fix(pretrain): default to drive_real; add GATE-TRAIN-010 falsifier#949
Merged
Conversation
) Contract training-loop-pretrain-v1 v1.3.0 → v1.4.0. Root cause (tasks #119 / #124 / #125): `extended_commands.rs` shipped `#[arg(long, default_value = "true")] synthetic: bool` with no `--synthetic=false` companion. Every `apr pretrain` invocation silently routed to drive_synthetic (scripted loss), so every "real-compute" capture was actually synthetic — val_loss values (18 at target=10, 4.2 at target=3) match the synthetic formula target × 1.8 linear-decayed. Fix: * extended_commands.rs: flip to `action = ArgAction::SetTrue` so absent = false (real), present = true (synthetic). * contracts/training-loop-pretrain-v1.yaml: - bump 1.3.0 → 1.4.0 - add INV-TRAIN-010 (CLI default routing must reach drive_real) - add GATE-TRAIN-010 (ship_blocking, binds AC-SHIP2-003) * crates/apr-cli/src/commands/pretrain.rs: - cli_pretrain_defaults_to_real_compute - cli_pretrain_synthetic_flag_routes_to_synthetic Parse on a 16 MiB worker thread because the `Commands` enum overflows the default 2 MiB test-thread stack in debug. Verification: * cargo test -p apr-cli --features training --lib pretrain → 8/8 pass (debug + release) * pv validate contracts/training-loop-pretrain-v1.yaml → 0 errors, 0 warnings Unblocks task #126 (MODEL-2 real-compute dispatch). Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
9 tasks
noahgift
added a commit
that referenced
this pull request
Apr 21, 2026
#961) Task #128. First real MODEL-2 from-scratch compute dispatch (post PR #949 canonical rebuild at commit 29607ed, 2026-04-21) reached drive_real and immediately fired `Embedding::forward token_id 50241 >= vocab_size 50000. N-09 OOB escape.` — provably garbage gradients on 257 out-of-range tokens. Root cause decomposes 3 ways: D1: tokenizer artifact at vocab=50_257 (GPT-2 standard: 50_000 merges + 256 bytes + 1 <|endoftext|>) violates BOTH tokenizer-bpe-v1 and llama-370m-sovereign-v1 contracts (both pin 50_000 exact) D2: Llama370MConfig::VOCAB_SIZE is a hardcoded const — `--vocab-size` CLI flag is silently ignored by the model-side config path D3: no pre-flight gate caught the mismatch → N-09 OOB escape guard masked garbage tokens as escape id, producing silent data corruption This PR ships the orthogonal D3 fix only. D1/D2 require an A/B decision (bump both contracts to 50_257, OR retrain tokenizer to 50_000) that is deferred to a follow-up. The pre-flight gate falsifies any future mismatch regardless of which number wins. Contract-first sequence: 1. contracts/model-families/llama-370m-sovereign-v1.yaml v1.3.0 → v1.4.0: new GATE-ARCH-370M-011 + FALSIFY-ARCH-370M-011, ship_blocking, binds_to AC-SHIP2-003. `pv validate` → 0 errors. 2. aprender-train: pure `assert_tokenizer_vocab_matches_model` helper (algorithm-level proof, no I/O) + `falsify_gate_arch_370m_011_*` test covering match, mismatch (50_257 vs 50_000), and edge cases. 3. apr-cli: `preflight_tokenizer_vocab_matches_model` wrapper reads <tokenizer_dir>/vocab.json, counts entries, delegates to pure helper. Called at top of `drive_real` before ShardBatchIter::new — refuses to dispatch a real training step on mismatch. Synthetic drive is intentionally exempt (never touches the real model). 4. Tests: 3 new CLI tests (accept match, reject 50257 vs 50000, reject missing vocab.json) + fix to existing real_mode_empty_dataset_dir test that now stages a valid vocab.json so the shard-iterator failure path is still reachable. Verification: cargo test -p aprender-train --lib llama_370m → 10 passed cargo test -p apr-cli --lib commands::pretrain → 11 passed pv validate contracts/model-families/llama-370m-sovereign-v1.yaml → valid Unblocks task #126 (re-dispatch) pending D1/D2 A/B decision. Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
apr pretrainshipped with#[arg(long, default_value = "true")] synthetic: booland no--synthetic=falsecompanion. Every invocation silently routed todrive_synthetic(scripted-loss), so tasks APR-FORMAT-002: APR v2 format spec for web-scale models #119 / Integrate trueno-viz for training visualization #124 / Integrate trueno-rag for enhanced text/document ML #125 captured synthetic output and mis-labeled it real compute. val_loss=18 at target=10 and val_loss=4.2 at target=3 match the synthetic formulatarget × 1.8linear-decayed — proving no real compute ran.action = ArgAction::SetTrue→ absent = false = drive_real, present = true = drive_synthetic.training-loop-pretrain-v1v1.3.0 → v1.4.0 with INV-TRAIN-010 (CLI default routing) and GATE-TRAIN-010 (ship_blocking, binds AC-SHIP2-003). Two parse-level falsifier tests assert the routing discriminator byte-for-byte.Why parse-level tests?
Existing
mode_*tests callrun(...)directly with explicitsynthetic: bool— they never exercise the clap surface, which is exactly why the defect slipped through task #105. The new tests driveCli::try_parse_fromso any future regression in the clap attribute fails a unit test before it reaches CI.Test plan
cargo test -p apr-cli --features training --lib pretrain→ 8/8 pass (debug + release)pv validate contracts/training-loop-pretrain-v1.yaml→ 0 errors, 0 warningscargo build -p apr-cli --features training→ cleanci / gate+workspace-testgreenUnblocks
Task #126 — MODEL-2 from-scratch pretrain on lambda-labs RTX 4090.
🤖 Generated with Claude Code