode 100644 index 0000000000..c628c04ca2 Binary files /dev/null and b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/final_model.int6.ptz differ diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/lossless_caps.py b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/lossless_caps.py new file mode 100644 index 0000000000..98e472f824 --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/lossless_caps.py @@ -0,0 +1,833 @@ +"""Lossless capitalization pre-encoding helpers. + +This module provides a narrow, reversible transform that only touches +ASCII capital letters `A-Z`. Each uppercase ASCII letter is rewritten as +``, where `sentinel` is a private-use Unicode +character that is escaped by doubling if it appears literally in the +input text. + +Example with the default sentinel `\\uE000`: + + "The NASA Launch" -> "\\uE000the \\uE000n\\uE000a\\uE000s\\uE000a \\uE000launch" + +The transform is intentionally simple for v1: + +- lowercase ASCII letters are unchanged +- uppercase ASCII letters become sentinel + lowercase letter +- non-ASCII characters are left untouched +- literal sentinel characters are escaped as sentinel + sentinel + +This makes the transform exactly invertible while allowing a downstream +tokenizer to reuse lowercase subwords across case variants. +""" + +from __future__ import annotations + +import json +from pathlib import Path +from typing import Callable, Iterable + +LOSSLESS_CAPS_V1 = "lossless_caps_v1" +LOSSLESS_CAPS_V2 = "lossless_caps_v2" +LOSSLESS_CAPS_V3 = "lossless_caps_v3" +LOSSLESS_CAPS_V4 = "lossless_caps_v4" +LOSSLESS_CAPS_V5 = "lossless_caps_v5" +LOSSLESS_CAPS_V6 = "lossless_caps_v6" +LOSSLESS_CAPS_V7 = "lossless_caps_v7" +LOSSLESS_CAPS_CASEOPS_V1 = "lossless_caps_caseops_v1" +IDENTITY = "identity" +DEFAULT_SENTINEL = "\uE000" +DEFAULT_V2_TITLE = "\uE001" +DEFAULT_V2_ALLCAPS = "\uE002" +DEFAULT_V2_CAPNEXT = "\uE003" +DEFAULT_V2_ESC = "\uE004" +DEFAULT_V5_TITLE_MIN_LEN = 7 +DEFAULT_V6_ALLCAPS_MIN_LEN = 3 +DEFAULT_V7_ALLCAPS_MIN_LEN = 4 + + +class LosslessCapsError(ValueError): + """Raised when a transformed string is malformed.""" + + +def _is_ascii_upper(ch: str) -> bool: + return "A" <= ch <= "Z" + + +def _is_ascii_lower(ch: str) -> bool: + return "a" <= ch <= "z" + + +def _is_ascii_alpha(ch: str) -> bool: + return _is_ascii_lower(ch) or _is_ascii_upper(ch) + + +def _validate_distinct_single_chars(*chars: str) -> None: + if any(len(ch) != 1 for ch in chars): + raise ValueError("all control characters must be exactly one character") + if len(set(chars)) != len(chars): + raise ValueError("control characters must be distinct") + + +def encode_lossless_caps_v1(text: str, *, sentinel: str = DEFAULT_SENTINEL) -> str: + """Encode ASCII capitals reversibly using a one-character sentinel.""" + if len(sentinel) != 1: + raise ValueError("sentinel must be exactly one character") + out: list[str] = [] + for ch in text: + if ch == sentinel: + out.append(sentinel) + out.append(sentinel) + elif _is_ascii_upper(ch): + out.append(sentinel) + out.append(ch.lower()) + else: + out.append(ch) + return "".join(out) + + +def decode_lossless_caps_v1(text: str, *, sentinel: str = DEFAULT_SENTINEL) -> str: + """Decode the `lossless_caps_v1` transform back to the original text.""" + if len(sentinel) != 1: + raise ValueError("sentinel must be exactly one character") + out: list[str] = [] + i = 0 + n = len(text) + while i < n: + ch = text[i] + if ch != sentinel: + out.append(ch) + i += 1 + continue + if i + 1 >= n: + raise LosslessCapsError("dangling capitalization sentinel at end of string") + nxt = text[i + 1] + if nxt == sentinel: + out.append(sentinel) + elif _is_ascii_lower(nxt): + out.append(nxt.upper()) + else: + raise LosslessCapsError( + f"invalid sentinel escape sequence {sentinel + nxt!r}; " + "expected doubled sentinel or sentinel + lowercase ASCII letter" + ) + i += 2 + return "".join(out) + + +def encode_lossless_caps_v2( + text: str, + *, + title: str = DEFAULT_V2_TITLE, + allcaps: str = DEFAULT_V2_ALLCAPS, + capnext: str = DEFAULT_V2_CAPNEXT, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Encode ASCII word capitalization with cheap word-level markers. + + Rules over maximal ASCII alphabetic runs: + - lowercase words stay unchanged + - TitleCase words become `title + lowercase(word)` + - ALLCAPS words become `allcaps + lowercase(word)` + - mixed-case words use: + - optional `title` when the first letter is uppercase + - `capnext + lowercase(letter)` for subsequent uppercase letters + - literal control characters are escaped as `esc + literal` + """ + _validate_distinct_single_chars(title, allcaps, capnext, esc) + controls = {title, allcaps, capnext, esc} + out: list[str] = [] + i = 0 + n = len(text) + while i < n: + ch = text[i] + if ch in controls: + out.append(esc) + out.append(ch) + i += 1 + continue + if not _is_ascii_alpha(ch): + out.append(ch) + i += 1 + continue + + j = i + 1 + while j < n and _is_ascii_alpha(text[j]): + j += 1 + word = text[i:j] + lower_word = word.lower() + + if word.islower(): + out.append(word) + elif len(word) >= 2 and word.isupper(): + out.append(allcaps) + out.append(lower_word) + elif _is_ascii_upper(word[0]) and word[1:].islower(): + out.append(title) + out.append(lower_word) + else: + if _is_ascii_upper(word[0]): + out.append(title) + out.append(lower_word[0]) + for orig_ch, lower_ch in zip(word[1:], lower_word[1:], strict=True): + if _is_ascii_upper(orig_ch): + out.append(capnext) + out.append(lower_ch) + i = j + return "".join(out) + + +def decode_lossless_caps_v2( + text: str, + *, + title: str = DEFAULT_V2_TITLE, + allcaps: str = DEFAULT_V2_ALLCAPS, + capnext: str = DEFAULT_V2_CAPNEXT, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Decode the `lossless_caps_v2` transform back to the original text.""" + _validate_distinct_single_chars(title, allcaps, capnext, esc) + out: list[str] = [] + pending_escape = False + pending_word_mode: str | None = None + active_allcaps = False + pending_capnext = False + in_ascii_word = False + + for ch in text: + if pending_escape: + if pending_word_mode is not None and not _is_ascii_alpha(ch): + raise LosslessCapsError("escaped control char cannot satisfy pending word capitalization mode") + out.append(ch) + pending_escape = False + if _is_ascii_alpha(ch): + in_ascii_word = True + else: + in_ascii_word = False + active_allcaps = False + continue + + if ch == esc: + pending_escape = True + continue + if ch == title: + if pending_word_mode is not None or in_ascii_word or pending_capnext: + raise LosslessCapsError("invalid title marker placement") + pending_word_mode = "title" + continue + if ch == allcaps: + if pending_word_mode is not None or in_ascii_word or pending_capnext: + raise LosslessCapsError("invalid allcaps marker placement") + pending_word_mode = "allcaps" + continue + if ch == capnext: + if pending_capnext: + raise LosslessCapsError("duplicate capnext marker") + pending_capnext = True + continue + + if _is_ascii_alpha(ch): + at_word_start = not in_ascii_word + if at_word_start: + if pending_word_mode == "allcaps": + out.append(ch.upper()) + active_allcaps = True + elif pending_word_mode == "title": + out.append(ch.upper()) + elif pending_capnext: + out.append(ch.upper()) + else: + out.append(ch) + pending_word_mode = None + pending_capnext = False + in_ascii_word = True + continue + + if pending_word_mode is not None: + raise LosslessCapsError("word capitalization marker leaked into the middle of a word") + if active_allcaps: + out.append(ch.upper()) + elif pending_capnext: + out.append(ch.upper()) + else: + out.append(ch) + pending_capnext = False + continue + + if pending_word_mode is not None or pending_capnext: + raise LosslessCapsError("capitalization marker not followed by an ASCII letter") + out.append(ch) + in_ascii_word = False + active_allcaps = False + + if pending_escape: + raise LosslessCapsError("dangling escape marker at end of string") + if pending_word_mode is not None or pending_capnext: + raise LosslessCapsError("dangling capitalization marker at end of string") + return "".join(out) + + +def encode_lossless_caps_v3( + text: str, + *, + title: str = DEFAULT_V2_TITLE, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Encode only common word-level capitalization patterns. + + Rules over maximal ASCII alphabetic runs: + - lowercase words stay unchanged + - TitleCase words become `title + lowercase(word)` + - ALLCAPS words become `allcaps + lowercase(word)` + - all other mixed-case words are left unchanged + - literal control characters are escaped as `esc + literal` + """ + _validate_distinct_single_chars(title, allcaps, esc) + controls = {title, allcaps, esc} + out: list[str] = [] + i = 0 + n = len(text) + while i < n: + ch = text[i] + if ch in controls: + out.append(esc) + out.append(ch) + i += 1 + continue + if not _is_ascii_alpha(ch): + out.append(ch) + i += 1 + continue + + j = i + 1 + while j < n and _is_ascii_alpha(text[j]): + j += 1 + word = text[i:j] + + if word.islower(): + out.append(word) + elif len(word) >= 2 and word.isupper(): + out.append(allcaps) + out.append(word.lower()) + elif _is_ascii_upper(word[0]) and word[1:].islower(): + out.append(title) + out.append(word.lower()) + else: + out.append(word) + i = j + return "".join(out) + + +def decode_lossless_caps_v3( + text: str, + *, + title: str = DEFAULT_V2_TITLE, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Decode the `lossless_caps_v3` transform back to the original text.""" + _validate_distinct_single_chars(title, allcaps, esc) + out: list[str] = [] + pending_escape = False + pending_word_mode: str | None = None + active_allcaps = False + in_ascii_word = False + + for ch in text: + if pending_escape: + if pending_word_mode is not None and not _is_ascii_alpha(ch): + raise LosslessCapsError("escaped control char cannot satisfy pending word capitalization mode") + out.append(ch) + pending_escape = False + if _is_ascii_alpha(ch): + in_ascii_word = True + else: + in_ascii_word = False + active_allcaps = False + continue + + if ch == esc: + pending_escape = True + continue + if ch == title: + if pending_word_mode is not None or in_ascii_word: + raise LosslessCapsError("invalid title marker placement") + pending_word_mode = "title" + continue + if ch == allcaps: + if pending_word_mode is not None or in_ascii_word: + raise LosslessCapsError("invalid allcaps marker placement") + pending_word_mode = "allcaps" + continue + + if _is_ascii_alpha(ch): + at_word_start = not in_ascii_word + if at_word_start: + if pending_word_mode == "allcaps": + out.append(ch.upper()) + active_allcaps = True + elif pending_word_mode == "title": + out.append(ch.upper()) + else: + out.append(ch) + pending_word_mode = None + in_ascii_word = True + continue + + if pending_word_mode is not None: + raise LosslessCapsError("word capitalization marker leaked into the middle of a word") + out.append(ch.upper() if active_allcaps else ch) + continue + + if pending_word_mode is not None: + raise LosslessCapsError("capitalization marker not followed by an ASCII letter") + out.append(ch) + in_ascii_word = False + active_allcaps = False + + if pending_escape: + raise LosslessCapsError("dangling escape marker at end of string") + if pending_word_mode is not None: + raise LosslessCapsError("dangling capitalization marker at end of string") + return "".join(out) + + +def encode_lossless_caps_v4( + text: str, + *, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Encode only ALLCAPS ASCII words, leaving all other case untouched.""" + _validate_distinct_single_chars(allcaps, esc) + controls = {allcaps, esc} + out: list[str] = [] + i = 0 + n = len(text) + while i < n: + ch = text[i] + if ch in controls: + out.append(esc) + out.append(ch) + i += 1 + continue + if not _is_ascii_alpha(ch): + out.append(ch) + i += 1 + continue + j = i + 1 + while j < n and _is_ascii_alpha(text[j]): + j += 1 + word = text[i:j] + if len(word) >= 2 and word.isupper(): + out.append(allcaps) + out.append(word.lower()) + else: + out.append(word) + i = j + return "".join(out) + + +def decode_lossless_caps_v4( + text: str, + *, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Decode the `lossless_caps_v4` transform back to the original text.""" + _validate_distinct_single_chars(allcaps, esc) + out: list[str] = [] + pending_escape = False + pending_allcaps = False + in_ascii_word = False + active_allcaps = False + + for ch in text: + if pending_escape: + if pending_allcaps and not _is_ascii_alpha(ch): + raise LosslessCapsError("escaped control char cannot satisfy pending allcaps mode") + out.append(ch) + pending_escape = False + if _is_ascii_alpha(ch): + in_ascii_word = True + else: + in_ascii_word = False + active_allcaps = False + continue + + if ch == esc: + pending_escape = True + continue + if ch == allcaps: + if pending_allcaps or in_ascii_word: + raise LosslessCapsError("invalid allcaps marker placement") + pending_allcaps = True + continue + + if _is_ascii_alpha(ch): + if not in_ascii_word: + active_allcaps = pending_allcaps + pending_allcaps = False + in_ascii_word = True + out.append(ch.upper() if active_allcaps else ch) + continue + + if pending_allcaps: + raise LosslessCapsError("allcaps marker not followed by an ASCII letter") + out.append(ch) + in_ascii_word = False + active_allcaps = False + + if pending_escape: + raise LosslessCapsError("dangling escape marker at end of string") + if pending_allcaps: + raise LosslessCapsError("dangling allcaps marker at end of string") + return "".join(out) + + +def encode_lossless_caps_v5( + text: str, + *, + title: str = DEFAULT_V2_TITLE, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, + title_min_len: int = DEFAULT_V5_TITLE_MIN_LEN, +) -> str: + """Encode ALLCAPS words and only sufficiently long TitleCase words.""" + _validate_distinct_single_chars(title, allcaps, esc) + controls = {title, allcaps, esc} + out: list[str] = [] + i = 0 + n = len(text) + while i < n: + ch = text[i] + if ch in controls: + out.append(esc) + out.append(ch) + i += 1 + continue + if not _is_ascii_alpha(ch): + out.append(ch) + i += 1 + continue + j = i + 1 + while j < n and _is_ascii_alpha(text[j]): + j += 1 + word = text[i:j] + if len(word) >= 2 and word.isupper(): + out.append(allcaps) + out.append(word.lower()) + elif len(word) >= title_min_len and _is_ascii_upper(word[0]) and word[1:].islower(): + out.append(title) + out.append(word.lower()) + else: + out.append(word) + i = j + return "".join(out) + + +def decode_lossless_caps_v5( + text: str, + *, + title: str = DEFAULT_V2_TITLE, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Decode the `lossless_caps_v5` transform back to the original text.""" + return decode_lossless_caps_v3(text, title=title, allcaps=allcaps, esc=esc) + + +def encode_lossless_caps_v6( + text: str, + *, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, + allcaps_min_len: int = DEFAULT_V6_ALLCAPS_MIN_LEN, +) -> str: + """Encode only ALLCAPS words with length >= allcaps_min_len.""" + _validate_distinct_single_chars(allcaps, esc) + controls = {allcaps, esc} + out: list[str] = [] + i = 0 + n = len(text) + while i < n: + ch = text[i] + if ch in controls: + out.append(esc) + out.append(ch) + i += 1 + continue + if not _is_ascii_alpha(ch): + out.append(ch) + i += 1 + continue + j = i + 1 + while j < n and _is_ascii_alpha(text[j]): + j += 1 + word = text[i:j] + if len(word) >= allcaps_min_len and word.isupper(): + out.append(allcaps) + out.append(word.lower()) + else: + out.append(word) + i = j + return "".join(out) + + +def decode_lossless_caps_v6( + text: str, + *, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Decode the `lossless_caps_v6` transform back to the original text.""" + return decode_lossless_caps_v4(text, allcaps=allcaps, esc=esc) + + +def encode_lossless_caps_v7( + text: str, + *, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, + allcaps_min_len: int = DEFAULT_V7_ALLCAPS_MIN_LEN, +) -> str: + """Encode only ALLCAPS words with length >= 4.""" + return encode_lossless_caps_v6( + text, + allcaps=allcaps, + esc=esc, + allcaps_min_len=allcaps_min_len, + ) + + +def decode_lossless_caps_v7( + text: str, + *, + allcaps: str = DEFAULT_V2_ALLCAPS, + esc: str = DEFAULT_V2_ESC, +) -> str: + """Decode the `lossless_caps_v7` transform back to the original text.""" + return decode_lossless_caps_v6(text, allcaps=allcaps, esc=esc) + + +def get_text_transform(name: str | None) -> Callable[[str], str]: + """Return the forward text transform for the given config name.""" + normalized = IDENTITY if name in {None, "", IDENTITY} else str(name) + if normalized == IDENTITY: + return lambda text: text + if normalized == LOSSLESS_CAPS_V1: + return encode_lossless_caps_v1 + if normalized == LOSSLESS_CAPS_V2: + return encode_lossless_caps_v2 + if normalized == LOSSLESS_CAPS_V3: + return encode_lossless_caps_v3 + if normalized == LOSSLESS_CAPS_V4: + return encode_lossless_caps_v4 + if normalized == LOSSLESS_CAPS_V5: + return encode_lossless_caps_v5 + if normalized == LOSSLESS_CAPS_V6: + return encode_lossless_caps_v6 + if normalized == LOSSLESS_CAPS_V7: + return encode_lossless_caps_v7 + if normalized == LOSSLESS_CAPS_CASEOPS_V1: + return encode_lossless_caps_v2 + raise ValueError(f"unsupported text_transform={name!r}") + + +def get_text_inverse_transform(name: str | None) -> Callable[[str], str]: + """Return the inverse transform for the given config name.""" + normalized = IDENTITY if name in {None, "", IDENTITY} else str(name) + if normalized == IDENTITY: + return lambda text: text + if normalized == LOSSLESS_CAPS_V1: + return decode_lossless_caps_v1 + if normalized == LOSSLESS_CAPS_V2: + return decode_lossless_caps_v2 + if normalized == LOSSLESS_CAPS_V3: + return decode_lossless_caps_v3 + if normalized == LOSSLESS_CAPS_V4: + return decode_lossless_caps_v4 + if normalized == LOSSLESS_CAPS_V5: + return decode_lossless_caps_v5 + if normalized == LOSSLESS_CAPS_V6: + return decode_lossless_caps_v6 + if normalized == LOSSLESS_CAPS_V7: + return decode_lossless_caps_v7 + if normalized == LOSSLESS_CAPS_CASEOPS_V1: + return decode_lossless_caps_v2 + raise ValueError(f"unsupported text_transform={name!r}") + + +def normalize_text_transform_name(name: str | None) -> str: + """Normalize empty/None transform names to the identity transform.""" + return IDENTITY if name in {None, "", IDENTITY} else str(name) + + +def get_text_transform_control_symbols(name: str | None) -> list[str]: + """Return reserved control symbols used by a transform, if any.""" + normalized = normalize_text_transform_name(name) + if normalized == IDENTITY: + return [] + if normalized == LOSSLESS_CAPS_V1: + return [DEFAULT_SENTINEL] + if normalized == LOSSLESS_CAPS_V2: + return [DEFAULT_V2_TITLE, DEFAULT_V2_ALLCAPS, DEFAULT_V2_CAPNEXT, DEFAULT_V2_ESC] + if normalized == LOSSLESS_CAPS_CASEOPS_V1: + return [DEFAULT_V2_TITLE, DEFAULT_V2_ALLCAPS, DEFAULT_V2_CAPNEXT, DEFAULT_V2_ESC] + if normalized in {LOSSLESS_CAPS_V3, LOSSLESS_CAPS_V5}: + return [DEFAULT_V2_TITLE, DEFAULT_V2_ALLCAPS, DEFAULT_V2_ESC] + if normalized in {LOSSLESS_CAPS_V4, LOSSLESS_CAPS_V6, LOSSLESS_CAPS_V7}: + return [DEFAULT_V2_ALLCAPS, DEFAULT_V2_ESC] + raise ValueError(f"unsupported text_transform={name!r}") + + +def infer_text_transform_from_manifest(tokenizer_path: str | Path) -> str: + """Best-effort lookup of a tokenizer's text transform from a local manifest.""" + tokenizer_path = Path(tokenizer_path).expanduser().resolve() + manifest_candidates = [ + tokenizer_path.parent.parent / "manifest.json", + tokenizer_path.parent / "manifest.json", + ] + for manifest_path in manifest_candidates: + if not manifest_path.is_file(): + continue + try: + payload = json.loads(manifest_path.read_text(encoding="utf-8")) + except (OSError, json.JSONDecodeError): + continue + tokenizers = payload.get("tokenizers") + if not isinstance(tokenizers, list): + continue + for tokenizer_meta in tokenizers: + if not isinstance(tokenizer_meta, dict): + continue + model_path = tokenizer_meta.get("model_path") or tokenizer_meta.get("path") + if not model_path: + continue + candidate = (manifest_path.parent / str(model_path)).resolve() + if candidate == tokenizer_path: + return normalize_text_transform_name(tokenizer_meta.get("text_transform")) + return IDENTITY + + +def surface_piece_original_byte_counts( + surfaces: Iterable[str], + *, + text_transform_name: str | None = None, + sentinel: str = DEFAULT_SENTINEL, +) -> list[int]: + """Return exact original UTF-8 byte counts contributed by each surface piece. + + `surfaces` must be the exact decoded text fragments emitted by SentencePiece + in order, e.g. `piece.surface` from `encode_as_immutable_proto`. + """ + normalized = normalize_text_transform_name(text_transform_name) + if normalized == IDENTITY: + return [len(surface.encode("utf-8")) for surface in surfaces] + if normalized == LOSSLESS_CAPS_V1: + if len(sentinel) != 1: + raise ValueError("sentinel must be exactly one character") + sentinel_bytes = len(sentinel.encode("utf-8")) + pending_sentinel = False + counts: list[int] = [] + for surface in surfaces: + piece_bytes = 0 + for ch in surface: + if pending_sentinel: + if ch == sentinel: + piece_bytes += sentinel_bytes + elif _is_ascii_lower(ch): + piece_bytes += 1 + else: + raise LosslessCapsError( + f"invalid continuation {ch!r} after capitalization sentinel" + ) + pending_sentinel = False + continue + if ch == sentinel: + pending_sentinel = True + else: + piece_bytes += len(ch.encode("utf-8")) + counts.append(piece_bytes) + if pending_sentinel: + raise LosslessCapsError("dangling capitalization sentinel across piece boundary") + return counts + if normalized not in {LOSSLESS_CAPS_V2, LOSSLESS_CAPS_V3, LOSSLESS_CAPS_V4, LOSSLESS_CAPS_V5, LOSSLESS_CAPS_V6, LOSSLESS_CAPS_V7, LOSSLESS_CAPS_CASEOPS_V1}: + raise ValueError(f"unsupported text_transform={text_transform_name!r}") + + title = DEFAULT_V2_TITLE + allcaps = DEFAULT_V2_ALLCAPS + capnext = DEFAULT_V2_CAPNEXT + esc = DEFAULT_V2_ESC + if normalized in {LOSSLESS_CAPS_V2, LOSSLESS_CAPS_CASEOPS_V1}: + _validate_distinct_single_chars(title, allcaps, capnext, esc) + elif normalized in {LOSSLESS_CAPS_V4, LOSSLESS_CAPS_V6, LOSSLESS_CAPS_V7}: + _validate_distinct_single_chars(allcaps, esc) + else: + _validate_distinct_single_chars(title, allcaps, esc) + pending_escape = False + pending_word_mode: str | None = None + active_allcaps = False + pending_capnext = False + in_ascii_word = False + counts: list[int] = [] + for surface in surfaces: + piece_bytes = 0 + for ch in surface: + if pending_escape: + if pending_word_mode is not None and not _is_ascii_alpha(ch): + raise LosslessCapsError("escaped control char cannot satisfy pending word capitalization mode") + piece_bytes += len(ch.encode("utf-8")) + pending_escape = False + if _is_ascii_alpha(ch): + in_ascii_word = True + else: + in_ascii_word = False + active_allcaps = False + continue + if ch == esc: + pending_escape = True + continue + if normalized in {LOSSLESS_CAPS_V2, LOSSLESS_CAPS_V3, LOSSLESS_CAPS_V5, LOSSLESS_CAPS_CASEOPS_V1} and ch == title: + if pending_word_mode is not None or in_ascii_word or pending_capnext: + raise LosslessCapsError("invalid title marker placement") + pending_word_mode = "title" + continue + if ch == allcaps: + if pending_word_mode is not None or in_ascii_word or pending_capnext: + raise LosslessCapsError("invalid allcaps marker placement") + pending_word_mode = "allcaps" + continue + if normalized in {LOSSLESS_CAPS_V2, LOSSLESS_CAPS_CASEOPS_V1} and ch == capnext: + if pending_capnext: + raise LosslessCapsError("duplicate capnext marker") + pending_capnext = True + continue + + if _is_ascii_alpha(ch): + at_word_start = not in_ascii_word + if at_word_start: + piece_bytes += 1 + active_allcaps = pending_word_mode == "allcaps" + pending_word_mode = None + pending_capnext = False + in_ascii_word = True + continue + if pending_word_mode is not None: + raise LosslessCapsError("word capitalization marker leaked into the middle of a word") + piece_bytes += 1 + pending_capnext = False + continue + + if pending_word_mode is not None or pending_capnext: + raise LosslessCapsError("capitalization marker not followed by an ASCII letter") + piece_bytes += len(ch.encode("utf-8")) + in_ascii_word = False + active_allcaps = False + counts.append(piece_bytes) + if pending_escape: + raise LosslessCapsError("dangling escape marker across piece boundary") + if pending_word_mode is not None or pending_capnext: + raise LosslessCapsError("dangling capitalization marker across piece boundary") + return counts diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/prepare_caseops_data.py b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/prepare_caseops_data.py new file mode 100644 index 0000000000..1c893ef988 --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/prepare_caseops_data.py @@ -0,0 +1,192 @@ +"""Parallel rebuild of prepare_caseops_data.py — multiprocessing tokenization. + +Bottleneck of the original is single-threaded SP encode + per-char byte +prefix-sum on val docs. With N workers via mp.Pool we get ~N× speedup. +On 28-vCPU pod, 16 workers cuts ~12h to ~45 min. + +Same CLI as prepare_caseops_data.py + extra --workers flag. +""" +import argparse +import json +import multiprocessing as mp +import os +import pathlib +import sys +from typing import Optional + +import numpy as np +import sentencepiece as spm + +sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) +from lossless_caps import ( + encode_lossless_caps_v2, + DEFAULT_V2_TITLE, + DEFAULT_V2_ALLCAPS, + DEFAULT_V2_CAPNEXT, + DEFAULT_V2_ESC, +) + +BOS_ID = 1 +SHARD_MAGIC = 20240520 +SHARD_VERSION = 1 +SHARD_TOKENS = 10_000_000 + +_LOSSLESS_V2_OPERATORS_CHARS: Optional[frozenset] = None +_worker_sp: Optional[spm.SentencePieceProcessor] = None + + +def _make_operators_set() -> frozenset: + return frozenset(( + DEFAULT_V2_TITLE, DEFAULT_V2_ALLCAPS, DEFAULT_V2_CAPNEXT, DEFAULT_V2_ESC, + )) + + +def _worker_init(sp_path: str) -> None: + global _worker_sp, _LOSSLESS_V2_OPERATORS_CHARS + _worker_sp = spm.SentencePieceProcessor(model_file=sp_path) + _LOSSLESS_V2_OPERATORS_CHARS = _make_operators_set() + + +def _byte_counts(transformed: str, piece_ids: list, pieces: list) -> np.ndarray: + n_chars = len(transformed) + prefix = np.zeros(n_chars + 1, dtype=np.int64) + running = 0 + ops = _LOSSLESS_V2_OPERATORS_CHARS + for idx, ch in enumerate(transformed): + if ch not in ops: + cp = ord(ch) + if cp < 0x80: + running += 1 + elif cp < 0x800: + running += 2 + elif cp < 0x10000: + running += 3 + else: + running += 4 + prefix[idx + 1] = running + counts = np.empty(len(piece_ids), dtype=np.uint16) + cursor_t = 0 + for i, piece in enumerate(pieces): + surface = piece.replace("▁", " ") + span_len = len(surface) + end = cursor_t + span_len + if end > n_chars: + end = n_chars + original_bytes = int(prefix[end] - prefix[cursor_t]) + cursor_t = end + counts[i] = max(0, min(65535, original_bytes)) + return counts + + +def _worker_process_doc(args: tuple) -> tuple: + """Worker: transform + tokenize one doc. Returns (doc_idx, token_ids, byte_counts_or_None).""" + doc_idx, text, is_val = args + sp = _worker_sp + transformed = encode_lossless_caps_v2(text) + piece_ids = sp.encode(transformed, out_type=int) + token_ids = [BOS_ID] + piece_ids + byte_counts = None + if is_val: + pieces = [sp.id_to_piece(int(pid)) for pid in piece_ids] + byte_counts = _byte_counts(transformed, piece_ids, pieces) + return doc_idx, token_ids, byte_counts + + +def _write_shard(path: pathlib.Path, arr: np.ndarray) -> None: + # 256 int32 header = 1024 bytes — matches kevclark/parameter-golf format + # and the load_data_shard expectations in train_gpt.py. + header = np.zeros(256, dtype=np.int32) + header[0] = SHARD_MAGIC + header[1] = SHARD_VERSION + header[2] = arr.size + with path.open("wb") as fh: + fh.write(header.tobytes()) + fh.write(arr.tobytes()) + + +def _iter_docs(docs_path: pathlib.Path): + with docs_path.open("r", encoding="utf-8") as fh: + for idx, line in enumerate(fh): + line = line.strip() + if not line: + continue + obj = json.loads(line) + yield idx, (obj["text"] if isinstance(obj, dict) else obj) + + +def main() -> None: + ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) + ap.add_argument("--docs", required=True, type=pathlib.Path) + ap.add_argument("--out", required=True, type=pathlib.Path) + ap.add_argument("--sp", required=True, type=pathlib.Path) + ap.add_argument("--val-docs", type=int, default=10_000) + ap.add_argument("--max-docs", type=int, default=0, help="Stop after N total docs (0 = process all). Limits disk usage.") + ap.add_argument("--workers", type=int, default=max(1, (os.cpu_count() or 8) - 4)) + ap.add_argument("--chunksize", type=int, default=64) + args = ap.parse_args() + + print(f"loading sp model: {args.sp}", flush=True) + sp_master = spm.SentencePieceProcessor(model_file=str(args.sp)) + print(f"loaded sp: vocab={sp_master.vocab_size()}", flush=True) + print(f"workers: {args.workers}", flush=True) + + train_out = args.out / "datasets" / "fineweb10B_sp8192_lossless_caps_caseops_v1_reserved" + train_out.mkdir(parents=True, exist_ok=True) + + val_buf_tokens: list = [] + val_buf_bytes: list = [] + train_buf: list = [] + val_written = 0 + train_written = 0 + n_docs = 0 + + def _build_args(): + for doc_idx, text in _iter_docs(args.docs): + if args.max_docs > 0 and doc_idx >= args.max_docs: + return + yield (doc_idx, text, doc_idx < args.val_docs) + + with mp.Pool(args.workers, initializer=_worker_init, initargs=(str(args.sp),)) as pool: + for doc_idx, token_ids, byte_counts in pool.imap( + _worker_process_doc, + _build_args(), + chunksize=args.chunksize, + ): + if doc_idx < args.val_docs: + val_buf_tokens.extend(token_ids) + val_buf_bytes.append(0) + val_buf_bytes.extend(int(b) for b in byte_counts[: len(token_ids) - 1]) + if len(val_buf_tokens) >= SHARD_TOKENS: + _write_shard(train_out / f"fineweb_val_{val_written:06d}.bin", + np.array(val_buf_tokens[:SHARD_TOKENS], dtype=np.uint16)) + _write_shard(train_out / f"fineweb_val_bytes_{val_written:06d}.bin", + np.array(val_buf_bytes[:SHARD_TOKENS], dtype=np.uint16)) + val_buf_tokens = val_buf_tokens[SHARD_TOKENS:] + val_buf_bytes = val_buf_bytes[SHARD_TOKENS:] + val_written += 1 + else: + train_buf.extend(token_ids) + if len(train_buf) >= SHARD_TOKENS: + _write_shard(train_out / f"fineweb_train_{train_written:06d}.bin", + np.array(train_buf[:SHARD_TOKENS], dtype=np.uint16)) + train_buf = train_buf[SHARD_TOKENS:] + train_written += 1 + n_docs += 1 + if n_docs % 10_000 == 0: + print(f" processed {n_docs} docs train_shards={train_written} val_shards={val_written}", flush=True) + + if val_buf_tokens: + _write_shard(train_out / f"fineweb_val_{val_written:06d}.bin", + np.array(val_buf_tokens, dtype=np.uint16)) + _write_shard(train_out / f"fineweb_val_bytes_{val_written:06d}.bin", + np.array(val_buf_bytes, dtype=np.uint16)) + if train_buf: + _write_shard(train_out / f"fineweb_train_{train_written:06d}.bin", + np.array(train_buf, dtype=np.uint16)) + + print(f"done. docs={n_docs} train_shards={train_written + (1 if train_buf else 0)} val_shards={val_written + (1 if val_buf_tokens else 0)}") + + +if __name__ == "__main__": + mp.set_start_method("spawn", force=True) + main() diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/submission.json b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/submission.json new file mode 100644 index 0000000000..ca8d5828e6 --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/submission.json @@ -0,0 +1,44 @@ +{ + "name": "Robert Walmsley", + "github_user": "robbiebusinessacc", + "val_bpb": 1.067773, + "val_bpb_std": 0.001062, + "seeds": [42, 314, 1337], + "per_seed_val_bpb": { + "42": 1.06686141, + "1337": 1.06893377, + "314": 1.06751568 + }, + "per_seed_eval_time_seconds": { + "42": 374.6, + "1337": 371.0, + "314": 327.7 + }, + "per_seed_total_artifact_bytes": { + "42": 15952086, + "1337": 15949941, + "314": 15951195 + }, + "track": "track_10min_16mb", + "base_pr": 1854, + "techniques": [ + "PR #1854 neural stack (PR #1797 base + Phased TTT score-first)", + "Budget-compliant trim: PHASED_TTT_PREFIX_DOCS=1500 (vs PR #1854's 2000) to fit 600s eval" + ], + "exploratory_not_claimed": [ + "Multibin-lambda PPM-D byte mixer (included in train_gpt.py; mix_bpb ~0.861) — not claimed as headline val_bpb due to open community question on byte-spread normalization vs Kraft compliance" + ], + "compliance": { + "track": "Track A/B (claim is on the standard token-level val_bpb, identical under both interpretations)", + "issue_1017_C1_causality": "Sliding-window eval scores each token from prefix only; Phased TTT is score-first", + "issue_1017_C2_normalized": "Standard val_bpb is total token NLL / log(2) / bytes; the underlying SP8192 softmax is a proper normalized token distribution", + "issue_1017_C3_score_before_update": "Phased TTT scores each chunk under no_grad before any optimizer step", + "issue_1017_C4_single_pass": "Single left-to-right pass over the val token stream", + "no_token_ngram_cache": true, + "no_slot": true, + "no_pre_quant_ttt": true, + "no_logit_bias": true, + "byte_ppm_mixture_claimed": false + }, + "delta_vs_merged_sota_pr1493": 0.013227 +} diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model new file mode 100644 index 0000000000..fffc8bb306 Binary files /dev/null and b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model differ diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_gpt.py b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_gpt.py new file mode 100644 index 0000000000..43d781f52b --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_gpt.py @@ -0,0 +1,3792 @@ +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)) + ttt_weight_decay = float(os.environ.get("TTT_WEIGHT_DECAY", 1.0)) + 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")) + # PPM-D byte mixture (port from #1835 anmarhindi 1.00136). Eval-time post- + # processing on already-scored token NLL. Score-first compliant: counts + # update AFTER per-byte mix log-prob is recorded. Same legality basis as + # PR #1145 normalized n-gram tilt (survived March 27 sweep). + ppm_enabled = bool(int(os.environ.get("PPM_ENABLED", "0"))) + ppm_order = int(os.environ.get("PPM_ORDER", 5)) + ppm_subset_tokens = int(os.environ.get("PPM_SUBSET_TOKENS", 5_000_000)) + ppm_lambda_hi = float(os.environ.get("PPM_LAMBDA_HI", 0.9)) + ppm_lambda_lo = float(os.environ.get("PPM_LAMBDA_LO", 0.05)) + ppm_conf_threshold = float(os.environ.get("PPM_CONF_THRESHOLD", 0.9)) + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + rank = int(os.environ.get("RANK", "0")) + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + is_main_process = rank == 0 + grad_accum_steps = 8 // world_size + # CaseOps integration: optional override of dataset root + tokenizer path. + # When CASEOPS_ENABLED=1, the wrapper loads a per-token byte sidecar + # (fineweb_val_bytes_*.bin, identical shard layout to val_*.bin) and uses + # it as the canonical raw-byte budget for BPB accounting. The sidecar + # REPLACES the build_sentencepiece_luts byte-counting path entirely. + caseops_enabled = bool(int(os.environ.get("CASEOPS_ENABLED", "0"))) + _default_caseops_data = os.path.join( + data_dir, + "datasets", + "fineweb10B_sp8192_caseops", + "datasets", + "datasets", + "fineweb10B_sp8192_lossless_caps_caseops_v1_reserved", + ) + _default_caseops_tok = os.path.join( + data_dir, + "datasets", + "fineweb10B_sp8192_caseops", + "datasets", + "tokenizers", + "fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model", + ) + if caseops_enabled: + datasets_dir = os.environ.get("DATA_PATH", _default_caseops_data) + tokenizer_path = os.environ.get("TOKENIZER_PATH", _default_caseops_tok) + else: + datasets_dir = os.environ.get( + "DATA_PATH", + os.path.join(data_dir, "datasets", f"fineweb10B_sp{vocab_size}"), + ) + tokenizer_path = os.environ.get( + "TOKENIZER_PATH", + os.path.join(data_dir, "tokenizers", f"fineweb_{vocab_size}_bpe.model"), + ) + train_files = os.path.join(datasets_dir, "fineweb_train_*.bin") + val_files = os.path.join(datasets_dir, "fineweb_val_*.bin") + val_bytes_files = os.path.join(datasets_dir, "fineweb_val_bytes_*.bin") + artifact_dir = os.environ.get("ARTIFACT_DIR", "") + logfile = ( + os.path.join(artifact_dir, f"{run_id}.txt") + if artifact_dir + else f"logs/{run_id}.txt" + ) + model_path = ( + os.path.join(artifact_dir, "final_model.pt") + if artifact_dir + else "final_model.pt" + ) + quantized_model_path = ( + os.path.join(artifact_dir, "final_model.int6.ptz") + if artifact_dir + else "final_model.int6.ptz" + ) + + +_logger_hparams = None + + +def set_logging_hparams(h): + global _logger_hparams + _logger_hparams = h + + +def log(msg, console=True): + if _logger_hparams is None: + print(msg) + return + if _logger_hparams.is_main_process: + if console: + print(msg) + if _logger_hparams.logfile is not None: + with open(_logger_hparams.logfile, "a", encoding="utf-8") as f: + print(msg, file=f) + + +class ValidationData: + def __init__(self, h, device): + self.sp = spm.SentencePieceProcessor(model_file=h.tokenizer_path) + if int(self.sp.vocab_size()) != h.vocab_size: + raise ValueError( + f"VOCAB_SIZE={h.vocab_size} does not match tokenizer vocab_size={int(self.sp.vocab_size())}" + ) + self.val_tokens = load_validation_tokens(h.val_files, h.eval_seq_len) + ( + self.base_bytes_lut, + self.has_leading_space_lut, + self.is_boundary_token_lut, + ) = build_sentencepiece_luts(self.sp, h.vocab_size, device) + # CaseOps: when enabled, load per-token byte sidecar and stash it as a + # CPU tensor aligned 1:1 with self.val_tokens. eval_val/eval_val_ttt + # branches use this as the canonical raw-byte budget per token. + self.caseops_enabled = bool(getattr(h, "caseops_enabled", False)) + self.val_bytes = None + if self.caseops_enabled: + self.val_bytes = load_validation_byte_sidecar( + h.val_bytes_files, h.eval_seq_len, self.val_tokens.numel() + ) + # PPM-D mixture needs per-token raw bytes to reconstruct the byte stream. + self.token_bytes_py = build_token_bytes_lut(self.sp, h.vocab_size) if bool(getattr(h, "ppm_enabled", False)) else None + + +def build_token_bytes_lut(sp, vocab_size): + """Per-token UTF-8 bytes (leading '▁' stripped). Used by PPM-D mixture + to reconstruct the byte stream from already-scored token NLLs.""" + sp_vocab_size = int(sp.vocab_size()) + table_size = max(sp_vocab_size, vocab_size) + out = [b""] * table_size + 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 + if sp.is_byte(token_id): + piece = sp.id_to_piece(token_id) + try: + out[token_id] = bytes([int(piece[3:-1], 16)]) + except Exception: + out[token_id] = b"" + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("▁"): + piece = piece[1:] + out[token_id] = piece.encode("utf-8") + return out + + +def build_sentencepiece_luts(sp, vocab_size, device): + sp_vocab_size = int(sp.vocab_size()) + assert ( + sp.piece_to_id("▁") != sp.unk_id() + ), "Tokenizer must have '▁' (space) as its own token for correct BPB byte counting" + table_size = max(sp_vocab_size, vocab_size) + base_bytes_np = np.zeros((table_size,), dtype=np.int16) + has_leading_space_np = np.zeros((table_size,), dtype=np.bool_) + is_boundary_token_np = np.ones((table_size,), dtype=np.bool_) + for token_id in range(sp_vocab_size): + if sp.is_control(token_id) or sp.is_unknown(token_id) or sp.is_unused(token_id): + continue + is_boundary_token_np[token_id] = False + if sp.is_byte(token_id): + base_bytes_np[token_id] = 1 + continue + piece = sp.id_to_piece(token_id) + if piece.startswith("▁"): + has_leading_space_np[token_id] = True + piece = piece[1:] + base_bytes_np[token_id] = len(piece.encode("utf-8")) + return ( + torch.tensor(base_bytes_np, dtype=torch.int16, device=device), + torch.tensor(has_leading_space_np, dtype=torch.bool, device=device), + torch.tensor(is_boundary_token_np, dtype=torch.bool, device=device), + ) + + +def load_validation_tokens(pattern, seq_len): + # Filter out CaseOps byte sidecar shards which share the val_*.bin glob. + files = [ + Path(p) + for p in sorted(glob.glob(pattern)) + if "_bytes_" not in Path(p).name + ] + if not files: + raise FileNotFoundError(f"No files found for pattern: {pattern}") + tokens = torch.cat([load_data_shard(file) for file in files]).contiguous() + usable = (tokens.numel() - 1) // seq_len * seq_len + if usable <= 0: + raise ValueError(f"Validation split is too short for TRAIN_SEQ_LEN={seq_len}") + return tokens[: usable + 1] + + +def load_validation_byte_sidecar(pattern, seq_len, expected_len): + """Load CaseOps per-token byte sidecar(s). Same shard layout as token shards + (256 int32 header + uint16 array). Each entry = canonical raw-text byte + budget for that token in the corresponding val shard. Returns a CPU + int16 tensor sliced to match expected_len (i.e. val_tokens length).""" + files = [Path(p) for p in sorted(glob.glob(pattern))] + if not files: + raise FileNotFoundError(f"No byte sidecar files for pattern: {pattern}") + shards = [load_data_shard(file) for file in files] + # load_data_shard returns uint16 — that's exactly what the sidecar stores. + bytes_full = torch.cat(shards).contiguous() + if bytes_full.numel() < expected_len: + raise ValueError( + f"Byte sidecar too short: {bytes_full.numel()} < val_tokens {expected_len}" + ) + return bytes_full[:expected_len].to(torch.int32) + + +def load_data_shard(file): + header_bytes = 256 * np.dtype(" 0: + pos = start + while pos < end: + seg_starts.append(pos) + pos += max_doc_len + else: + seg_starts.append(start) + boundaries = seg_starts + [total_len] + padded_len = get_next_multiple_of_n(len(boundaries), bucket_size) + cu = torch.full((padded_len,), total_len, dtype=torch.int32, device=device) + cu[: len(boundaries)] = torch.tensor(boundaries, dtype=torch.int32, device=device) + seg_ends = seg_starts[1:] + [total_len] + max_seqlen = max(end - start for start, end in zip(seg_starts, seg_ends)) + return cu, max_seqlen + +class DocumentPackingLoader: + _shard_pool = ThreadPoolExecutor(1) + + def __init__(self, h, device, cu_bucket_size=64): + self.rank = h.rank + self.world_size = h.world_size + self.device = device + self.cu_bucket_size = cu_bucket_size + self.max_seq_len = h.train_seq_len + all_files = [Path(p) for p in sorted(glob.glob(h.train_files))] + if not all_files: + raise FileNotFoundError(f"No files found for pattern: {h.train_files}") + self.files = all_files + self.file_iter = iter(self.files) + self._init_shard(load_data_shard(next(self.file_iter))) + self._next_shard = self._submit_next_shard() + self._batch_pool = ThreadPoolExecutor(1) + self._next_batch = None + + def _init_shard(self, tokens): + global BOS_ID + self.tokens = tokens + self.shard_size = tokens.numel() + if BOS_ID is None: + BOS_ID = 1 + self.bos_idx = ( + (tokens == BOS_ID).nonzero(as_tuple=True)[0].to(torch.int64).cpu().numpy() + ) + self.cursor = int(self.bos_idx[0]) + + def _submit_next_shard(self): + try: + path = next(self.file_iter) + return self._shard_pool.submit(load_data_shard, path) + except StopIteration: + return None + + def _advance_shard(self): + if self._next_shard is None: + self.file_iter = iter(self.files) + self._next_shard = self._shard_pool.submit( + load_data_shard, next(self.file_iter) + ) + self._init_shard(self._next_shard.result()) + self._next_shard = self._submit_next_shard() + + def _local_doc_starts(self, local_start, total_len): + lo = np.searchsorted(self.bos_idx, local_start, side="left") + hi = np.searchsorted(self.bos_idx, local_start + total_len, side="left") + return (self.bos_idx[lo:hi] - local_start).tolist() + + def _prepare_batch(self, num_tokens_local, max_seq_len): + per_rank_span = num_tokens_local + 1 + global_span = per_rank_span * self.world_size + while self.cursor + global_span > self.shard_size: + self._advance_shard() + local_start = self.cursor + self.rank * per_rank_span + buf = self.tokens[local_start : local_start + per_rank_span] + inputs = buf[:-1].to(dtype=torch.int64).pin_memory() + targets = buf[1:].to(dtype=torch.int64).pin_memory() + starts = self._local_doc_starts(local_start, inputs.numel()) + cu_seqlens, max_seqlen = _build_cu_seqlens( + starts, inputs.numel(), inputs.device, max_seq_len, self.cu_bucket_size + ) + cu_seqlens = cu_seqlens.pin_memory() + self.cursor += global_span + return inputs, targets, cu_seqlens, max_seqlen + + def next_batch(self, global_tokens, grad_accum_steps): + num_tokens_local = global_tokens // (self.world_size * grad_accum_steps) + if self._next_batch is not None: + inputs, targets, cu_seqlens, max_seqlen = self._next_batch.result() + else: + inputs, targets, cu_seqlens, max_seqlen = self._prepare_batch( + num_tokens_local, self.max_seq_len + ) + self._next_batch = self._batch_pool.submit( + self._prepare_batch, num_tokens_local, self.max_seq_len + ) + return ( + inputs[None].to(self.device, non_blocking=True), + targets[None].to(self.device, non_blocking=True), + cu_seqlens.to(self.device, non_blocking=True), + max_seqlen, + ) + + +class ShuffledSequenceLoader: + def __init__(self, h, device): + self.world_size = h.world_size + self.seq_len = h.train_seq_len + self.device = device + all_files = [Path(p) for p in sorted(glob.glob(h.train_files))] + if not all_files: + raise FileNotFoundError(f"No files found for pattern: {h.train_files}") + self.files = all_files[h.rank :: h.world_size] + self.rng = np.random.Generator(np.random.PCG64(h.rank)) + self.num_tokens = [_read_num_tokens(f) for f in self.files] + self.start_inds = [[] for _ in self.files] + for si in range(len(self.files)): + self._reset_shard(si) + + def _reset_shard(self, si): + max_phase = min( + self.seq_len - 1, max(0, self.num_tokens[si] - self.seq_len - 1) + ) + phase = int(self.rng.integers(max_phase + 1)) if max_phase > 0 else 0 + num_sequences = (self.num_tokens[si] - 1 - phase) // self.seq_len + sequence_order = self.rng.permutation(num_sequences) + self.start_inds[si] = (phase + sequence_order * self.seq_len).tolist() + + def next_batch(self, global_tokens, grad_accum_steps): + device_tokens = global_tokens // (self.world_size * grad_accum_steps) + device_batch_size = device_tokens // self.seq_len + remaining = np.array([len(s) for s in self.start_inds], dtype=np.float64) + x = torch.empty((device_batch_size, self.seq_len), dtype=torch.int64) + y = torch.empty((device_batch_size, self.seq_len), dtype=torch.int64) + for bi in range(device_batch_size): + total = remaining.sum() + if total <= 0: + for si in range(len(self.files)): + self._reset_shard(si) + remaining = np.array( + [len(s) for s in self.start_inds], dtype=np.float64 + ) + total = remaining.sum() + probs = remaining / total + si = int(self.rng.choice(len(self.files), p=probs)) + start_ind = self.start_inds[si].pop() + remaining[si] -= 1 + mm = _get_shard_memmap(self.files[si]) + window = torch.as_tensor( + np.array(mm[start_ind : start_ind + self.seq_len + 1], dtype=np.int64) + ) + x[bi] = window[:-1] + y[bi] = window[1:] + return x.to(self.device, non_blocking=True), y.to( + self.device, non_blocking=True + ) + + +class RMSNorm(nn.Module): + def __init__(self, eps=None): + super().__init__() + self.eps = eps + + def forward(self, x): + return F.rms_norm(x, (x.size(-1),), eps=self.eps) + + +class CastedLinear(nn.Linear): + def forward(self, x): + w = self.weight.to(x.dtype) + bias = self.bias.to(x.dtype) if self.bias is not None else None + return F.linear(x, w, bias) + + +@triton.jit +def linear_leaky_relu_square_kernel( + a_desc, + b_desc, + c_desc, + aux_desc, + M, + N, + K, + BLOCK_SIZE_M: tl.constexpr, + BLOCK_SIZE_N: tl.constexpr, + BLOCK_SIZE_K: tl.constexpr, + NUM_SMS: tl.constexpr, + FORWARD: tl.constexpr, +): + dtype = tl.bfloat16 + start_pid = tl.program_id(axis=0) + num_pid_m = tl.cdiv(M, BLOCK_SIZE_M) + num_pid_n = tl.cdiv(N, BLOCK_SIZE_N) + k_tiles = tl.cdiv(K, BLOCK_SIZE_K) + num_tiles = num_pid_m * num_pid_n + tile_id_c = start_pid - NUM_SMS + for tile_id in tl.range(start_pid, num_tiles, NUM_SMS, flatten=True): + pid_m = tile_id // num_pid_n + pid_n = tile_id % num_pid_n + offs_am = pid_m * BLOCK_SIZE_M + offs_bn = pid_n * BLOCK_SIZE_N + accumulator = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32) + for ki in range(k_tiles): + offs_k = ki * BLOCK_SIZE_K + a = a_desc.load([offs_am, offs_k]) + b = b_desc.load([offs_bn, offs_k]) + accumulator = tl.dot(a, b.T, accumulator) + tile_id_c += NUM_SMS + offs_am_c = offs_am + offs_bn_c = offs_bn + acc = tl.reshape(accumulator, (BLOCK_SIZE_M, 2, BLOCK_SIZE_N // 2)) + acc = tl.permute(acc, (0, 2, 1)) + acc0, acc1 = tl.split(acc) + c0 = acc0.to(dtype) + c1 = acc1.to(dtype) + if not FORWARD: + pre0 = aux_desc.load([offs_am_c, offs_bn_c]) + pre1 = aux_desc.load([offs_am_c, offs_bn_c + BLOCK_SIZE_N // 2]) + c0 = c0 * tl.where(pre0 > 0, 2.0 * pre0, 0.5 * pre0) + c1 = c1 * tl.where(pre1 > 0, 2.0 * pre1, 0.5 * pre1) + c_desc.store([offs_am_c, offs_bn_c], c0) + c_desc.store([offs_am_c, offs_bn_c + BLOCK_SIZE_N // 2], c1) + if FORWARD: + aux0 = tl.where(c0 > 0, c0, 0.5 * c0) + aux1 = tl.where(c1 > 0, c1, 0.5 * c1) + aux_desc.store([offs_am_c, offs_bn_c], aux0 * aux0) + aux_desc.store([offs_am_c, offs_bn_c + BLOCK_SIZE_N // 2], aux1 * aux1) + + +def linear_leaky_relu_square(a, b, aux=None): + M, K = a.shape + N, K2 = b.shape + assert K == K2 + c = torch.empty((M, N), device=a.device, dtype=a.dtype) + forward = aux is None + if aux is None: + aux = torch.empty((M, N), device=a.device, dtype=a.dtype) + num_sms = torch.cuda.get_device_properties(a.device).multi_processor_count + BLOCK_SIZE_M, BLOCK_SIZE_N, BLOCK_SIZE_K = 128, 256, 64 + num_stages = 4 if forward else 3 + a_desc = TensorDescriptor.from_tensor(a, [BLOCK_SIZE_M, BLOCK_SIZE_K]) + b_desc = TensorDescriptor.from_tensor(b, [BLOCK_SIZE_N, BLOCK_SIZE_K]) + c_desc = TensorDescriptor.from_tensor(c, [BLOCK_SIZE_M, BLOCK_SIZE_N // 2]) + aux_desc = TensorDescriptor.from_tensor(aux, [BLOCK_SIZE_M, BLOCK_SIZE_N // 2]) + grid = lambda _meta: ( + min(num_sms, triton.cdiv(M, BLOCK_SIZE_M) * triton.cdiv(N, BLOCK_SIZE_N)), + ) + linear_leaky_relu_square_kernel[grid]( + a_desc, + b_desc, + c_desc, + aux_desc, + M, + N, + K, + BLOCK_SIZE_M=BLOCK_SIZE_M, + BLOCK_SIZE_N=BLOCK_SIZE_N, + BLOCK_SIZE_K=BLOCK_SIZE_K, + NUM_SMS=num_sms, + FORWARD=forward, + num_stages=num_stages, + num_warps=8, + ) + if forward: + return c, aux + return c + + +class FusedLinearLeakyReLUSquareFunction(torch.autograd.Function): + @staticmethod + def forward(ctx, x, w1, w2): + x_flat = x.reshape(-1, x.shape[-1]) + pre, post = linear_leaky_relu_square(x_flat, w1) + out = F.linear(post, w2) + ctx.save_for_backward(x, w1, w2, pre, post) + return out.view(*x.shape[:-1], out.shape[-1]) + + @staticmethod + def backward(ctx, grad_output): + x, w1, w2, pre, post = ctx.saved_tensors + x_flat = x.reshape(-1, x.shape[-1]) + grad_output_flat = grad_output.reshape(-1, grad_output.shape[-1]) + dw2 = grad_output_flat.T @ post + dpre = linear_leaky_relu_square(grad_output_flat, w2.T.contiguous(), aux=pre) + dw1 = dpre.T @ x_flat + dx = dpre @ w1 + return dx.view_as(x), dw1, dw2 + + +FusedLeakyReLUSquareMLP = FusedLinearLeakyReLUSquareFunction.apply + + +class Rotary(nn.Module): + def __init__(self, dim, base=1e4, train_seq_len=1024, rope_dims=0, yarn=True): + super().__init__() + self.dim = dim + self.base = base + self.train_seq_len = train_seq_len + self.yarn = yarn + self.rope_dims = rope_dims if rope_dims > 0 else dim + inv_freq = 1.0 / base ** ( + torch.arange(0, self.rope_dims, 2, dtype=torch.float32) / self.rope_dims + ) + self.register_buffer("inv_freq", inv_freq, persistent=False) + self._seq_len_cached = 0 + self._cos_cached = None + self._sin_cached = None + + def forward(self, seq_len, device, dtype): + if ( + self._cos_cached is None + or self._sin_cached is None + or self._seq_len_cached < seq_len + or self._cos_cached.device != device + ): + rd = self.rope_dims + if self.yarn and seq_len > self.train_seq_len: + scale = seq_len / self.train_seq_len + new_base = self.base * scale ** (rd / (rd - 2)) + inv_freq = 1.0 / new_base ** ( + torch.arange(0, rd, 2, dtype=torch.float32, device=device) / rd + ) + else: + inv_freq = self.inv_freq.float().to(device) + t = torch.arange(seq_len, device=device, dtype=torch.float32) + freqs = torch.outer(t, inv_freq) + self._cos_cached = freqs.cos()[None, :, None, :] + self._sin_cached = freqs.sin()[None, :, None, :] + self._seq_len_cached = seq_len + return self._cos_cached[:, :seq_len].to(dtype=dtype), self._sin_cached[:, :seq_len].to(dtype=dtype) + + +def apply_rotary_emb(x, cos, sin, rope_dims=0): + if rope_dims > 0 and rope_dims < x.size(-1): + x_rope, x_pass = x[..., :rope_dims], x[..., rope_dims:] + half = rope_dims // 2 + x1, x2 = x_rope[..., :half], x_rope[..., half:] + x_rope = torch.cat((x1 * cos + x2 * sin, x1 * -sin + x2 * cos), dim=-1) + return torch.cat((x_rope, x_pass), dim=-1) + half = x.size(-1) // 2 + x1, x2 = x[..., :half], x[..., half:] + return torch.cat((x1 * cos + x2 * sin, x1 * -sin + x2 * cos), dim=-1) + + +class CausalSelfAttention(nn.Module): + def __init__( + self, dim, num_heads, num_kv_heads, rope_base, qk_gain_init, train_seq_len, yarn=True, + 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=0.5).square() + self._last_down_input = hidden.detach() if getattr(self, "_calib", False) else None + return F.linear(hidden, down_w.to(x.dtype)) + + +class Block(nn.Module): + def __init__( + self, + dim, + num_heads, + num_kv_heads, + mlp_mult, + rope_base, + qk_gain_init, + train_seq_len, + layer_idx=0, + ln_scale=False, + yarn=True, + 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)) + 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). Inline gate compute with .contiguous() on the slice fed + # to the projection so torch.compile fullgraph is happy. lam=0 + W=0 -> identity + # at init. This block runs unconditionally on the smear path; the cat keeps + # position 0 untouched so causality holds. + if self.smear_gate_enabled: + sl = self.smear_lambda.to(dtype=x.dtype) + gate_in = x[:, 1:, : self.smear_window].contiguous() + g = sl * torch.sigmoid(self.smear_gate(gate_in)) + x = torch.cat([x[:, :1], x[:, 1:] + g * x[:, :-1]], dim=1) + x = F.rms_norm(x, (x.size(-1),)) + x0 = x + skips = [] + 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 forward_logits(self, input_ids, cu_seqlens=None, max_seqlen=0): + hidden = self._forward_hidden(input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + logits_proj = self._project_logits(hidden) + return self.logit_softcap * torch.tanh(logits_proj / self.logit_softcap) + + def forward(self, input_ids, target_ids, cu_seqlens=None, max_seqlen=0): + hidden = self._forward_hidden(input_ids, cu_seqlens=cu_seqlens, max_seqlen=max_seqlen) + logits_proj = self._project_logits(hidden) + flat_targets = target_ids.reshape(-1) + # 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): + x = self.tok_emb(input_ids) + # SmearGate on the TTT path — same inline compute as forward_logits. + if self.smear_gate_enabled: + sl = self.smear_lambda.to(dtype=x.dtype) + gate_in = x[:, 1:, : self.smear_window].contiguous() + g = sl * torch.sigmoid(self.smear_gate(gate_in)) + x = torch.cat([x[:, :1], x[:, 1:] + g * x[:, :-1]], dim=1) + x = F.rms_norm(x, (x.size(-1),)) + x0 = x + skips = [] + enc_iter = ( + self.encoder_indices + if self.looping_active + else list(range(self.num_encoder_layers)) + ) + dec_iter = ( + self.decoder_indices + if self.looping_active + else list( + range( + self.num_encoder_layers, + self.num_encoder_layers + self.num_decoder_layers, + ) + ) + ) + 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) + logits = self.logit_softcap * torch.tanh(logits / self.logit_softcap) + bsz, sl, V = logits.shape + return F.cross_entropy( + logits.float().reshape(-1, V), target_ids.reshape(-1), reduction="none" + ).reshape(bsz, sl) + + def _block_with_lora(self, block, x, x0, lora, slot, q_w, k_w, v_w, out_w, up_w, down_w): + mix = block.resid_mix.to(dtype=x.dtype) + x_in = mix[0][None, None, :] * x + mix[1][None, None, :] * x0 + n = block.attn_norm(x_in) * block.ln_scale_factor + attn = block.attn + bsz, seqlen, dim = n.shape + # 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.bfloat16()) + if pg.shape[0] > m["B"]: + pg[m["B"] :].zero_() + fut = dist.reduce_scatter_tensor( + m["shard"], pg, op=dist.ReduceOp.AVG, async_op=True + ) + self._rs_futures.append(fut) + + @torch.no_grad() + def step(self, closure=None): + loss = None + if closure is not None: + with torch.enable_grad(): + loss = closure() + if not self._built: + self._build() + for group in self.param_groups: + lr = group["lr"] + momentum = group["momentum"] + backend_steps = group["backend_steps"] + nesterov = group["nesterov"] + wd = group.get("weight_decay", 0.0) + row_normalize = group.get("row_normalize", False) + prev_ag_handle = None + prev_m = None + sharded = self._distributed and hasattr(self, "_rs_futures") + for idx, m in enumerate(self._bank_meta): + p = m["p"] + if p.grad is None: + continue + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m["p"] + upd = prev_m["full_update"][: prev_m["B"]] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m["scale"]) + if sharded and self._rs_futures[idx] is not None: + self._rs_futures[idx].wait() + g = m["shard"] + buf = m["shard_mom"] + else: + g = p.grad.bfloat16() + state = self.state[p] + if "momentum_buffer" not in state: + state["momentum_buffer"] = torch.zeros_like(g) + buf = state["momentum_buffer"] + buf.mul_(momentum).add_(g) + if nesterov: + update = g.add(buf, alpha=momentum) + else: + update = buf + if row_normalize: + rn = update.float().norm(dim=-1, keepdim=True).clamp_min(1e-07) + update = update / rn.to(update.dtype) + update = zeropower_via_newtonschulz5(update, steps=backend_steps) + if sharded: + prev_ag_handle = dist.all_gather_into_tensor( + m["full_update"], update, async_op=True + ) + prev_m = m + else: + if wd > 0.0: + p.data.mul_(1.0 - lr * wd) + p.add_(update.to(dtype=p.dtype), alpha=-lr * m["scale"]) + if prev_ag_handle is not None: + prev_ag_handle.wait() + pp = prev_m["p"] + upd = prev_m["full_update"][: prev_m["B"]] + if wd > 0.0: + pp.data.mul_(1.0 - lr * wd) + pp.add_(upd.to(dtype=pp.dtype), alpha=-lr * prev_m["scale"]) + if hasattr(self, "_rs_futures"): + del self._rs_futures + return loss + + +CONTROL_TENSOR_NAME_PATTERNS = tuple( + pattern + for pattern in os.environ.get( + "CONTROL_TENSOR_NAME_PATTERNS", + "attn_scale,attn_scales,mlp_scale,mlp_scales,resid_mix,resid_mixes,q_gain,skip_weight,skip_weights,skip_gates,parallel_post_lambdas,parallel_resid_lambdas,attn_gate_proj,attn_gate_w,smear_gate,smear_lambda", + ).split(",") + if pattern +) + + +PACKED_REPLICATED_GRAD_MAX_NUMEL = 1 << 15 + + +class Optimizers: + def __init__(self, h, base_model): + matrix_params = [ + base_model.qo_bank, + base_model.kv_bank, + base_model.mlp_up_bank, + base_model.mlp_down_bank, + ] + block_named_params = list(base_model.blocks.named_parameters()) + scalar_params = [ + p + for (name, p) in block_named_params + if p.ndim < 2 + or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ] + if base_model.skip_weights.numel() > 0: + scalar_params.append(base_model.skip_weights) + if base_model.skip_gates is not None and base_model.skip_gates.numel() > 0: + scalar_params.append(base_model.skip_gates) + if base_model.parallel_post_lambdas is not None: + scalar_params.append(base_model.parallel_post_lambdas) + if base_model.parallel_resid_lambdas is not None: + scalar_params.append(base_model.parallel_resid_lambdas) + # SmearGate params live on GPT root (not in .blocks), so add them by hand. + # Both are tiny (gate_window scalars + 1 lambda). Optimized via scalar Adam. + if getattr(base_model, "smear_gate_enabled", False): + scalar_params.append(base_model.smear_gate.weight) + scalar_params.append(base_model.smear_lambda) + token_lr = h.tied_embed_lr if h.tie_embeddings else h.embed_lr + tok_params = [ + {"params": [base_model.tok_emb.weight], "lr": token_lr, "base_lr": token_lr} + ] + self.optimizer_tok = torch.optim.AdamW( + tok_params, + betas=(h.beta1, h.beta2), + eps=h.adam_eps, + weight_decay=h.embed_wd, + fused=True, + ) + self.optimizer_muon = Muon( + matrix_params, + lr=h.matrix_lr, + momentum=h.muon_momentum, + backend_steps=h.muon_backend_steps, + weight_decay=h.muon_wd, + row_normalize=h.muon_row_normalize, + ) + for group in self.optimizer_muon.param_groups: + group["base_lr"] = h.matrix_lr + self.optimizer_scalar = torch.optim.AdamW( + [{"params": scalar_params, "lr": h.scalar_lr, "base_lr": h.scalar_lr}], + betas=(h.beta1, h.beta2), + eps=h.adam_eps, + weight_decay=h.adam_wd, + fused=True, + ) + self.optimizers = [ + self.optimizer_tok, + self.optimizer_muon, + self.optimizer_scalar, + ] + self.replicated_params = list(tok_params[0]["params"]) + self.replicated_params.extend(scalar_params) + self.replicated_large_params = [] + self.replicated_packed_params = [] + for p in self.replicated_params: + if p.numel() <= PACKED_REPLICATED_GRAD_MAX_NUMEL: + self.replicated_packed_params.append(p) + else: + self.replicated_large_params.append(p) + + def __iter__(self): + return iter(self.optimizers) + + def zero_grad_all(self): + for opt in self.optimizers: + opt.zero_grad(set_to_none=True) + + def _all_reduce_packed_grads(self): + grads_by_key = collections.defaultdict(list) + for p in self.replicated_packed_params: + if p.grad is not None: + grads_by_key[(p.grad.device, p.grad.dtype)].append(p.grad) + for grads in grads_by_key.values(): + flat = torch.empty( + sum(g.numel() for g in grads), + device=grads[0].device, + dtype=grads[0].dtype, + ) + offset = 0 + for g in grads: + n = g.numel() + flat[offset : offset + n].copy_(g.contiguous().view(-1)) + offset += n + dist.all_reduce(flat, op=dist.ReduceOp.AVG) + offset = 0 + for g in grads: + n = g.numel() + g.copy_(flat[offset : offset + n].view_as(g)) + offset += n + + def step(self, distributed=False): + self.optimizer_muon.launch_reduce_scatters() + if distributed: + reduce_handles = [ + dist.all_reduce(p.grad, op=dist.ReduceOp.AVG, async_op=True) + for p in self.replicated_large_params + if p.grad is not None + ] + self._all_reduce_packed_grads() + for handle in reduce_handles: + handle.wait() + self.optimizer_tok.step() + self.optimizer_scalar.step() + self.optimizer_muon.step() + self.zero_grad_all() + + +def restore_fp32_params(model): + for module in model.modules(): + if isinstance(module, CastedLinear): + module.float() + for name, param in model.named_parameters(): + if ( + param.ndim < 2 + or any(pattern in name for pattern in CONTROL_TENSOR_NAME_PATTERNS) + ) and param.dtype != torch.float32: + param.data = param.data.float() + if hasattr(model, "qo_bank") and model.qo_bank is not None: + model.qo_bank.data = model.qo_bank.data.float() + model.kv_bank.data = model.kv_bank.data.float() + model.mlp_up_bank.data = model.mlp_up_bank.data.float() + model.mlp_down_bank.data = model.mlp_down_bank.data.float() + + +def collect_hessians(model, train_loader, h, device, n_calibration_batches=64): + hessians = {} + hooks = [] + for i, block in enumerate(model.blocks): + block.attn._calib = True + block.mlp._calib = True + block.mlp.use_fused = False + + def make_attn_hook(layer_idx): + def hook_fn(module, inp, out): + x = inp[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + for suffix in ["c_q", "c_k", "c_v"]: + name = f"blocks.{layer_idx}.attn.{suffix}.weight" + if name not in hessians: + hessians[name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(x.T, x) + y = module._last_proj_input + if y is not None: + y = y.float() + if y.ndim == 3: + y = y.reshape(-1, y.shape[-1]) + name = f"blocks.{layer_idx}.attn.proj.weight" + if name not in hessians: + hessians[name] = torch.zeros( + y.shape[1], y.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(y.T, y) + return hook_fn + + def make_mlp_hook(layer_idx): + def hook_fn(module, inp, out): + x = inp[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + name = f"blocks.{layer_idx}.mlp.fc.weight" + if name not in hessians: + hessians[name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(x.T, x) + h_act = module._last_down_input + if h_act is not None: + h_act = h_act.float() + if h_act.ndim == 3: + h_act = h_act.reshape(-1, h_act.shape[-1]) + name = f"blocks.{layer_idx}.mlp.proj.weight" + if name not in hessians: + hessians[name] = torch.zeros( + h_act.shape[1], h_act.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(h_act.T, h_act) + return hook_fn + + for i, block in enumerate(model.blocks): + hooks.append(block.attn.register_forward_hook(make_attn_hook(i))) + hooks.append(block.mlp.register_forward_hook(make_mlp_hook(i))) + + # Hessian hooks for embedding factorization projection layers + def make_linear_input_hook(weight_name): + def hook_fn(module, inp, out): + x = inp[0].detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + if weight_name not in hessians: + hessians[weight_name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[weight_name].addmm_(x.T, x) + return hook_fn + + if model.tie_embeddings: + hook_module = model.final_norm + + def make_output_hook(name): + def hook_fn(module, inp, out): + x = out.detach().float() + if x.ndim == 3: + x = x.reshape(-1, x.shape[-1]) + if name not in hessians: + hessians[name] = torch.zeros( + x.shape[1], x.shape[1], dtype=torch.float32, device=device + ) + hessians[name].addmm_(x.T, x) + return hook_fn + + hooks.append( + hook_module.register_forward_hook(make_output_hook("tok_emb.weight")) + ) + model.eval() + with torch.no_grad(): + for _ in range(n_calibration_batches): + x, _ = train_loader.next_batch(h.train_batch_tokens, h.grad_accum_steps) + model.forward_logits(x) + for hook in hooks: + hook.remove() + for i, block in enumerate(model.blocks): + block.attn._calib = False + block.mlp._calib = False + block.mlp.use_fused = True + for name in hessians: + hessians[name] = hessians[name].cpu() / n_calibration_batches + return hessians + + +def gptq_quantize_weight(w, H, clip_sigmas=3.0, clip_range=63, block_size=128): + W_orig = w.float().clone() + rows, cols = W_orig.shape + H = H.float().clone() + dead = torch.diag(H) == 0 + H[dead, dead] = 1 + damp = 0.01 * H.diag().mean() + H.diagonal().add_(damp) + perm = torch.argsort(H.diag(), descending=True) + invperm = torch.argsort(perm) + W_perm = W_orig[:, perm].clone() + W_perm[:, dead[perm]] = 0 + H = H[perm][:, perm] + Hinv = torch.cholesky_inverse(torch.linalg.cholesky(H)) + Hinv = torch.linalg.cholesky(Hinv, upper=True) + row_std = W_orig.std(dim=1) + s = (clip_sigmas * row_std / clip_range).clamp_min(1e-10).to(torch.float16) + sf = s.float() + Q = torch.zeros(rows, cols, dtype=torch.int8) + W_work = W_perm.clone() + for i1 in range(0, cols, block_size): + i2 = min(i1 + block_size, cols) + W_block = W_work[:, i1:i2].clone() + Hinv_block = Hinv[i1:i2, i1:i2] + Err = torch.zeros(rows, i2 - i1) + for j in range(i2 - i1): + w_col = W_block[:, j] + d = Hinv_block[j, j] + q_col = torch.clamp(torch.round(w_col / sf), -clip_range, clip_range) + Q[:, i1 + j] = q_col.to(torch.int8) + err = (w_col - q_col.float() * sf) / d + Err[:, j] = err + W_block[:, j:] -= err.unsqueeze(1) * Hinv_block[j, j:].unsqueeze(0) + if i2 < cols: + W_work[:, i2:] -= Err @ Hinv[i1:i2, i2:] + return Q[:, invperm], s + + +def _quantize_gate_int8_row(w): + # Symmetric int8-per-row quantization for small gate tensors. w shape + # (R, C) -> (R,) scales in fp16, int8 values in [-127, 127]. Single scale + # per row keeps accuracy high while halving storage vs fp16. + W = w.float().contiguous() + row_max = W.abs().amax(dim=1).clamp_min(1e-10) + s = (row_max / 127.0).to(torch.float16) + sf = s.float().view(-1, 1) + q = torch.clamp(torch.round(W / sf), -127, 127).to(torch.int8) + return q, s + + +def _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 gptq_mixed_quantize(state_dict, hessians, h): + result = {} + meta = {} + quant_gate = bool(getattr(h, "gated_attn_quant_gate", False)) + lqer_on = bool(getattr(h, "lqer_enabled", False)) + lqer_cands = {} + 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 + ret = gptq_quantize_weight( + t, hessians[name], clip_sigmas=cs, clip_range=clip_range + ) + q, s = ret + result[name + ".q"] = q + result[name + ".scale"] = s + meta[name] = f"gptq (int{bits})" + if lqer_on: + W_q = q.float() * s.float().view(-1, 1) + E = t.float() - W_q + lqer_cands[name] = (E, float(E.norm())) + if lqer_on and lqer_cands: + 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 "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" + + +def _byte_shuffle(data, stride=2): + if stride <= 1 or len(data) < stride: + return data + src = np.frombuffer(data, dtype=np.uint8) + n = len(src) + out = np.empty(n, dtype=np.uint8) + dest_off = 0 + for pos in range(stride): + chunk = src[pos::stride] + out[dest_off : dest_off + len(chunk)] = chunk + dest_off += len(chunk) + return _BSHF_MAGIC + bytes([stride]) + out.tobytes() + + +def _byte_unshuffle(data): + if len(data) < 5 or data[:4] != _BSHF_MAGIC: + return data + stride = data[4] + if stride < 2: + return data[5:] + payload = np.frombuffer(data, dtype=np.uint8, offset=5) + n = len(payload) + out = np.empty(n, dtype=np.uint8) + src_off = 0 + for pos in range(stride): + chunk_len = n // stride + (1 if pos < n % stride else 0) + out[pos::stride][:chunk_len] = payload[src_off : src_off + chunk_len] + src_off += chunk_len + return out.tobytes() + + +def _compress(data, compressor): + data = _byte_shuffle(data) + if compressor == "lzma": + return lzma.compress(data, preset=6) + elif compressor == "brotli": + import brotli + + return brotli.compress(data, quality=11) + raise ValueError(f"Unknown compressor: {compressor!r}") + + +def _decompress(data, compressor): + if compressor == "lzma": + raw = lzma.decompress(data) + elif compressor == "brotli": + import brotli + + raw = brotli.decompress(data) + else: + raise ValueError(f"Unknown compressor: {compressor!r}") + raw = _byte_unshuffle(raw) + return raw + + +def _unbank_state_dict(state_dict, num_layers): + sd = {} + n = num_layers + for k, v in state_dict.items(): + t = v.detach().cpu() if v is not None else None + if k == "qo_bank": + for i in range(n): + sd[f"blocks.{i}.attn.c_q.weight"] = t[i] + sd[f"blocks.{i}.attn.proj.weight"] = t[n + i] + elif k == "kv_bank": + for i in range(n): + sd[f"blocks.{i}.attn.c_k.weight"] = t[i] + sd[f"blocks.{i}.attn.c_v.weight"] = t[n + i] + elif k == "mlp_up_bank": + for i in range(n): + sd[f"blocks.{i}.mlp.fc.weight"] = t[i] + elif k == "mlp_down_bank": + for i in range(n): + sd[f"blocks.{i}.mlp.proj.weight"] = t[i] + else: + if t is not None: + sd[k] = t + return sd + + +def _rebank_state_dict(flat_sd, num_layers, model_dim, kv_dim, hidden_dim): + sd = {} + n = num_layers + sd["qo_bank"] = torch.zeros(2 * n, model_dim, model_dim) + sd["kv_bank"] = torch.zeros(2 * n, kv_dim, model_dim) + for i in range(n): + sd["qo_bank"][i] = flat_sd[f"blocks.{i}.attn.c_q.weight"] + sd["qo_bank"][n + i] = flat_sd[f"blocks.{i}.attn.proj.weight"] + sd["kv_bank"][i] = flat_sd[f"blocks.{i}.attn.c_k.weight"] + sd["kv_bank"][n + i] = flat_sd[f"blocks.{i}.attn.c_v.weight"] + sd["mlp_up_bank"] = torch.zeros(n, hidden_dim, model_dim) + sd["mlp_down_bank"] = torch.zeros(n, model_dim, hidden_dim) + for i in range(n): + sd["mlp_up_bank"][i] = flat_sd[f"blocks.{i}.mlp.fc.weight"] + sd["mlp_down_bank"][i] = flat_sd[f"blocks.{i}.mlp.proj.weight"] + for k, v in flat_sd.items(): + if not ( + k.startswith("blocks.") + and any( + p in k + for p in [ + ".attn.c_q.", ".attn.c_k.", ".attn.c_v.", + ".attn.proj.", ".mlp.fc.", ".mlp.proj.", + ] + ) + ): + sd[k] = v + return sd + + + +def _compressed_code_size(code): + code_raw = code.encode("utf-8") + minified = subprocess.run( + ["pyminify", "--no-rename-locals", "--no-hoist-literals", "--remove-literal-statements", "-"], + input=code_raw, capture_output=True, check=True, + ).stdout + compressed = lzma.compress(minified) + encoded = base64.b85encode(compressed) + wrapper = b'import lzma as L,base64 as B\nexec(L.decompress(B.b85decode("' + encoded + b'")))\n' + return len(code_raw), len(wrapper) + + +def serialize(h, base_model, code): + code_bytes_uncompressed, code_bytes = _compressed_code_size(code) + if h.is_main_process: + torch.save(base_model.state_dict(), h.model_path) + model_bytes = os.path.getsize(h.model_path) + log(f"Serialized model: {model_bytes} bytes") + log(f"Code size (uncompressed): {code_bytes_uncompressed} bytes") + log(f"Code size (compressed): {code_bytes} bytes") + sd_cpu = _unbank_state_dict(base_model.state_dict(), h.num_layers) + device = torch.device("cuda", h.local_rank) + t0 = time.perf_counter() + calib_loader = ShuffledSequenceLoader(h, device) + log("GPTQ:collecting Hessians from calibration data...") + hessians = collect_hessians( + base_model, + calib_loader, + h, + device, + n_calibration_batches=h.gptq_calibration_batches, + ) + log(f"GPTQ:collected {len(hessians)} Hessians in {time.perf_counter()-t0:.1f}s") + quant_result, quant_meta = gptq_mixed_quantize(sd_cpu, hessians, h) + quant_buf = io.BytesIO() + torch.save({"w": quant_result, "m": quant_meta}, quant_buf) + quant_raw = quant_buf.getvalue() + quant_blob = _compress(quant_raw, h.compressor) + quant_file_bytes = len(quant_blob) + bytes_total = quant_file_bytes + code_bytes + if h.is_main_process: + with open(h.quantized_model_path, "wb") as f: + f.write(quant_blob) + log(f"Serialized model quantized+{h.compressor}: {quant_file_bytes} bytes") + log(f"Total submission size quantized+{h.compressor}: {bytes_total} bytes") + return bytes_total, quant_file_bytes + + +def deserialize(h, device): + eval_model = GPT(h).to(device).bfloat16() + restore_fp32_params(eval_model) + flat_template = _unbank_state_dict(eval_model.state_dict(), h.num_layers) + with open(h.quantized_model_path, "rb") as f: + quant_blob_disk = f.read() + quant_state = torch.load( + io.BytesIO(_decompress(quant_blob_disk, h.compressor)), map_location="cpu" + ) + deq_flat = dequantize_mixed(quant_state["w"], quant_state["m"], flat_template) + head_dim = h.model_dim // h.num_heads + kv_dim = h.num_kv_heads * head_dim + hidden_dim = int(h.mlp_mult * h.model_dim) + deq_state = _rebank_state_dict(deq_flat, h.num_layers, h.model_dim, kv_dim, hidden_dim) + eval_model.load_state_dict(deq_state, strict=True) + return eval_model + + +def _loss_bpb(loss_sum, token_count, byte_count): + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_count.item()) + return val_loss, val_bpb + + +def eval_val(h, device, val_data, model, forward_logits_fn=None): + seq_len = h.eval_seq_len + local_batch_tokens = h.val_batch_tokens // (h.world_size * h.grad_accum_steps) + if local_batch_tokens < seq_len: + raise ValueError( + f"VAL_BATCH_SIZE must provide at least one sequence per rank; got VAL_BATCH_SIZE={h.val_batch_tokens}, WORLD_SIZE={h.world_size}, GRAD_ACCUM_STEPS={h.grad_accum_steps}, seq_len={seq_len}" + ) + local_batch_seqs = local_batch_tokens // seq_len + total_seqs = (val_data.val_tokens.numel() - 1) // seq_len + seq_start = total_seqs * h.rank // h.world_size + seq_end = total_seqs * (h.rank + 1) // h.world_size + + # TODO: Don't truncate this. + seq_end = seq_start + ((seq_end - seq_start) // local_batch_seqs) * local_batch_seqs + + val_loss_sum = torch.zeros((), device=device, dtype=torch.float64) + val_token_count = torch.zeros((), device=device, dtype=torch.float64) + val_byte_count = torch.zeros((), device=device, dtype=torch.float64) + run_forward_logits = ( + (model.module.forward_logits if hasattr(model, "module") else model.forward_logits) + if forward_logits_fn is None + else forward_logits_fn + ) + model.eval() + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + with torch.no_grad(): + for batch_seq_start in range(seq_start, seq_end, local_batch_seqs): + batch_seq_end = min(batch_seq_start + local_batch_seqs, seq_end) + raw_start = batch_seq_start * seq_len + raw_end = batch_seq_end * seq_len + 1 + local = val_data.val_tokens[raw_start:raw_end].to( + device=device, dtype=torch.int64, non_blocking=True + ) + x = local[:-1] + y = local[1:] + bos_pos = (x == BOS_ID).nonzero(as_tuple=True)[0].tolist() + cu_seqlens, max_seqlen = _build_cu_seqlens( + bos_pos, x.numel(), x.device, h.eval_seq_len, 64 + ) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + logits = run_forward_logits( + x[None], cu_seqlens=cu_seqlens, max_seqlen=max_seqlen + ).detach() + per_token_loss = F.cross_entropy( + logits.reshape(-1, logits.size(-1)).float(), + y.reshape(-1), + reduction="none", + ) + val_loss_sum += per_token_loss.to(torch.float64).sum() + val_token_count += float(y.numel()) + prev_ids = x + tgt_ids = y + if val_data.caseops_enabled and val_data.val_bytes is not None: + # CaseOps: read per-token byte budget from sidecar at the same + # global positions as the target tokens y. raw_start/raw_end + # span [raw_start, raw_end), x = local[:-1], y = local[1:], + # so y is at sidecar positions [raw_start + 1, raw_end). + sidecar_slice = val_data.val_bytes[raw_start + 1 : raw_end].to( + device=device, dtype=torch.int32, non_blocking=True + ) + val_byte_count += sidecar_slice.to(torch.float64).sum() + else: + token_bytes = val_data.base_bytes_lut[tgt_ids].to(dtype=torch.int16) + token_bytes += ( + val_data.has_leading_space_lut[tgt_ids] + & ~val_data.is_boundary_token_lut[prev_ids] + ).to(dtype=torch.int16) + val_byte_count += token_bytes.to(torch.float64).sum() + if dist.is_available() and dist.is_initialized(): + dist.all_reduce(val_loss_sum, op=dist.ReduceOp.SUM) + dist.all_reduce(val_token_count, op=dist.ReduceOp.SUM) + dist.all_reduce(val_byte_count, op=dist.ReduceOp.SUM) + model.train() + return _loss_bpb(val_loss_sum, val_token_count, val_byte_count) + + +def _find_docs(all_tokens): + bos_positions = (all_tokens == BOS_ID).nonzero(as_tuple=True)[0].numpy() + docs = [] + for i in range(len(bos_positions)): + start = int(bos_positions[i]) + end = ( + int(bos_positions[i + 1]) + if i + 1 < len(bos_positions) + else all_tokens.numel() + ) + if i + 1 < len(bos_positions): + end += 1 + assert end - start >= 2 + docs.append((start, end - start)) + return docs + + +def _build_ttt_global_batches(doc_entries, h, ascending=False): + batch_size = h.ttt_batch_size + global_doc_entries = sorted(doc_entries, key=lambda x: x[1][1]) + global_batches = [ + global_doc_entries[i : i + batch_size] + for i in range(0, len(global_doc_entries), batch_size) + ] + indexed = list(enumerate(global_batches)) + if not ascending: + indexed.sort(key=lambda ib: -max(dl for _, (_, dl) in ib[1])) + return indexed + + +def _init_batch_counter(path): + with open(path, "wb") as f: + f.write((0).to_bytes(4, "little")) + + +def _claim_next_batch(counter_path, queue_len): + try: + with open(counter_path, "r+b") as f: + fcntl.flock(f, fcntl.LOCK_EX) + idx = int.from_bytes(f.read(4), "little") + f.seek(0) + f.write((idx + 1).to_bytes(4, "little")) + f.flush() + except FileNotFoundError: + return queue_len + return idx + + +def _compute_chunk_window(ci, pred_len, num_chunks, chunk_size, eval_seq_len): + chunk_end = pred_len if ci == num_chunks - 1 else (ci + 1) * chunk_size + win_start = max(0, chunk_end - eval_seq_len) + win_len = chunk_end - win_start + chunk_start = ci * chunk_size + chunk_offset = chunk_start - win_start + chunk_len = chunk_end - chunk_start + return win_start, win_len, chunk_offset, chunk_len + + +def _accumulate_bpb( + ptl, + x, + y, + chunk_offsets, + chunk_lens, + pos_idx, + base_bytes_lut, + has_leading_space_lut, + is_boundary_token_lut, + loss_sum, + byte_sum, + token_count, + y_bytes=None, +): + pos = pos_idx[: x.size(1)].unsqueeze(0) + mask = ( + (chunk_lens.unsqueeze(1) > 0) + & (pos >= chunk_offsets.unsqueeze(1)) + & (pos < (chunk_offsets + chunk_lens).unsqueeze(1)) + ) + mask_f64 = mask.to(torch.float64) + if y_bytes is not None: + tok_bytes = y_bytes.to(torch.float64) + else: + tok_bytes = base_bytes_lut[y].to(torch.float64) + tok_bytes += (has_leading_space_lut[y] & ~is_boundary_token_lut[x]).to( + torch.float64 + ) + loss_sum += (ptl.to(torch.float64) * mask_f64).sum() + byte_sum += (tok_bytes * mask_f64).sum() + token_count += chunk_lens.to(torch.float64).sum() + + +def _loss_bpb_from_sums(loss_sum, token_count, byte_sum): + val_loss = (loss_sum / token_count).item() + val_bpb = val_loss / math.log(2.0) * (token_count.item() / byte_sum.item()) + return val_loss, val_bpb + + +def _add_to_counter(path, delta): + try: + with open(path, "r+b") as f: + fcntl.flock(f, fcntl.LOCK_EX) + cur = int.from_bytes(f.read(8), "little", signed=True) + cur += int(delta) + f.seek(0) + f.write(int(cur).to_bytes(8, "little", signed=True)) + f.flush() + return cur + except FileNotFoundError: + return int(delta) + + +def _init_int64_counter(path): + with open(path, "wb") as f: + f.write((0).to_bytes(8, "little", signed=True)) + + +def _select_ttt_doc_entries(docs, h): + doc_entries = list(enumerate(docs)) + if h.val_doc_fraction < 1.0: + sample_n = max(1, int(round(len(docs) * h.val_doc_fraction))) + sampled_indices = sorted( + random.Random(h.seed).sample(range(len(docs)), sample_n) + ) + return [(i, docs[i]) for i in sampled_indices] + return doc_entries + + +def _ppm_mixture_bpb( + target_ids, + prev_ids, + nll_nats, + token_bytes_lut, + has_leading_space_lut_np, + is_boundary_token_lut_np, + order=5, + lambda_hi=0.9, # kept for arg-compat; not used in multibin + lambda_lo=0.05, # kept for arg-compat; not used in multibin + conf_threshold=0.9, # kept for arg-compat; not used in multibin + log_prefix="ppm_mix", +): + _ln = math.log + LOG2 = _ln(2.0) + UNIFORM_LOGP = _ln(1.0 / 256.0) + + # Multibin schedule. Override via env if needed for tuning. + BINS_ENV = os.environ.get("MIX_BINS", "") + if BINS_ENV: + # Format: "thresh:lambda,thresh:lambda,..." + BINS = [tuple(float(x) for x in pair.split(":")) for pair in BINS_ENV.split(",")] + else: + BINS = [ + (0.95, 0.02), + (0.85, 0.10), + (0.75, 0.40), + (0.0, 1.0), + ] + + num_tokens = len(target_ids) + byte_stream = [] + byte_nn_logp = [] + for i in range(num_tokens): + tid = int(target_ids[i]) + pid = int(prev_ids[i]) + tb = token_bytes_lut[tid] if 0 <= tid < len(token_bytes_lut) else b"" + has_space = bool(has_leading_space_lut_np[tid]) if 0 <= tid < len(has_leading_space_lut_np) else False + prev_is_boundary = pid < 0 or ( + bool(is_boundary_token_lut_np[pid]) if 0 <= pid < len(is_boundary_token_lut_np) else True + ) + include_space = has_space and not prev_is_boundary + n_bytes = len(tb) + (1 if include_space else 0) + if n_bytes == 0: + continue + token_logp = -float(nll_nats[i]) + per_byte_logp = token_logp / n_bytes + if include_space: + byte_stream.append(0x20) + byte_nn_logp.append(per_byte_logp) + for b in tb: + byte_stream.append(b) + byte_nn_logp.append(per_byte_logp) + + total_bytes = len(byte_stream) + if total_bytes == 0: + return (0.0, 0.0, 0.0) + + ctx_counts = {} + mix_nll = 0.0 + nn_nll = 0.0 + ppm_nll = 0.0 + window = bytearray() + bin_hits = [0] * len(BINS) + + for t in range(total_bytes): + b = byte_stream[t] + ppm_log_p = None + confidence = 0.0 + seen_any = False + escape_log_prob = 0.0 + for K in range(min(order, len(window)), -1, -1): + ctx = bytes(window[-K:]) if K > 0 else b"" + counts = ctx_counts.get(ctx) + if counts is None: + continue + unique = len(counts) + total = sum(counts.values()) + denom = total + unique + if not seen_any: + max_count = max(counts.values()) + confidence = max_count / denom + seen_any = True + if b in counts: + prob_here = counts[b] / denom + ppm_log_p = escape_log_prob + _ln(prob_here) + break + escape_log_prob += _ln(unique / denom) if unique > 0 else 0.0 + if ppm_log_p is None: + ppm_log_p = escape_log_prob + UNIFORM_LOGP + + nn_log_p = byte_nn_logp[t] + + # Multibin gate + lam = BINS[-1][1] + bin_idx = len(BINS) - 1 + for bi, (thresh, l) in enumerate(BINS): + if confidence >= thresh: + lam = l + bin_idx = bi + break + bin_hits[bin_idx] += 1 + + if lam <= 0.0: + log_mix = ppm_log_p + elif lam >= 1.0: + log_mix = nn_log_p + else: + a = _ln(lam) + nn_log_p + c = _ln(1.0 - lam) + ppm_log_p + log_mix = max(a, c) + math.log1p(math.exp(-abs(a - c))) + + mix_nll -= log_mix + nn_nll -= nn_log_p + ppm_nll -= ppm_log_p + + for K in range(0, min(order, len(window)) + 1): + ctx = bytes(window[-K:]) if K > 0 else b"" + d = ctx_counts.get(ctx) + if d is None: + d = {} + ctx_counts[ctx] = d + d[b] = d.get(b, 0) + 1 + window.append(b) + if len(window) > order: + del window[0] + + mix_bpb = mix_nll / total_bytes / LOG2 + ppm_bpb = ppm_nll / total_bytes / LOG2 + nn_bpb = nn_nll / total_bytes / LOG2 + + bin_pcts = " ".join(f"bin{i}_pct={bin_hits[i]/total_bytes:.3f}" for i in range(len(BINS))) + print( + f"{log_prefix} bytes={total_bytes} mix_bpb={mix_bpb:.6f} " + f"ppm_only={ppm_bpb:.6f} nn_only={nn_bpb:.6f} {bin_pcts}", + flush=True, + ) + + return mix_bpb, ppm_bpb, nn_bpb + + +def train_val_ttt_global_sgd_distributed(h, device, val_data, base_model, val_tokens, batch_seqs=None): + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + base_model.eval() + seq_len = h.eval_seq_len + total_tokens = val_tokens.numel() - 1 + ttt_chunk = h.global_ttt_chunk_tokens + batch_seqs = h.global_ttt_batch_seqs if batch_seqs is None else batch_seqs + num_chunks = (total_tokens + ttt_chunk - 1) // ttt_chunk + ttt_params = [p for p in base_model.parameters()] + for p in ttt_params: + p.requires_grad_(True) + optimizer = torch.optim.SGD( + ttt_params, lr=h.global_ttt_lr, momentum=h.global_ttt_momentum + ) + t_start = time.perf_counter() + for ci in range(num_chunks): + chunk_start = ci * ttt_chunk + chunk_end = min((ci + 1) * ttt_chunk, total_tokens) + is_last_chunk = ci == num_chunks - 1 + if is_last_chunk or h.global_ttt_epochs <= 0: + continue + base_model.train() + chunk_seqs = (chunk_end - chunk_start) // seq_len + if chunk_seqs <= 0: + continue + warmup_chunks = max(0, min(h.global_ttt_warmup_chunks, num_chunks - 1)) + if warmup_chunks > 0 and ci < warmup_chunks: + warmup_denom = max(warmup_chunks - 1, 1) + warmup_t = ci / warmup_denom + lr_now = ( + h.global_ttt_warmup_start_lr + + (h.global_ttt_lr - h.global_ttt_warmup_start_lr) * warmup_t + ) + else: + decay_steps = max(num_chunks - 1 - warmup_chunks, 1) + decay_ci = max(ci - warmup_chunks, 0) + lr_now = h.global_ttt_lr * 0.5 * ( + 1.0 + math.cos(math.pi * decay_ci / decay_steps) + ) + for pg in optimizer.param_groups: + pg["lr"] = lr_now + my_seq_s = chunk_seqs * h.rank // h.world_size + my_seq_e = chunk_seqs * (h.rank + 1) // h.world_size + my_chunk_seqs = my_seq_e - my_seq_s + for _ in range(h.global_ttt_epochs): + for bs in range(0, my_chunk_seqs, batch_seqs): + be = min(bs + batch_seqs, my_chunk_seqs) + actual_bs = my_seq_s + bs + start_tok = chunk_start + actual_bs * seq_len + end_tok = chunk_start + (my_seq_s + be) * seq_len + 1 + if end_tok > val_tokens.numel(): + continue + local = val_tokens[start_tok:end_tok].to(device=device, dtype=torch.int64) + x_flat = local[:-1] + y_flat = local[1:] + optimizer.zero_grad(set_to_none=True) + with torch.enable_grad(): + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + if h.global_ttt_respect_doc_boundaries: + bos_pos = (x_flat == BOS_ID).nonzero(as_tuple=True)[0].tolist() + cu_seqlens, max_seqlen = _build_cu_seqlens( + bos_pos, x_flat.numel(), x_flat.device, h.eval_seq_len, 64 + ) + loss = base_model( + x_flat[None], + y_flat[None], + cu_seqlens=cu_seqlens, + max_seqlen=max_seqlen, + ) + else: + x = x_flat.reshape(-1, seq_len) + y = y_flat.reshape(-1, seq_len) + loss = base_model(x, y) + loss.backward() + if dist.is_available() and dist.is_initialized(): + for p in ttt_params: + if p.grad is not None: + dist.all_reduce(p.grad, op=dist.ReduceOp.SUM) + p.grad.mul_(1.0 / h.world_size) + if h.global_ttt_grad_clip > 0: + torch.nn.utils.clip_grad_norm_(ttt_params, h.global_ttt_grad_clip) + optimizer.step() + base_model.eval() + if h.rank == 0: + elapsed = time.perf_counter() - t_start + log( + f"tttg: c{ci+1}/{num_chunks} lr:{lr_now:.6f} t:{elapsed:.1f}s" + ) + for p in base_model.parameters(): + p.requires_grad_(True) + base_model.eval() + + +def eval_val_ttt_phased(h, base_model, device, val_data, forward_ttt_train): + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + base_model.eval() + for p in base_model.parameters(): + p.requires_grad_(False) + all_tokens = val_data.val_tokens + all_tokens_idx = all_tokens.to(torch.int32) + docs = _find_docs(all_tokens) + doc_entries = _select_ttt_doc_entries(docs, h) + prefix_doc_limit = max(0, min(len(doc_entries), int(h.phased_ttt_prefix_docs))) + num_phases = max(1, int(h.phased_ttt_num_phases)) + phase_boundaries = [] + for pi in range(num_phases): + boundary = prefix_doc_limit * (pi + 1) // num_phases + phase_boundaries.append(boundary) + current_phase = 0 + current_phase_boundary = phase_boundaries[0] + log( + "ttt_phased:" + f" total_docs:{len(doc_entries)} prefix_docs:{prefix_doc_limit} " + f"suffix_docs:{len(doc_entries) - prefix_doc_limit}" + f" num_phases:{num_phases} boundaries:{phase_boundaries}" + ) + # PPM mixture scratch buffers (port from #1835). + collect_ppm = bool(getattr(h, "ppm_enabled", False)) and val_data.token_bytes_py is not None + if collect_ppm: + ppm_capacity = int(all_tokens.numel()) + ppm_pos_nll = torch.zeros((ppm_capacity,), dtype=torch.float64, device=device) + ppm_pos_tgt = torch.zeros((ppm_capacity,), dtype=torch.int64, device=device) + ppm_pos_prev = torch.zeros((ppm_capacity,), dtype=torch.int64, device=device) + ppm_pos_written = torch.zeros((ppm_capacity,), dtype=torch.int32, device=device) + 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) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + per_tok_loss = forward_ttt_train(x, y, lora=cur_lora) + # CaseOps sidecar-driven byte budget. Mirror the index pattern + # used to build y from all_tokens: y[b, j] corresponds to the + # token at global position tok_starts[b] + 1 + j (when valid). + y_bytes_arg = None + if val_data.caseops_enabled and val_data.val_bytes is not None: + y_idx = ( + tok_starts.unsqueeze(1) + + 1 + + col_idx[:context_size].unsqueeze(0) + ) + y_idx = y_idx.clamp_(max=val_data.val_bytes.numel() - 1) + y_bytes_arg = val_data.val_bytes[y_idx].to( + device=device, dtype=torch.int32, non_blocking=True + ) + # Mirror the `valid` masking used for y so out-of-range tokens + # contribute zero bytes (matches y=0 substitution above). + y_bytes_arg = torch.where( + valid, y_bytes_arg, torch.zeros_like(y_bytes_arg) + ) + with torch.no_grad(): + _accumulate_bpb( + per_tok_loss, + x, + y, + chunk_offsets, + chunk_lens, + ctx_pos, + val_data.base_bytes_lut, + val_data.has_leading_space_lut, + val_data.is_boundary_token_lut, + loss_sum, + byte_sum, + token_count, + y_bytes=y_bytes_arg, + ) + if collect_ppm: + # Capture this chunk's per-token NLL into position-indexed + # buffer. Score-first: this is the FIRST scoring of these + # tokens (TTT update happens after, in needs_train branch). + for b in range(bsz): + if not active[b]: + continue + c_off = int(chunk_offsets_cpu[b].item()) + c_len = int(chunk_lens_cpu[b].item()) + if c_len <= 0: + continue + abs_start = int(tok_starts[b].item()) + c_off + abs_end = abs_start + c_len + if abs_end > ppm_capacity: + abs_end = ppm_capacity + c_len = abs_end - abs_start + if c_len <= 0: + continue + ppm_pos_nll[abs_start:abs_end] = per_tok_loss[b, c_off:c_off + c_len].to(torch.float64) + ppm_pos_tgt[abs_start:abs_end] = y[b, c_off:c_off + c_len].to(torch.int64) + ppm_pos_prev[abs_start:abs_end] = x[b, c_off:c_off + c_len].to(torch.int64) + ppm_pos_written[abs_start:abs_end] = 1 + 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) + if collect_ppm: + dist.all_reduce(ppm_pos_nll, op=dist.ReduceOp.SUM) + dist.all_reduce(ppm_pos_tgt, op=dist.ReduceOp.SUM) + dist.all_reduce(ppm_pos_prev, op=dist.ReduceOp.SUM) + dist.all_reduce(ppm_pos_written, op=dist.ReduceOp.SUM) + if collect_ppm and h.rank == 0: + try: + written_mask = (ppm_pos_written > 0).cpu().numpy() + subset_cap = min(int(h.ppm_subset_tokens), ppm_capacity) + tgt_full = ppm_pos_tgt.detach().cpu().numpy().astype(np.int64) + prev_full = ppm_pos_prev.detach().cpu().numpy().astype(np.int64) + nll_full = ppm_pos_nll.detach().cpu().numpy().astype(np.float64) + sel_idx = np.flatnonzero(written_mask[:subset_cap]) + if sel_idx.size > 0: + tgt_np = tgt_full[sel_idx] + prev_np = prev_full[sel_idx] + nll_np = nll_full[sel_idx] + hls_np = val_data.has_leading_space_lut.detach().cpu().numpy().astype(bool) + isb_np = val_data.is_boundary_token_lut.detach().cpu().numpy().astype(bool) + t0 = time.perf_counter() + _ppm_mixture_bpb( + target_ids=tgt_np, + prev_ids=prev_np, + nll_nats=nll_np, + token_bytes_lut=val_data.token_bytes_py, + has_leading_space_lut_np=hls_np, + is_boundary_token_lut_np=isb_np, + order=h.ppm_order, + lambda_hi=h.ppm_lambda_hi, + lambda_lo=h.ppm_lambda_lo, + conf_threshold=h.ppm_conf_threshold, + ) + log(f"ppm_mix_time:{time.perf_counter() - t0:.1f}s subset={sel_idx.size} tokens") + except Exception as e: + log(f"ppm_mix:FAILED {type(e).__name__}: {e}") + 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 + + def step_fn(step, lr_scale): + optimizers.zero_grad_all() + train_loss = torch.zeros((), device=device) + for micro_step in range(h.grad_accum_steps): + x, y, cu_seqlens, _max_seqlen = train_loader.next_batch( + h.train_batch_tokens, h.grad_accum_steps + ) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + loss = model(x, y, cu_seqlens=cu_seqlens, max_seqlen=h.train_seq_len) + train_loss += loss.detach() + (loss / h.grad_accum_steps).backward() + train_loss /= h.grad_accum_steps + frac = ( + min(step / h.muon_momentum_warmup_steps, 1.0) + if h.muon_momentum_warmup_steps > 0 + else 1.0 + ) + muon_momentum = ( + 1 - frac + ) * h.muon_momentum_warmup_start + frac * h.muon_momentum + for group in optimizers.optimizer_muon.param_groups: + group["momentum"] = muon_momentum + for opt in optimizers: + for group in opt.param_groups: + group["lr"] = group["base_lr"] * lr_scale + if h.grad_clip_norm > 0: + torch.nn.utils.clip_grad_norm_(base_model.parameters(), h.grad_clip_norm) + optimizers.step(distributed=h.distributed) + return train_loss + + if h.warmup_steps > 0: + initial_model_state = { + name: tensor.detach().cpu().clone() + for (name, tensor) in base_model.state_dict().items() + } + initial_optimizer_states = [ + copy.deepcopy(opt.state_dict()) for opt in optimizers + ] + model.train() + num_tokens_local = h.train_batch_tokens // h.world_size + for blk in base_model.blocks: + blk.attn.rotary(num_tokens_local, device, torch.bfloat16) + cu_bucket_size = train_loader.cu_bucket_size + warmup_cu_buckets = tuple(cu_bucket_size * i for i in range(1, 5)) + warmup_cu_iters = 3 + x, y, cu_seqlens, _ = train_loader.next_batch( + h.train_batch_tokens, h.grad_accum_steps + ) + log(f"warmup_cu_buckets:{','.join(str(b) for b in warmup_cu_buckets)} iters_each:{warmup_cu_iters}") + def _run_cu_bucket_warmup(): + for bucket_len in warmup_cu_buckets: + boundaries = list(range(0, x.size(1), max(h.train_seq_len, 1))) + if boundaries[-1] != x.size(1): + boundaries.append(x.size(1)) + cu = torch.full((bucket_len,), x.size(1), dtype=torch.int32, device=device) + cu[: len(boundaries)] = torch.tensor(boundaries, dtype=torch.int32, device=device) + for _ in range(warmup_cu_iters): + optimizers.zero_grad_all() + with torch.autocast(device_type="cuda", dtype=torch.bfloat16, enabled=True): + wloss = model(x, y, cu_seqlens=cu, max_seqlen=h.train_seq_len) + (wloss / h.grad_accum_steps).backward() + optimizers.zero_grad_all() + _run_cu_bucket_warmup() + if h.num_loops > 0: + base_model.looping_active = True + _run_cu_bucket_warmup() + base_model.looping_active = False + for warmup_step in range(h.warmup_steps): + step_fn(warmup_step, 1.0) + if ( + warmup_step <= 5 + or (warmup_step + 1) % 10 == 0 + or warmup_step + 1 == h.warmup_steps + ): + log(f"warmup_step: {warmup_step+1}/{h.warmup_steps}") + if h.num_loops > 0: + base_model.looping_active = True + log( + f"loop_warmup:enabled encoder:{base_model.encoder_indices} decoder:{base_model.decoder_indices}" + ) + for warmup_step in range(h.warmup_steps): + step_fn(warmup_step, 1.0) + if ( + warmup_step <= 5 + or (warmup_step + 1) % 10 == 0 + or warmup_step + 1 == h.warmup_steps + ): + log(f"loop_warmup_step: {warmup_step+1}/{h.warmup_steps}") + base_model.looping_active = False + base_model.load_state_dict(initial_model_state, strict=True) + for (opt, state) in zip(optimizers, initial_optimizer_states, strict=True): + opt.load_state_dict(state) + optimizers.zero_grad_all() + train_loader = DocumentPackingLoader(h, device) + ema_state = { + name: t.detach().float().clone() + for (name, t) in base_model.state_dict().items() + } + ema_decay = h.ema_decay + training_time_ms = 0.0 + stop_after_step = None + torch.cuda.synchronize() + t0 = time.perf_counter() + step = 0 + while True: + last_step = ( + step == h.iterations + or stop_after_step is not None + and step >= stop_after_step + ) + should_validate = ( + last_step or h.val_loss_every > 0 and step % h.val_loss_every == 0 + ) + if should_validate: + torch.cuda.synchronize() + training_time_ms += 1e3 * (time.perf_counter() - t0) + val_loss, val_bpb = eval_val( + h, device, val_data, model, compiled_forward_logits + ) + log( + f"{step}/{h.iterations} val_loss: {val_loss:.4f} val_bpb: {val_bpb:.4f}" + ) + torch.cuda.synchronize() + t0 = time.perf_counter() + if last_step: + if stop_after_step is not None and step < h.iterations: + log( + f"stopping_early: wallclock_cap train_time: {training_time_ms:.0f}ms step: {step}/{h.iterations}" + ) + break + elapsed_ms = training_time_ms + 1e3 * (time.perf_counter() - t0) + frac = training_frac(step, elapsed_ms) + scale = lr_mul(frac) + if ( + h.num_loops > 0 + and not base_model.looping_active + and frac >= h.enable_looping_at + ): + base_model.looping_active = True + log( + f"layer_loop:enabled step:{step} frac:{frac:.3f} encoder:{base_model.encoder_indices} decoder:{base_model.decoder_indices}" + ) + train_loss = step_fn(step, scale) + with torch.no_grad(): + for (name, t) in base_model.state_dict().items(): + ema_state[name].mul_(ema_decay).add_( + t.detach().float(), alpha=1.0 - ema_decay + ) + step += 1 + approx_training_time_ms = training_time_ms + 1e3 * (time.perf_counter() - t0) + should_log_train = h.train_log_every > 0 and ( + step <= 5 or step % h.train_log_every == 0 or stop_after_step is not None + ) + if should_log_train: + tok_per_sec = step * h.train_batch_tokens / (approx_training_time_ms / 1e3) + log( + f"{step}/{h.iterations} train_loss: {train_loss.item():.4f} train_time: {approx_training_time_ms/60000:.1f}m tok/s: {tok_per_sec:.0f}" + ) + reached_cap = ( + max_wallclock_ms is not None and approx_training_time_ms >= max_wallclock_ms + ) + if h.distributed and max_wallclock_ms is not None: + reached_cap_tensor = torch.tensor(int(reached_cap), device=device) + dist.all_reduce(reached_cap_tensor, op=dist.ReduceOp.MAX) + reached_cap = bool(reached_cap_tensor.item()) + if stop_after_step is None and reached_cap: + stop_after_step = step + log( + f"peak memory allocated: {torch.cuda.max_memory_allocated()//1024//1024} MiB reserved: {torch.cuda.max_memory_reserved()//1024//1024} MiB" + ) + log("ema:applying EMA weights") + current_state = base_model.state_dict() + avg_state = { + name: t.to(dtype=current_state[name].dtype) for (name, t) in ema_state.items() + } + base_model.load_state_dict(avg_state, strict=True) + return base_model, compiled_model, compiled_forward_logits + + +def train_and_eval(h, device): + random.seed(h.seed) + np.random.seed(h.seed) + torch.manual_seed(h.seed) + torch.cuda.manual_seed_all(h.seed) + if h.artifact_dir and h.is_main_process: + os.makedirs(h.artifact_dir, exist_ok=True) + val_data = ValidationData(h, device) + 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" + 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}") + 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) + + _fwd_ttt_compiled_inner = None + + def _fwd_ttt(input_ids, target_ids, lora): + nonlocal _fwd_ttt_compiled_inner + if _fwd_ttt_compiled_inner is None: + _fwd_ttt_compiled_inner = torch.compile(_fwd_ttt_inner, dynamic=True) + return _fwd_ttt_compiled_inner(input_ids, target_ids, lora=lora) + + fwd_ttt_compiled = _fwd_ttt + log(f"ttt_lora:warming up compile (random tokens, no val data)") + global BOS_ID + if BOS_ID is None: + BOS_ID = 1 + t_warmup = time.perf_counter() + warmup_bszes = [h.ttt_batch_size] + for bsz in warmup_bszes: + wl = BatchedTTTLoRA( + bsz, ttt_model, h.ttt_lora_rank, + k_lora=h.ttt_k_lora, mlp_lora=h.ttt_mlp_lora, o_lora=h.ttt_o_lora, + ).to(device) + wo = torch.optim.AdamW( + wl.parameters(), + lr=h.ttt_lora_lr, + betas=(h.ttt_beta1, h.ttt_beta2), + eps=1e-10, + weight_decay=h.ttt_weight_decay, + fused=True, + ) + for ctx_len in (h.ttt_chunk_size, h.ttt_eval_seq_len): + xw = torch.randint(0, h.vocab_size, (bsz, ctx_len), device=device, dtype=torch.int64) + yw = torch.randint(0, h.vocab_size, (bsz, ctx_len), device=device, dtype=torch.int64) + with torch.autocast(device_type="cuda", dtype=torch.bfloat16): + ptl = fwd_ttt_compiled(xw, yw, lora=wl) + ptl[:, : min(h.ttt_chunk_size, ctx_len)].mean(dim=-1).sum().backward() + wo.step() + wo.zero_grad(set_to_none=True) + del wl, wo + torch.cuda.empty_cache() + compile_elapsed = time.perf_counter() - t_warmup + log(f"ttt_lora:compile warmup done ({compile_elapsed:.1f}s)") + log("\nbeginning TTT eval timer") + torch.cuda.synchronize() + t_ttt = time.perf_counter() + ttt_val_loss, ttt_val_bpb = eval_val_ttt_phased( + h, ttt_model, device, val_data, forward_ttt_train=fwd_ttt_compiled + ) + torch.cuda.synchronize() + ttt_eval_elapsed = time.perf_counter() - t_ttt + log( + "quantized_ttt_phased " + f"val_loss:{ttt_val_loss:.8f} val_bpb:{ttt_val_bpb:.8f} " + f"eval_time:{1e3*ttt_eval_elapsed:.0f}ms" + ) + log(f"total_eval_time:{ttt_eval_elapsed:.1f}s") + del ttt_model + + +def main(): + world_size = int(os.environ.get("WORLD_SIZE", "1")) + local_rank = int(os.environ.get("LOCAL_RANK", "0")) + distributed = "RANK" in os.environ and "WORLD_SIZE" in os.environ + if not torch.cuda.is_available(): + raise RuntimeError("CUDA is required") + if world_size <= 0: + raise ValueError(f"WORLD_SIZE must be positive, got {world_size}") + if 8 % world_size != 0: + raise ValueError( + f"WORLD_SIZE={world_size} must divide 8 so grad_accum_steps stays integral" + ) + device = torch.device("cuda", local_rank) + torch.cuda.set_device(device) + if distributed: + dist.init_process_group(backend="nccl", device_id=device) + dist.barrier() + torch.backends.cuda.matmul.allow_tf32 = True + torch.backends.cudnn.allow_tf32 = True + torch.set_float32_matmul_precision("high") + from torch.backends.cuda import ( + enable_cudnn_sdp, + enable_flash_sdp, + enable_math_sdp, + enable_mem_efficient_sdp, + ) + + enable_cudnn_sdp(False) + enable_flash_sdp(True) + enable_mem_efficient_sdp(False) + enable_math_sdp(False) + torch._dynamo.config.optimize_ddp = False + torch._dynamo.config.cache_size_limit = 16 + h = Hyperparameters() + set_logging_hparams(h) + if h.is_main_process: + os.makedirs(h.artifact_dir if h.artifact_dir else "logs", exist_ok=True) + log(100 * "=", console=False) + log("Hyperparameters:", console=True) + for (k, v) in sorted(vars(type(h)).items()): + if not k.startswith("_"): + log(f" {k}: {v}", console=True) + log("=" * 100, console=False) + log("Source code:", console=False) + log("=" * 100, console=False) + with open(__file__, "r", encoding="utf-8") as _src: + log(_src.read(), console=False) + log("=" * 100, console=False) + log(f"Running Python {sys.version}", console=False) + log(f"Running PyTorch {torch.__version__}", console=False) + log("=" * 100, console=False) + train_and_eval(h, device) + if distributed: + dist.destroy_process_group() + + +if __name__ == "__main__": + main() diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed1337.log b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed1337.log new file mode 100644 index 0000000000..c8c001406b --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed1337.log @@ -0,0 +1,407 @@ +W0428 02:27:09.521000 256406 torch/distributed/run.py:803] +W0428 02:27:09.521000 256406 torch/distributed/run.py:803] ***************************************** +W0428 02:27:09.521000 256406 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. +W0428 02:27:09.521000 256406 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + beta1: 0.9 + beta2: 0.95 + caseops_enabled: True + compressor: brotli + data_dir: ./data/ + datasets_dir: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2048 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.01 + gated_attn_quant_gate: False + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 0.5 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/434de36e-20e4-49fc-87de-0279000947bc.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_rank: 4 + lqer_top_k: 3 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 3 + phased_ttt_prefix_docs: 1500 + ppm_conf_threshold: 0.9 + ppm_enabled: True + ppm_lambda_hi: 0.9 + ppm_lambda_lo: 0.05 + ppm_order: 5 + ppm_subset_tokens: 4000000 + qk_gain_init: 5.0 + quantized_model_path: final_model.int6.ptz + rank: 0 + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: 434de36e-20e4-49fc-87de-0279000947bc + scalar_lr: 0.02 + seed: 1337 + skip_gates_enabled: True + smear_gate_enabled: True + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 1.0 + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: /workspace/work_mixer/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_train_*.bin + train_log_every: 500 + train_seq_len: 2048 + ttt_batch_size: 64 + ttt_beta1: 0.0 + ttt_beta2: 0.999 + ttt_chunk_size: 48 + ttt_enabled: True + ttt_eval_batches: + ttt_eval_seq_len: 2048 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 1.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 488 +val_tokens: 9662464 +model_params:35945671 +gptq:reserving 0s, effective=599500ms +warmup_cu_buckets:64,128,192,256 iters_each:3 +warmup_step: 1/20 +warmup_step: 2/20 +warmup_step: 3/20 +warmup_step: 4/20 +warmup_step: 5/20 +warmup_step: 6/20 +warmup_step: 10/20 +warmup_step: 20/20 +loop_warmup:enabled encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +loop_warmup_step: 1/20 +loop_warmup_step: 2/20 +loop_warmup_step: 3/20 +loop_warmup_step: 4/20 +loop_warmup_step: 5/20 +loop_warmup_step: 6/20 +loop_warmup_step: 10/20 +loop_warmup_step: 20/20 +0/20000 val_loss: 9.0060 val_bpb: 4.1902 +1/20000 train_loss: 9.0078 train_time: 0.0m tok/s: 12215410 +2/20000 train_loss: 12.9761 train_time: 0.0m tok/s: 11617057 +3/20000 train_loss: 10.2157 train_time: 0.0m tok/s: 10392935 +4/20000 train_loss: 8.6963 train_time: 0.0m tok/s: 9828042 +5/20000 train_loss: 7.8629 train_time: 0.0m tok/s: 9533054 +500/20000 train_loss: 2.7282 train_time: 0.8m tok/s: 8380311 +1000/20000 train_loss: 2.7678 train_time: 1.6m tok/s: 8363231 +1500/20000 train_loss: 2.5991 train_time: 2.4m tok/s: 8360657 +2000/20000 train_loss: 2.6856 train_time: 3.1m tok/s: 8360288 +layer_loop:enabled step:2231 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: 2.4801 train_time: 4.1m tok/s: 7949577 +3000/20000 train_loss: 2.5469 train_time: 5.3m tok/s: 7393438 +3500/20000 train_loss: 2.4304 train_time: 6.5m tok/s: 7082339 +4000/20000 train_loss: 2.4313 train_time: 7.6m tok/s: 6867110 +4000/20000 val_loss: 2.4114 val_bpb: 1.1220 +4500/20000 train_loss: 2.3121 train_time: 8.8m tok/s: 6709403 +5000/20000 train_loss: 2.4542 train_time: 9.9m tok/s: 6588149 +5020/20000 val_loss: 2.3279 val_bpb: 1.0831 +stopping_early: wallclock_cap train_time: 599676ms step: 5020/20000 +peak memory allocated: 41709 MiB reserved: 47026 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.30363318 val_bpb:1.07181015 eval_time:6602ms +Serialized model: 135417533 bytes +Code size (uncompressed): 161565 bytes +Code size (compressed): 33305 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 3.5s +Quantized weights: + gptq (int6): blocks.attn.c_k.weight, blocks.attn.c_q.weight, blocks.attn.c_v.weight, blocks.attn.proj.weight, blocks.mlp.fc.weight, blocks.mlp.proj.weight + gptq (int6)+lqer_asym: blocks.mlp.fc.weight + gptq (int7)+lqer_asym: tok_emb.weight + passthrough (float16): blocks.attn.attn_gate_w, blocks.attn.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, skip_gates, skip_weights, smear_gate.weight, smear_lambda +Serialized model quantized+brotli: 15916636 bytes +Total submission size quantized+brotli: 15949941 bytes +diagnostic quantized val_loss:2.32387621 val_bpb:1.08122861 eval_time:7069ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (107.8s) + +beginning TTT eval timer +ttt_phased: total_docs:10000 prefix_docs:1500 suffix_docs:8500 num_phases:3 boundaries:[500, 1000, 1500] +ttp: b156/157 bl:2.2468 bb:1.0835 rl:2.2468 rb:1.0835 dl:7999-19941 gd:0 +ttpp: phase:1/3 pd:912 gd:500 t:215.0s +tttg: c1/38 lr:0.001000 t:0.3s +tttg: c2/38 lr:0.000998 t:0.4s +tttg: c3/38 lr:0.000993 t:0.5s +tttg: c4/38 lr:0.000984 t:0.6s +tttg: c5/38 lr:0.000971 t:0.7s +tttg: c6/38 lr:0.000956 t:0.8s +tttg: c7/38 lr:0.000937 t:0.8s +tttg: c8/38 lr:0.000914 t:0.9s +tttg: c9/38 lr:0.000889 t:1.0s +tttg: c10/38 lr:0.000861 t:1.1s +tttg: c11/38 lr:0.000830 t:1.2s +tttg: c12/38 lr:0.000797 t:1.2s +tttg: c13/38 lr:0.000762 t:1.3s +tttg: c14/38 lr:0.000725 t:1.4s +tttg: c15/38 lr:0.000686 t:1.5s +tttg: c16/38 lr:0.000646 t:1.5s +tttg: c17/38 lr:0.000605 t:1.6s +tttg: c18/38 lr:0.000564 t:1.7s +tttg: c19/38 lr:0.000521 t:1.8s +tttg: c20/38 lr:0.000479 t:1.9s +tttg: c21/38 lr:0.000436 t:1.9s +tttg: c22/38 lr:0.000395 t:2.0s +tttg: c23/38 lr:0.000354 t:2.1s +tttg: c24/38 lr:0.000314 t:2.2s +tttg: c25/38 lr:0.000275 t:2.2s +tttg: c26/38 lr:0.000238 t:2.3s +tttg: c27/38 lr:0.000203 t:2.4s +tttg: c28/38 lr:0.000170 t:2.5s +tttg: c29/38 lr:0.000139 t:2.6s +tttg: c30/38 lr:0.000111 t:2.6s +tttg: c31/38 lr:0.000086 t:2.7s +tttg: c32/38 lr:0.000063 t:2.8s +tttg: c33/38 lr:0.000044 t:2.9s +tttg: c34/38 lr:0.000029 t:3.0s +tttg: c35/38 lr:0.000016 t:3.0s +tttg: c36/38 lr:0.000007 t:3.1s +tttg: c37/38 lr:0.000002 t:3.2s +ttpr: phase:1/3 t:220.3s +ttp: b142/157 bl:2.3953 bb:1.0832 rl:2.2677 rb:1.0835 dl:1892-1953 gd:0 +ttpp: phase:2/3 pd:1488 gd:1000 t:278.1s +tttg: c1/61 lr:0.001000 t:0.1s +tttg: c2/61 lr:0.000999 t:0.2s +tttg: c3/61 lr:0.000997 t:0.2s +tttg: c4/61 lr:0.000994 t:0.3s +tttg: c5/61 lr:0.000989 t:0.4s +tttg: c6/61 lr:0.000983 t:0.5s +tttg: c7/61 lr:0.000976 t:0.5s +tttg: c8/61 lr:0.000967 t:0.6s +tttg: c9/61 lr:0.000957 t:0.7s +tttg: c10/61 lr:0.000946 t:0.8s +tttg: c11/61 lr:0.000933 t:0.8s +tttg: c12/61 lr:0.000919 t:2.5s +tttg: c13/61 lr:0.000905 t:2.6s +tttg: c14/61 lr:0.000889 t:2.7s +tttg: c15/61 lr:0.000872 t:2.8s +tttg: c16/61 lr:0.000854 t:2.8s +tttg: c17/61 lr:0.000835 t:2.9s +tttg: c18/61 lr:0.000815 t:3.0s +tttg: c19/61 lr:0.000794 t:3.1s +tttg: c20/61 lr:0.000772 t:3.1s +tttg: c21/61 lr:0.000750 t:3.2s +tttg: c22/61 lr:0.000727 t:3.3s +tttg: c23/61 lr:0.000703 t:3.4s +tttg: c24/61 lr:0.000679 t:3.4s +tttg: c25/61 lr:0.000655 t:3.5s +tttg: c26/61 lr:0.000629 t:3.6s +tttg: c27/61 lr:0.000604 t:3.7s +tttg: c28/61 lr:0.000578 t:3.8s +tttg: c29/61 lr:0.000552 t:3.8s +tttg: c30/61 lr:0.000526 t:3.9s +tttg: c31/61 lr:0.000500 t:4.0s +tttg: c32/61 lr:0.000474 t:4.1s +tttg: c33/61 lr:0.000448 t:4.2s +tttg: c34/61 lr:0.000422 t:4.2s +tttg: c35/61 lr:0.000396 t:4.3s +tttg: c36/61 lr:0.000371 t:4.4s +tttg: c37/61 lr:0.000345 t:4.5s +tttg: c38/61 lr:0.000321 t:4.5s +tttg: c39/61 lr:0.000297 t:4.6s +tttg: c40/61 lr:0.000273 t:4.7s +tttg: c41/61 lr:0.000250 t:4.8s +tttg: c42/61 lr:0.000228 t:4.9s +tttg: c43/61 lr:0.000206 t:4.9s +tttg: c44/61 lr:0.000185 t:5.0s +tttg: c45/61 lr:0.000165 t:5.1s +tttg: c46/61 lr:0.000146 t:5.2s +tttg: c47/61 lr:0.000128 t:5.2s +tttg: c48/61 lr:0.000111 t:5.3s +tttg: c49/61 lr:0.000095 t:5.4s +tttg: c50/61 lr:0.000081 t:5.5s +tttg: c51/61 lr:0.000067 t:5.6s +tttg: c52/61 lr:0.000054 t:5.6s +tttg: c53/61 lr:0.000043 t:5.7s +tttg: c54/61 lr:0.000033 t:5.8s +tttg: c55/61 lr:0.000024 t:5.9s +tttg: c56/61 lr:0.000017 t:5.9s +tttg: c57/61 lr:0.000011 t:6.0s +tttg: c58/61 lr:0.000006 t:6.1s +tttg: c59/61 lr:0.000003 t:6.2s +tttg: c60/61 lr:0.000001 t:6.3s +ttpr: phase:2/3 t:286.5s +ttp: b127/157 bl:2.4007 bb:1.0881 rl:2.2794 rb:1.0839 dl:1299-1329 gd:0 +ttpp: phase:3/3 pd:2000 gd:1500 t:289.3s +tttg: c1/82 lr:0.001000 t:0.1s +tttg: c2/82 lr:0.001000 t:0.1s +tttg: c3/82 lr:0.000998 t:0.2s +tttg: c4/82 lr:0.000997 t:0.3s +tttg: c5/82 lr:0.000994 t:0.4s +tttg: c6/82 lr:0.000991 t:0.4s +tttg: c7/82 lr:0.000987 t:0.5s +tttg: c8/82 lr:0.000982 t:0.6s +tttg: c9/82 lr:0.000976 t:0.7s +tttg: c10/82 lr:0.000970 t:0.8s +tttg: c11/82 lr:0.000963 t:0.8s +tttg: c12/82 lr:0.000955 t:0.9s +tttg: c13/82 lr:0.000947 t:1.0s +tttg: c14/82 lr:0.000938 t:1.1s +tttg: c15/82 lr:0.000928 t:1.1s +tttg: c16/82 lr:0.000918 t:1.2s +tttg: c17/82 lr:0.000907 t:1.3s +tttg: c18/82 lr:0.000895 t:1.4s +tttg: c19/82 lr:0.000883 t:1.5s +tttg: c20/82 lr:0.000870 t:1.5s +tttg: c21/82 lr:0.000857 t:1.6s +tttg: c22/82 lr:0.000843 t:1.7s +tttg: c23/82 lr:0.000829 t:1.8s +tttg: c24/82 lr:0.000814 t:1.8s +tttg: c25/82 lr:0.000799 t:1.9s +tttg: c26/82 lr:0.000783 t:2.0s +tttg: c27/82 lr:0.000767 t:2.1s +tttg: c28/82 lr:0.000750 t:2.1s +tttg: c29/82 lr:0.000733 t:2.2s +tttg: c30/82 lr:0.000716 t:2.3s +tttg: c31/82 lr:0.000698 t:2.4s +tttg: c32/82 lr:0.000680 t:2.5s +tttg: c33/82 lr:0.000662 t:2.5s +tttg: c34/82 lr:0.000643 t:2.6s +tttg: c35/82 lr:0.000625 t:2.7s +tttg: c36/82 lr:0.000606 t:2.8s +tttg: c37/82 lr:0.000587 t:2.9s +tttg: c38/82 lr:0.000568 t:2.9s +tttg: c39/82 lr:0.000548 t:3.0s +tttg: c40/82 lr:0.000529 t:3.1s +tttg: c41/82 lr:0.000510 t:3.2s +tttg: c42/82 lr:0.000490 t:3.2s +tttg: c43/82 lr:0.000471 t:3.3s +tttg: c44/82 lr:0.000452 t:3.4s +tttg: c45/82 lr:0.000432 t:3.5s +tttg: c46/82 lr:0.000413 t:3.5s +tttg: c47/82 lr:0.000394 t:3.6s +tttg: c48/82 lr:0.000375 t:3.7s +tttg: c49/82 lr:0.000357 t:3.8s +tttg: c50/82 lr:0.000338 t:3.9s +tttg: c51/82 lr:0.000320 t:4.0s +tttg: c52/82 lr:0.000302 t:4.0s +tttg: c53/82 lr:0.000284 t:4.1s +tttg: c54/82 lr:0.000267 t:4.2s +tttg: c55/82 lr:0.000250 t:4.3s +tttg: c56/82 lr:0.000233 t:4.3s +tttg: c57/82 lr:0.000217 t:4.4s +tttg: c58/82 lr:0.000201 t:4.5s +tttg: c59/82 lr:0.000186 t:4.6s +tttg: c60/82 lr:0.000171 t:4.7s +tttg: c61/82 lr:0.000157 t:4.7s +tttg: c62/82 lr:0.000143 t:4.8s +tttg: c63/82 lr:0.000130 t:4.9s +tttg: c64/82 lr:0.000117 t:5.0s +tttg: c65/82 lr:0.000105 t:5.1s +tttg: c66/82 lr:0.000093 t:5.1s +tttg: c67/82 lr:0.000082 t:5.2s +tttg: c68/82 lr:0.000072 t:5.3s +tttg: c69/82 lr:0.000062 t:5.4s +tttg: c70/82 lr:0.000053 t:5.5s +tttg: c71/82 lr:0.000045 t:5.5s +tttg: c72/82 lr:0.000037 t:5.6s +tttg: c73/82 lr:0.000030 t:5.7s +tttg: c74/82 lr:0.000024 t:5.8s +tttg: c75/82 lr:0.000018 t:5.9s +tttg: c76/82 lr:0.000013 t:5.9s +tttg: c77/82 lr:0.000009 t:6.0s +tttg: c78/82 lr:0.000006 t:6.1s +tttg: c79/82 lr:0.000003 t:6.2s +tttg: c80/82 lr:0.000002 t:6.3s +tttg: c81/82 lr:0.000000 t:6.3s +ttpr: phase:3/3 t:297.7s +ttp: b121/157 bl:2.3022 bb:1.0585 rl:2.2810 rb:1.0820 dl:1144-1164 gd:1 +ttp: b113/157 bl:2.2375 bb:1.0236 rl:2.2785 rb:1.0786 dl:965-988 gd:1 +ttp: b106/157 bl:2.3305 bb:1.0477 rl:2.2810 rb:1.0770 dl:849-860 gd:1 +ttp: b102/157 bl:2.3576 bb:1.0827 rl:2.2843 rb:1.0773 dl:795-808 gd:1 +ttp: b94/157 bl:2.2896 bb:1.0659 rl:2.2845 rb:1.0769 dl:695-707 gd:1 +ttp: b82/157 bl:2.2691 bb:1.0404 rl:2.2840 rb:1.0758 dl:565-575 gd:1 +ttp: b78/157 bl:2.3267 bb:1.0668 rl:2.2851 rb:1.0755 dl:528-536 gd:1 +ttp: b66/157 bl:2.2534 bb:1.0915 rl:2.2845 rb:1.0759 dl:436-442 gd:1 +ttp: b60/157 bl:2.3746 bb:1.0866 rl:2.2861 rb:1.0761 dl:393-399 gd:1 +ttp: b54/157 bl:2.4066 bb:1.1258 rl:2.2881 rb:1.0769 dl:356-362 gd:1 +ttp: b49/157 bl:2.4400 bb:1.1287 rl:2.2904 rb:1.0777 dl:327-333 gd:1 +ttp: b43/157 bl:2.3303 bb:1.1350 rl:2.2909 rb:1.0784 dl:296-300 gd:1 +ttp: b37/157 bl:2.5166 bb:1.1744 rl:2.2936 rb:1.0796 dl:266-269 gd:1 +ttp: b31/157 bl:2.4341 bb:1.1370 rl:2.2951 rb:1.0802 dl:240-244 gd:1 +ttp: b24/157 bl:2.4826 bb:1.1577 rl:2.2968 rb:1.0809 dl:210-215 gd:1 +ttp: b17/157 bl:2.4620 bb:1.1728 rl:2.2981 rb:1.0816 dl:179-183 gd:1 +ttp: b10/157 bl:2.5684 bb:1.2296 rl:2.2998 rb:1.0825 dl:148-153 gd:1 +ttp: b3/157 bl:2.7263 bb:1.2738 rl:2.3018 rb:1.0834 dl:107-114 gd:1 +ppm_mix bytes=13517716 mix_bpb=0.862121 ppm_only=2.139538 nn_only=0.969095 bin0_pct=0.136 bin1_pct=0.125 bin2_pct=0.085 bin3_pct=0.654 +ppm_mix_time:58.6s subset=4000000 tokens +quantized_ttt_phased val_loss:2.29665817 val_bpb:1.06893377 eval_time:370990ms +total_eval_time:371.0s diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed314.log b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed314.log new file mode 100644 index 0000000000..96cbb131ca --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed314.log @@ -0,0 +1,408 @@ +W0428 02:49:35.159000 275901 torch/distributed/run.py:803] +W0428 02:49:35.159000 275901 torch/distributed/run.py:803] ***************************************** +W0428 02:49:35.159000 275901 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. +W0428 02:49:35.159000 275901 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + beta1: 0.9 + beta2: 0.95 + caseops_enabled: True + compressor: brotli + data_dir: ./data/ + datasets_dir: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2048 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.01 + gated_attn_quant_gate: False + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 0.5 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/4afef68d-94b3-4181-ac3d-00834c9124d3.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_rank: 4 + lqer_top_k: 3 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 3 + phased_ttt_prefix_docs: 1500 + ppm_conf_threshold: 0.9 + ppm_enabled: True + ppm_lambda_hi: 0.9 + ppm_lambda_lo: 0.05 + ppm_order: 5 + ppm_subset_tokens: 4000000 + qk_gain_init: 5.0 + quantized_model_path: final_model.int6.ptz + rank: 0 + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: 4afef68d-94b3-4181-ac3d-00834c9124d3 + scalar_lr: 0.02 + seed: 314 + skip_gates_enabled: True + smear_gate_enabled: True + sparse_attn_gate_enabled: True + sparse_attn_gate_init_std: 0.0 + sparse_attn_gate_scale: 1.0 + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: /workspace/work_mixer/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_train_*.bin + train_log_every: 500 + train_seq_len: 2048 + ttt_batch_size: 64 + ttt_beta1: 0.0 + ttt_beta2: 0.999 + ttt_chunk_size: 48 + ttt_enabled: True + ttt_eval_batches: + ttt_eval_seq_len: 2048 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 1.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 488 +val_tokens: 9662464 +model_params:35945671 +gptq:reserving 0s, effective=599500ms +warmup_cu_buckets:64,128,192,256 iters_each:3 +warmup_step: 1/20 +warmup_step: 2/20 +warmup_step: 3/20 +warmup_step: 4/20 +warmup_step: 5/20 +warmup_step: 6/20 +warmup_step: 10/20 +warmup_step: 20/20 +loop_warmup:enabled encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +loop_warmup_step: 1/20 +loop_warmup_step: 2/20 +loop_warmup_step: 3/20 +loop_warmup_step: 4/20 +loop_warmup_step: 5/20 +loop_warmup_step: 6/20 +loop_warmup_step: 10/20 +loop_warmup_step: 20/20 +0/20000 val_loss: 8.9968 val_bpb: 4.1860 +1/20000 train_loss: 8.9992 train_time: 0.0m tok/s: 12227506 +2/20000 train_loss: 12.8309 train_time: 0.0m tok/s: 11607043 +3/20000 train_loss: 10.1387 train_time: 0.0m tok/s: 10401114 +4/20000 train_loss: 8.6111 train_time: 0.0m tok/s: 9834310 +5/20000 train_loss: 7.8120 train_time: 0.0m tok/s: 9541011 +500/20000 train_loss: 2.7145 train_time: 0.8m tok/s: 8390584 +1000/20000 train_loss: 2.7695 train_time: 1.6m tok/s: 8365799 +1500/20000 train_loss: 2.5869 train_time: 2.4m tok/s: 8360616 +2000/20000 train_loss: 2.6795 train_time: 3.1m tok/s: 8358807 +layer_loop:enabled step:2230 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: 2.4727 train_time: 4.1m tok/s: 7946915 +3000/20000 train_loss: 2.5374 train_time: 5.3m tok/s: 7413686 +3500/20000 train_loss: 2.4209 train_time: 6.5m tok/s: 7098373 +4000/20000 train_loss: 2.4280 train_time: 7.6m tok/s: 6880521 +4000/20000 val_loss: 2.4093 val_bpb: 1.1210 +4500/20000 train_loss: 2.3115 train_time: 8.8m tok/s: 6719805 +5000/20000 train_loss: 2.4485 train_time: 9.9m tok/s: 6596988 +5025/20000 val_loss: 2.3262 val_bpb: 1.0823 +stopping_early: wallclock_cap train_time: 599577ms step: 5025/20000 +peak memory allocated: 41709 MiB reserved: 47026 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.30116619 val_bpb:1.07066233 eval_time:6418ms +Serialized model: 135417533 bytes +Code size (uncompressed): 161565 bytes +Code size (compressed): 33305 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 3.5s +Quantized weights: + gptq (int6): blocks.attn.c_k.weight, blocks.attn.c_q.weight, blocks.attn.c_v.weight, blocks.attn.proj.weight, blocks.mlp.fc.weight, blocks.mlp.proj.weight + gptq (int6)+lqer_asym: blocks.mlp.fc.weight + gptq (int7)+lqer_asym: tok_emb.weight + passthrough (float16): blocks.attn.attn_gate_w, blocks.attn.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, skip_gates, skip_weights, smear_gate.weight, smear_lambda +Serialized model quantized+brotli: 15917890 bytes +Total submission size quantized+brotli: 15951195 bytes +diagnostic quantized val_loss:2.32103418 val_bpb:1.07990631 eval_time:7094ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (104.2s) + +beginning TTT eval timer +ttt_phased: total_docs:10000 prefix_docs:1500 suffix_docs:8500 num_phases:3 boundaries:[500, 1000, 1500] +ttp: b157/157 bl:2.0310 bb:1.0027 rl:2.0310 rb:1.0027 dl:20000-97114 gd:0 +ttpp: phase:1/3 pd:912 gd:500 t:166.6s +tttg: c1/38 lr:0.001000 t:0.3s +tttg: c2/38 lr:0.000998 t:0.4s +tttg: c3/38 lr:0.000993 t:0.5s +tttg: c4/38 lr:0.000984 t:0.6s +tttg: c5/38 lr:0.000971 t:0.7s +tttg: c6/38 lr:0.000956 t:0.8s +tttg: c7/38 lr:0.000937 t:0.8s +tttg: c8/38 lr:0.000914 t:0.9s +tttg: c9/38 lr:0.000889 t:1.0s +tttg: c10/38 lr:0.000861 t:1.1s +tttg: c11/38 lr:0.000830 t:1.2s +tttg: c12/38 lr:0.000797 t:1.2s +tttg: c13/38 lr:0.000762 t:1.3s +tttg: c14/38 lr:0.000725 t:1.4s +tttg: c15/38 lr:0.000686 t:1.5s +tttg: c16/38 lr:0.000646 t:1.5s +tttg: c17/38 lr:0.000605 t:1.6s +tttg: c18/38 lr:0.000564 t:1.7s +tttg: c19/38 lr:0.000521 t:1.8s +tttg: c20/38 lr:0.000479 t:1.8s +tttg: c21/38 lr:0.000436 t:1.9s +tttg: c22/38 lr:0.000395 t:2.0s +tttg: c23/38 lr:0.000354 t:2.1s +tttg: c24/38 lr:0.000314 t:2.2s +tttg: c25/38 lr:0.000275 t:2.2s +tttg: c26/38 lr:0.000238 t:2.3s +tttg: c27/38 lr:0.000203 t:2.4s +tttg: c28/38 lr:0.000170 t:2.5s +tttg: c29/38 lr:0.000139 t:2.5s +tttg: c30/38 lr:0.000111 t:2.6s +tttg: c31/38 lr:0.000086 t:2.7s +tttg: c32/38 lr:0.000063 t:2.8s +tttg: c33/38 lr:0.000044 t:2.8s +tttg: c34/38 lr:0.000029 t:2.9s +tttg: c35/38 lr:0.000016 t:3.0s +tttg: c36/38 lr:0.000007 t:3.1s +tttg: c37/38 lr:0.000002 t:3.1s +ttpr: phase:1/3 t:171.9s +ttp: b136/157 bl:2.3440 bb:1.0710 rl:2.0862 rb:1.0155 dl:1597-1633 gd:0 +ttpp: phase:2/3 pd:1488 gd:1000 t:229.1s +tttg: c1/61 lr:0.001000 t:0.1s +tttg: c2/61 lr:0.000999 t:0.2s +tttg: c3/61 lr:0.000997 t:0.2s +tttg: c4/61 lr:0.000994 t:0.3s +tttg: c5/61 lr:0.000989 t:0.4s +tttg: c6/61 lr:0.000983 t:0.5s +tttg: c7/61 lr:0.000976 t:0.5s +tttg: c8/61 lr:0.000967 t:0.6s +tttg: c9/61 lr:0.000957 t:0.7s +tttg: c10/61 lr:0.000946 t:0.8s +tttg: c11/61 lr:0.000933 t:0.8s +tttg: c12/61 lr:0.000919 t:0.9s +tttg: c13/61 lr:0.000905 t:1.0s +tttg: c14/61 lr:0.000889 t:1.1s +tttg: c15/61 lr:0.000872 t:1.2s +tttg: c16/61 lr:0.000854 t:1.2s +tttg: c17/61 lr:0.000835 t:1.3s +tttg: c18/61 lr:0.000815 t:1.4s +tttg: c19/61 lr:0.000794 t:1.5s +tttg: c20/61 lr:0.000772 t:1.5s +tttg: c21/61 lr:0.000750 t:1.6s +tttg: c22/61 lr:0.000727 t:1.7s +tttg: c23/61 lr:0.000703 t:1.8s +tttg: c24/61 lr:0.000679 t:1.9s +tttg: c25/61 lr:0.000655 t:1.9s +tttg: c26/61 lr:0.000629 t:2.0s +tttg: c27/61 lr:0.000604 t:2.1s +tttg: c28/61 lr:0.000578 t:2.2s +tttg: c29/61 lr:0.000552 t:2.2s +tttg: c30/61 lr:0.000526 t:2.3s +tttg: c31/61 lr:0.000500 t:2.4s +tttg: c32/61 lr:0.000474 t:2.5s +tttg: c33/61 lr:0.000448 t:2.6s +tttg: c34/61 lr:0.000422 t:2.6s +tttg: c35/61 lr:0.000396 t:2.7s +tttg: c36/61 lr:0.000371 t:2.8s +tttg: c37/61 lr:0.000345 t:2.9s +tttg: c38/61 lr:0.000321 t:2.9s +tttg: c39/61 lr:0.000297 t:3.0s +tttg: c40/61 lr:0.000273 t:3.1s +tttg: c41/61 lr:0.000250 t:3.2s +tttg: c42/61 lr:0.000228 t:3.3s +tttg: c43/61 lr:0.000206 t:3.3s +tttg: c44/61 lr:0.000185 t:3.4s +tttg: c45/61 lr:0.000165 t:3.5s +tttg: c46/61 lr:0.000146 t:3.6s +tttg: c47/61 lr:0.000128 t:3.7s +tttg: c48/61 lr:0.000111 t:3.7s +tttg: c49/61 lr:0.000095 t:3.8s +tttg: c50/61 lr:0.000081 t:3.9s +tttg: c51/61 lr:0.000067 t:4.0s +tttg: c52/61 lr:0.000054 t:4.1s +tttg: c53/61 lr:0.000043 t:4.1s +tttg: c54/61 lr:0.000033 t:4.2s +tttg: c55/61 lr:0.000024 t:4.3s +tttg: c56/61 lr:0.000017 t:4.4s +tttg: c57/61 lr:0.000011 t:4.4s +tttg: c58/61 lr:0.000006 t:4.5s +tttg: c59/61 lr:0.000003 t:4.6s +tttg: c60/61 lr:0.000001 t:4.7s +ttpr: phase:2/3 t:235.9s +ttp: b131/157 bl:2.3328 bb:1.0640 rl:2.1196 rb:1.0225 dl:1426-1451 gd:0 +ttpp: phase:3/3 pd:2000 gd:1500 t:238.7s +tttg: c1/82 lr:0.001000 t:0.1s +tttg: c2/82 lr:0.001000 t:0.1s +tttg: c3/82 lr:0.000998 t:0.2s +tttg: c4/82 lr:0.000997 t:0.3s +tttg: c5/82 lr:0.000994 t:0.4s +tttg: c6/82 lr:0.000991 t:0.4s +tttg: c7/82 lr:0.000987 t:0.5s +tttg: c8/82 lr:0.000982 t:0.6s +tttg: c9/82 lr:0.000976 t:0.7s +tttg: c10/82 lr:0.000970 t:0.8s +tttg: c11/82 lr:0.000963 t:0.8s +tttg: c12/82 lr:0.000955 t:0.9s +tttg: c13/82 lr:0.000947 t:1.0s +tttg: c14/82 lr:0.000938 t:1.1s +tttg: c15/82 lr:0.000928 t:1.1s +tttg: c16/82 lr:0.000918 t:1.2s +tttg: c17/82 lr:0.000907 t:1.3s +tttg: c18/82 lr:0.000895 t:1.4s +tttg: c19/82 lr:0.000883 t:1.5s +tttg: c20/82 lr:0.000870 t:1.5s +tttg: c21/82 lr:0.000857 t:1.6s +tttg: c22/82 lr:0.000843 t:1.7s +tttg: c23/82 lr:0.000829 t:1.8s +tttg: c24/82 lr:0.000814 t:1.9s +tttg: c25/82 lr:0.000799 t:1.9s +tttg: c26/82 lr:0.000783 t:2.0s +tttg: c27/82 lr:0.000767 t:2.1s +tttg: c28/82 lr:0.000750 t:2.2s +tttg: c29/82 lr:0.000733 t:3.9s +tttg: c30/82 lr:0.000716 t:4.0s +tttg: c31/82 lr:0.000698 t:4.1s +tttg: c32/82 lr:0.000680 t:4.1s +tttg: c33/82 lr:0.000662 t:4.2s +tttg: c34/82 lr:0.000643 t:4.3s +tttg: c35/82 lr:0.000625 t:4.4s +tttg: c36/82 lr:0.000606 t:4.4s +tttg: c37/82 lr:0.000587 t:4.5s +tttg: c38/82 lr:0.000568 t:4.6s +tttg: c39/82 lr:0.000548 t:4.7s +tttg: c40/82 lr:0.000529 t:4.8s +tttg: c41/82 lr:0.000510 t:4.8s +tttg: c42/82 lr:0.000490 t:4.9s +tttg: c43/82 lr:0.000471 t:5.0s +tttg: c44/82 lr:0.000452 t:5.1s +tttg: c45/82 lr:0.000432 t:5.2s +tttg: c46/82 lr:0.000413 t:5.2s +tttg: c47/82 lr:0.000394 t:5.3s +tttg: c48/82 lr:0.000375 t:5.4s +tttg: c49/82 lr:0.000357 t:5.5s +tttg: c50/82 lr:0.000338 t:5.5s +tttg: c51/82 lr:0.000320 t:5.6s +tttg: c52/82 lr:0.000302 t:5.7s +tttg: c53/82 lr:0.000284 t:5.8s +tttg: c54/82 lr:0.000267 t:5.8s +tttg: c55/82 lr:0.000250 t:5.9s +tttg: c56/82 lr:0.000233 t:6.0s +tttg: c57/82 lr:0.000217 t:6.1s +tttg: c58/82 lr:0.000201 t:6.2s +tttg: c59/82 lr:0.000186 t:6.3s +tttg: c60/82 lr:0.000171 t:6.3s +tttg: c61/82 lr:0.000157 t:6.4s +tttg: c62/82 lr:0.000143 t:6.5s +tttg: c63/82 lr:0.000130 t:6.6s +tttg: c64/82 lr:0.000117 t:6.6s +tttg: c65/82 lr:0.000105 t:6.7s +tttg: c66/82 lr:0.000093 t:6.8s +tttg: c67/82 lr:0.000082 t:6.9s +tttg: c68/82 lr:0.000072 t:7.0s +tttg: c69/82 lr:0.000062 t:7.0s +tttg: c70/82 lr:0.000053 t:7.1s +tttg: c71/82 lr:0.000045 t:7.2s +tttg: c72/82 lr:0.000037 t:7.3s +tttg: c73/82 lr:0.000030 t:7.3s +tttg: c74/82 lr:0.000024 t:7.4s +tttg: c75/82 lr:0.000018 t:7.5s +tttg: c76/82 lr:0.000013 t:7.6s +tttg: c77/82 lr:0.000009 t:7.7s +tttg: c78/82 lr:0.000006 t:7.7s +tttg: c79/82 lr:0.000003 t:7.8s +tttg: c80/82 lr:0.000002 t:7.9s +tttg: c81/82 lr:0.000000 t:8.0s +ttpr: phase:3/3 t:248.8s +ttp: b118/157 bl:2.3518 bb:1.0793 rl:2.1411 rb:1.0280 dl:1069-1090 gd:1 +ttp: b117/157 bl:2.3456 bb:1.0595 rl:2.1580 rb:1.0307 dl:1043-1068 gd:1 +ttp: b110/157 bl:2.3501 bb:1.0473 rl:2.1710 rb:1.0319 dl:915-929 gd:1 +ttp: b99/157 bl:2.3461 bb:1.0474 rl:2.1802 rb:1.0328 dl:754-766 gd:1 +ttp: b95/157 bl:2.3464 bb:1.0908 rl:2.1880 rb:1.0356 dl:707-718 gd:1 +ttp: b85/157 bl:2.3532 bb:1.0895 rl:2.1943 rb:1.0377 dl:596-607 gd:1 +ttp: b77/157 bl:2.3647 bb:1.1090 rl:2.1998 rb:1.0400 dl:520-528 gd:1 +ttp: b72/157 bl:2.3257 bb:1.0737 rl:2.2035 rb:1.0410 dl:481-488 gd:1 +ttp: b60/157 bl:2.3692 bb:1.0842 rl:2.2073 rb:1.0420 dl:393-399 gd:1 +ttp: b52/157 bl:2.3923 bb:1.1066 rl:2.2109 rb:1.0433 dl:344-350 gd:1 +ttp: b47/157 bl:2.4003 bb:1.1463 rl:2.2143 rb:1.0451 dl:315-322 gd:1 +ttp: b42/157 bl:2.4170 bb:1.1378 rl:2.2176 rb:1.0466 dl:290-296 gd:1 +ttp: b37/157 bl:2.5112 bb:1.1718 rl:2.2219 rb:1.0485 dl:266-269 gd:1 +ttp: b32/157 bl:2.3553 bb:1.1338 rl:2.2236 rb:1.0496 dl:244-248 gd:1 +ttp: b23/157 bl:2.4093 bb:1.1158 rl:2.2257 rb:1.0503 dl:206-210 gd:1 +ttp: b18/157 bl:2.5196 bb:1.2283 rl:2.2285 rb:1.0520 dl:184-187 gd:1 +ttp: b13/157 bl:2.5262 bb:1.2666 rl:2.2311 rb:1.0537 dl:162-166 gd:1 +ttp: b8/157 bl:2.5861 bb:1.2543 rl:2.2336 rb:1.0551 dl:138-143 gd:1 +ttp: b1/157 bl:2.7454 bb:1.1896 rl:2.2359 rb:1.0558 dl:69-97 gd:1 +ppm_mix bytes=13517716 mix_bpb=0.860991 ppm_only=2.139538 nn_only=0.967720 bin0_pct=0.136 bin1_pct=0.125 bin2_pct=0.085 bin3_pct=0.654 +ppm_mix_time:60.5s subset=4000000 tokens +quantized_ttt_phased val_loss:2.29361134 val_bpb:1.06751568 eval_time:327745ms +total_eval_time:327.7s diff --git a/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed42.log b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed42.log new file mode 100644 index 0000000000..78f43f0b65 --- /dev/null +++ b/records/track_10min_16mb/2026-04-28_PR1854_BudgetCompliant_1.0678/train_seed42.log @@ -0,0 +1,407 @@ +W0428 02:04:00.932000 236850 torch/distributed/run.py:803] +W0428 02:04:00.932000 236850 torch/distributed/run.py:803] ***************************************** +W0428 02:04:00.932000 236850 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. +W0428 02:04:00.932000 236850 torch/distributed/run.py:803] ***************************************** +Hyperparameters: + adam_eps: 1e-08 + adam_wd: 0.02 + artifact_dir: + attn_clip_sigmas: 13.0 + attn_out_gate_enabled: False + attn_out_gate_src: proj + beta1: 0.9 + beta2: 0.95 + caseops_enabled: True + compressor: brotli + data_dir: ./data/ + datasets_dir: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved + distributed: True + ema_decay: 0.9965 + embed_bits: 7 + embed_clip_sigmas: 15.0 + embed_lr: 0.6 + embed_wd: 0.085 + enable_looping_at: 0.35 + eval_seq_len: 2048 + eval_stride: 64 + fused_ce_enabled: True + gate_window: 12 + gated_attn_enabled: False + gated_attn_init_std: 0.01 + gated_attn_quant_gate: False + global_ttt_batch_seqs: 32 + global_ttt_chunk_tokens: 32768 + global_ttt_epochs: 1 + global_ttt_grad_clip: 1.0 + global_ttt_lr: 0.001 + global_ttt_momentum: 0.9 + global_ttt_respect_doc_boundaries: True + global_ttt_warmup_chunks: 0 + global_ttt_warmup_start_lr: 0.0 + gptq_calibration_batches: 16 + gptq_reserve_seconds: 0.5 + grad_accum_steps: 1 + grad_clip_norm: 0.3 + is_main_process: True + iterations: 20000 + ln_scale: True + local_rank: 0 + logfile: logs/0b6f94e5-2cc2-4987-920a-dddd1a780b69.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_rank: 4 + lqer_top_k: 3 + matrix_bits: 6 + matrix_clip_sigmas: 12.85 + matrix_lr: 0.026 + max_wallclock_seconds: 600.0 + min_lr: 0.1 + mlp_clip_sigmas: 12.0 + mlp_mult: 4.0 + model_dim: 512 + model_path: final_model.pt + muon_backend_steps: 5 + muon_momentum: 0.97 + muon_momentum_warmup_start: 0.92 + muon_momentum_warmup_steps: 1500 + muon_row_normalize: True + muon_wd: 0.095 + num_heads: 8 + num_kv_heads: 4 + num_layers: 11 + num_loops: 2 + parallel_final_lane: mean + parallel_start_layer: 8 + phased_ttt_num_phases: 3 + phased_ttt_prefix_docs: 1500 + ppm_conf_threshold: 0.9 + ppm_enabled: True + ppm_lambda_hi: 0.9 + ppm_lambda_lo: 0.05 + ppm_order: 5 + ppm_subset_tokens: 4000000 + qk_gain_init: 5.0 + quantized_model_path: final_model.int6.ptz + rank: 0 + rope_base: 10000.0 + rope_dims: 16 + rope_train_seq_len: 2048 + rope_yarn: False + run_id: 0b6f94e5-2cc2-4987-920a-dddd1a780b69 + 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: 1.0 + tie_embeddings: True + tied_embed_init_std: 0.005 + tied_embed_lr: 0.03 + tokenizer_path: /workspace/work_mixer/tokenizers/fineweb_8192_bpe_lossless_caps_caseops_v1_reserved.model + train_batch_tokens: 786432 + train_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_train_*.bin + train_log_every: 500 + train_seq_len: 2048 + ttt_batch_size: 64 + ttt_beta1: 0.0 + ttt_beta2: 0.999 + ttt_chunk_size: 48 + ttt_enabled: True + ttt_eval_batches: + ttt_eval_seq_len: 2048 + ttt_grad_steps: 1 + ttt_k_lora: True + ttt_lora_lr: 0.0001 + ttt_lora_rank: 96 + ttt_mlp_lora: True + ttt_o_lora: True + ttt_optimizer: adam + ttt_weight_decay: 1.0 + val_batch_tokens: 524288 + val_bytes_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_bytes_*.bin + val_doc_fraction: 1.0 + val_files: /workspace/parameter-golf/data/datasets/fineweb10B_sp8192_caseops/datasets/datasets/fineweb10B_sp8192_lossless_caps_caseops_v1_reserved/fineweb_val_*.bin + val_loss_every: 4000 + vocab_size: 8192 + warmdown_frac: 0.75 + warmup_steps: 20 + world_size: 8 + xsa_last_n: 11 +train_shards: 488 +val_tokens: 9662464 +model_params:35945671 +gptq:reserving 0s, effective=599500ms +warmup_cu_buckets:64,128,192,256 iters_each:3 +warmup_step: 1/20 +warmup_step: 2/20 +warmup_step: 3/20 +warmup_step: 4/20 +warmup_step: 5/20 +warmup_step: 6/20 +warmup_step: 10/20 +warmup_step: 20/20 +loop_warmup:enabled encoder:[0, 1, 2, 3, 4, 5, 3, 4] decoder:[5, 3, 4, 5, 6, 7, 8, 9, 10] +loop_warmup_step: 1/20 +loop_warmup_step: 2/20 +loop_warmup_step: 3/20 +loop_warmup_step: 4/20 +loop_warmup_step: 5/20 +loop_warmup_step: 6/20 +loop_warmup_step: 10/20 +loop_warmup_step: 20/20 +0/20000 val_loss: 9.0065 val_bpb: 4.1905 +1/20000 train_loss: 9.0073 train_time: 0.0m tok/s: 12230131 +2/20000 train_loss: 12.8046 train_time: 0.0m tok/s: 11603246 +3/20000 train_loss: 10.1376 train_time: 0.0m tok/s: 10356655 +4/20000 train_loss: 8.6391 train_time: 0.0m tok/s: 9795431 +5/20000 train_loss: 7.8854 train_time: 0.0m tok/s: 9515298 +500/20000 train_loss: 2.7226 train_time: 0.8m tok/s: 8392481 +1000/20000 train_loss: 2.7599 train_time: 1.6m tok/s: 8368441 +1500/20000 train_loss: 2.5913 train_time: 2.4m tok/s: 8363707 +2000/20000 train_loss: 2.6812 train_time: 3.1m tok/s: 8361218 +layer_loop:enabled step:2230 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: 2.4761 train_time: 4.1m tok/s: 7946317 +3000/20000 train_loss: 2.5362 train_time: 5.3m tok/s: 7413166 +3500/20000 train_loss: 2.4193 train_time: 6.5m tok/s: 7097366 +4000/20000 train_loss: 2.4262 train_time: 7.6m tok/s: 6879319 +4000/20000 val_loss: 2.4081 val_bpb: 1.1204 +4500/20000 train_loss: 2.3062 train_time: 8.8m tok/s: 6719632 +5000/20000 train_loss: 2.4491 train_time: 9.9m tok/s: 6597281 +5025/20000 val_loss: 2.3245 val_bpb: 1.0815 +stopping_early: wallclock_cap train_time: 599550ms step: 5025/20000 +peak memory allocated: 41709 MiB reserved: 47026 MiB +ema:applying EMA weights +diagnostic pre-quantization post-ema val_loss:2.29898278 val_bpb:1.06964646 eval_time:6493ms +Serialized model: 135417533 bytes +Code size (uncompressed): 161565 bytes +Code size (compressed): 33305 bytes +GPTQ:collecting Hessians from calibration data... +GPTQ:collected 67 Hessians in 3.5s +Quantized weights: + gptq (int6): blocks.attn.c_k.weight, blocks.attn.c_q.weight, blocks.attn.c_v.weight, blocks.attn.proj.weight, blocks.mlp.fc.weight, blocks.mlp.proj.weight + gptq (int6)+lqer_asym: blocks.mlp.fc.weight + gptq (int7)+lqer_asym: tok_emb.weight + passthrough (float16): blocks.attn.attn_gate_w, blocks.attn.q_gain, blocks.attn_scale, blocks.mlp_scale, blocks.resid_mix, parallel_post_lambdas, parallel_resid_lambdas, skip_gates, skip_weights, smear_gate.weight, smear_lambda +Serialized model quantized+brotli: 15918781 bytes +Total submission size quantized+brotli: 15952086 bytes +diagnostic quantized val_loss:2.31962379 val_bpb:1.07925009 eval_time:7028ms +ttt_lora:warming up compile (random tokens, no val data) +ttt_lora:compile warmup done (104.8s) + +beginning TTT eval timer +ttt_phased: total_docs:10000 prefix_docs:1500 suffix_docs:8500 num_phases:3 boundaries:[500, 1000, 1500] +ttp: b156/157 bl:2.2437 bb:1.0820 rl:2.2437 rb:1.0820 dl:7999-19941 gd:0 +ttpp: phase:1/3 pd:912 gd:500 t:215.6s +tttg: c1/38 lr:0.001000 t:0.3s +tttg: c2/38 lr:0.000998 t:0.4s +tttg: c3/38 lr:0.000993 t:0.5s +tttg: c4/38 lr:0.000984 t:0.5s +tttg: c5/38 lr:0.000971 t:0.6s +tttg: c6/38 lr:0.000956 t:0.7s +tttg: c7/38 lr:0.000937 t:0.8s +tttg: c8/38 lr:0.000914 t:0.8s +tttg: c9/38 lr:0.000889 t:0.9s +tttg: c10/38 lr:0.000861 t:1.0s +tttg: c11/38 lr:0.000830 t:1.1s +tttg: c12/38 lr:0.000797 t:1.2s +tttg: c13/38 lr:0.000762 t:1.2s +tttg: c14/38 lr:0.000725 t:1.3s +tttg: c15/38 lr:0.000686 t:1.4s +tttg: c16/38 lr:0.000646 t:1.5s +tttg: c17/38 lr:0.000605 t:1.6s +tttg: c18/38 lr:0.000564 t:1.6s +tttg: c19/38 lr:0.000521 t:1.7s +tttg: c20/38 lr:0.000479 t:1.8s +tttg: c21/38 lr:0.000436 t:1.9s +tttg: c22/38 lr:0.000395 t:2.0s +tttg: c23/38 lr:0.000354 t:2.0s +tttg: c24/38 lr:0.000314 t:2.1s +tttg: c25/38 lr:0.000275 t:2.2s +tttg: c26/38 lr:0.000238 t:2.3s +tttg: c27/38 lr:0.000203 t:2.3s +tttg: c28/38 lr:0.000170 t:2.4s +tttg: c29/38 lr:0.000139 t:2.5s +tttg: c30/38 lr:0.000111 t:2.6s +tttg: c31/38 lr:0.000086 t:2.7s +tttg: c32/38 lr:0.000063 t:2.7s +tttg: c33/38 lr:0.000044 t:2.8s +tttg: c34/38 lr:0.000029 t:2.9s +tttg: c35/38 lr:0.000016 t:3.0s +tttg: c36/38 lr:0.000007 t:3.0s +tttg: c37/38 lr:0.000002 t:3.1s +ttpr: phase:1/3 t:220.9s +ttp: b141/157 bl:2.3362 bb:1.0777 rl:2.2564 rb:1.0814 dl:1833-1891 gd:0 +ttpp: phase:2/3 pd:1488 gd:1000 t:278.5s +tttg: c1/61 lr:0.001000 t:0.1s +tttg: c2/61 lr:0.000999 t:0.1s +tttg: c3/61 lr:0.000997 t:0.2s +tttg: c4/61 lr:0.000994 t:0.3s +tttg: c5/61 lr:0.000989 t:0.4s +tttg: c6/61 lr:0.000983 t:0.5s +tttg: c7/61 lr:0.000976 t:0.5s +tttg: c8/61 lr:0.000967 t:0.6s +tttg: c9/61 lr:0.000957 t:0.7s +tttg: c10/61 lr:0.000946 t:0.8s +tttg: c11/61 lr:0.000933 t:0.8s +tttg: c12/61 lr:0.000919 t:0.9s +tttg: c13/61 lr:0.000905 t:1.0s +tttg: c14/61 lr:0.000889 t:1.1s +tttg: c15/61 lr:0.000872 t:1.1s +tttg: c16/61 lr:0.000854 t:1.2s +tttg: c17/61 lr:0.000835 t:1.3s +tttg: c18/61 lr:0.000815 t:1.4s +tttg: c19/61 lr:0.000794 t:1.5s +tttg: c20/61 lr:0.000772 t:1.5s +tttg: c21/61 lr:0.000750 t:1.6s +tttg: c22/61 lr:0.000727 t:1.7s +tttg: c23/61 lr:0.000703 t:1.8s +tttg: c24/61 lr:0.000679 t:1.9s +tttg: c25/61 lr:0.000655 t:1.9s +tttg: c26/61 lr:0.000629 t:2.0s +tttg: c27/61 lr:0.000604 t:2.1s +tttg: c28/61 lr:0.000578 t:2.2s +tttg: c29/61 lr:0.000552 t:2.3s +tttg: c30/61 lr:0.000526 t:2.3s +tttg: c31/61 lr:0.000500 t:2.4s +tttg: c32/61 lr:0.000474 t:2.5s +tttg: c33/61 lr:0.000448 t:2.6s +tttg: c34/61 lr:0.000422 t:2.7s +tttg: c35/61 lr:0.000396 t:2.7s +tttg: c36/61 lr:0.000371 t:2.8s +tttg: c37/61 lr:0.000345 t:2.9s +tttg: c38/61 lr:0.000321 t:3.0s +tttg: c39/61 lr:0.000297 t:3.0s +tttg: c40/61 lr:0.000273 t:3.1s +tttg: c41/61 lr:0.000250 t:3.2s +tttg: c42/61 lr:0.000228 t:3.3s +tttg: c43/61 lr:0.000206 t:3.4s +tttg: c44/61 lr:0.000185 t:3.4s +tttg: c45/61 lr:0.000165 t:3.5s +tttg: c46/61 lr:0.000146 t:3.6s +tttg: c47/61 lr:0.000128 t:3.7s +tttg: c48/61 lr:0.000111 t:3.8s +tttg: c49/61 lr:0.000095 t:3.8s +tttg: c50/61 lr:0.000081 t:3.9s +tttg: c51/61 lr:0.000067 t:4.0s +tttg: c52/61 lr:0.000054 t:4.1s +tttg: c53/61 lr:0.000043 t:4.2s +tttg: c54/61 lr:0.000033 t:4.2s +tttg: c55/61 lr:0.000024 t:4.3s +tttg: c56/61 lr:0.000017 t:4.4s +tttg: c57/61 lr:0.000011 t:4.5s +tttg: c58/61 lr:0.000006 t:4.5s +tttg: c59/61 lr:0.000003 t:4.6s +tttg: c60/61 lr:0.000001 t:4.7s +ttpr: phase:2/3 t:285.3s +ttp: b133/157 bl:2.2975 bb:1.0283 rl:2.2605 rb:1.0758 dl:1484-1516 gd:0 +ttpp: phase:3/3 pd:2000 gd:1500 t:289.7s +tttg: c1/82 lr:0.001000 t:0.1s +tttg: c2/82 lr:0.001000 t:0.2s +tttg: c3/82 lr:0.000998 t:0.2s +tttg: c4/82 lr:0.000997 t:0.3s +tttg: c5/82 lr:0.000994 t:0.4s +tttg: c6/82 lr:0.000991 t:0.5s +tttg: c7/82 lr:0.000987 t:0.5s +tttg: c8/82 lr:0.000982 t:0.6s +tttg: c9/82 lr:0.000976 t:0.7s +tttg: c10/82 lr:0.000970 t:0.8s +tttg: c11/82 lr:0.000963 t:0.9s +tttg: c12/82 lr:0.000955 t:0.9s +tttg: c13/82 lr:0.000947 t:1.0s +tttg: c14/82 lr:0.000938 t:1.1s +tttg: c15/82 lr:0.000928 t:1.2s +tttg: c16/82 lr:0.000918 t:1.2s +tttg: c17/82 lr:0.000907 t:1.3s +tttg: c18/82 lr:0.000895 t:1.4s +tttg: c19/82 lr:0.000883 t:1.5s +tttg: c20/82 lr:0.000870 t:1.6s +tttg: c21/82 lr:0.000857 t:1.6s +tttg: c22/82 lr:0.000843 t:1.7s +tttg: c23/82 lr:0.000829 t:1.8s +tttg: c24/82 lr:0.000814 t:1.9s +tttg: c25/82 lr:0.000799 t:1.9s +tttg: c26/82 lr:0.000783 t:2.0s +tttg: c27/82 lr:0.000767 t:2.1s +tttg: c28/82 lr:0.000750 t:2.2s +tttg: c29/82 lr:0.000733 t:2.3s +tttg: c30/82 lr:0.000716 t:2.3s +tttg: c31/82 lr:0.000698 t:2.4s +tttg: c32/82 lr:0.000680 t:2.5s +tttg: c33/82 lr:0.000662 t:2.6s +tttg: c34/82 lr:0.000643 t:2.6s +tttg: c35/82 lr:0.000625 t:2.7s +tttg: c36/82 lr:0.000606 t:2.8s +tttg: c37/82 lr:0.000587 t:2.9s +tttg: c38/82 lr:0.000568 t:3.0s +tttg: c39/82 lr:0.000548 t:3.0s +tttg: c40/82 lr:0.000529 t:3.1s +tttg: c41/82 lr:0.000510 t:3.2s +tttg: c42/82 lr:0.000490 t:3.3s +tttg: c43/82 lr:0.000471 t:3.4s +tttg: c44/82 lr:0.000452 t:3.4s +tttg: c45/82 lr:0.000432 t:3.5s +tttg: c46/82 lr:0.000413 t:3.6s +tttg: c47/82 lr:0.000394 t:3.7s +tttg: c48/82 lr:0.000375 t:3.7s +tttg: c49/82 lr:0.000357 t:3.8s +tttg: c50/82 lr:0.000338 t:3.9s +tttg: c51/82 lr:0.000320 t:4.0s +tttg: c52/82 lr:0.000302 t:4.1s +tttg: c53/82 lr:0.000284 t:4.1s +tttg: c54/82 lr:0.000267 t:4.2s +tttg: c55/82 lr:0.000250 t:4.3s +tttg: c56/82 lr:0.000233 t:4.4s +tttg: c57/82 lr:0.000217 t:4.4s +tttg: c58/82 lr:0.000201 t:4.5s +tttg: c59/82 lr:0.000186 t:4.6s +tttg: c60/82 lr:0.000171 t:4.7s +tttg: c61/82 lr:0.000157 t:4.8s +tttg: c62/82 lr:0.000143 t:4.8s +tttg: c63/82 lr:0.000130 t:4.9s +tttg: c64/82 lr:0.000117 t:5.0s +tttg: c65/82 lr:0.000105 t:5.1s +tttg: c66/82 lr:0.000093 t:5.2s +tttg: c67/82 lr:0.000082 t:5.2s +tttg: c68/82 lr:0.000072 t:5.3s +tttg: c69/82 lr:0.000062 t:5.4s +tttg: c70/82 lr:0.000053 t:5.5s +tttg: c71/82 lr:0.000045 t:5.6s +tttg: c72/82 lr:0.000037 t:5.6s +tttg: c73/82 lr:0.000030 t:5.7s +tttg: c74/82 lr:0.000024 t:5.8s +tttg: c75/82 lr:0.000018 t:5.9s +tttg: c76/82 lr:0.000013 t:6.0s +tttg: c77/82 lr:0.000009 t:6.0s +tttg: c78/82 lr:0.000006 t:6.1s +tttg: c79/82 lr:0.000003 t:6.2s +tttg: c80/82 lr:0.000002 t:6.3s +tttg: c81/82 lr:0.000000 t:6.4s +ttpr: phase:3/3 t:298.1s +ttp: b120/157 bl:2.2811 bb:1.0645 rl:2.2619 rb:1.0750 dl:1115-1143 gd:1 +ttp: b117/157 bl:2.3477 bb:1.0604 rl:2.2672 rb:1.0741 dl:1043-1068 gd:1 +ttp: b107/157 bl:2.3794 bb:1.0637 rl:2.2725 rb:1.0736 dl:860-876 gd:1 +ttp: b97/157 bl:2.2406 bb:1.0373 rl:2.2713 rb:1.0721 dl:730-742 gd:1 +ttp: b90/157 bl:2.3832 bb:1.0760 rl:2.2750 rb:1.0722 dl:648-659 gd:1 +ttp: b87/157 bl:2.2448 bb:1.0533 rl:2.2741 rb:1.0717 dl:618-629 gd:1 +ttp: b74/157 bl:2.2813 bb:1.0633 rl:2.2743 rb:1.0715 dl:495-503 gd:1 +ttp: b68/157 bl:2.3811 bb:1.0728 rl:2.2766 rb:1.0715 dl:450-458 gd:1 +ttp: b61/157 bl:2.3364 bb:1.1246 rl:2.2777 rb:1.0725 dl:399-406 gd:1 +ttp: b53/157 bl:2.3444 bb:1.1199 rl:2.2788 rb:1.0732 dl:350-356 gd:1 +ttp: b46/157 bl:2.3848 bb:1.1191 rl:2.2803 rb:1.0739 dl:310-314 gd:1 +ttp: b39/157 bl:2.3541 bb:1.1288 rl:2.2812 rb:1.0745 dl:275-279 gd:1 +ttp: b34/157 bl:2.4553 bb:1.1526 rl:2.2831 rb:1.0754 dl:253-256 gd:1 +ttp: b27/157 bl:2.3655 bb:1.1156 rl:2.2839 rb:1.0758 dl:224-227 gd:1 +ttp: b19/157 bl:2.4477 bb:1.2146 rl:2.2853 rb:1.0769 dl:187-192 gd:1 +ttp: b15/157 bl:2.5027 bb:1.2128 rl:2.2869 rb:1.0778 dl:171-175 gd:1 +ttp: b8/157 bl:2.5842 bb:1.2534 rl:2.2886 rb:1.0789 dl:138-143 gd:1 +ttp: b1/157 bl:2.7534 bb:1.1930 rl:2.2904 rb:1.0793 dl:69-97 gd:1 +ppm_mix bytes=13517716 mix_bpb=0.860499 ppm_only=2.139538 nn_only=0.967113 bin0_pct=0.136 bin1_pct=0.125 bin2_pct=0.085 bin3_pct=0.654 +ppm_mix_time:58.4s subset=4000000 tokens +quantized_ttt_phased val_loss:2.29220561 val_bpb:1.06686141 eval_time:374612ms +total_eval_time:374.6s