contract(trace-ffn-sub-block-gguf-v1): v1.0.0 PROPOSED scaffold — SHIP-007 layer-3 H1/H2 unblock by noahgift · Pull Request #1532 · paiml/aprender

noahgift · 2026-05-06T10:57:20Z

Summary (CORRECTED)

Authors the GGUF-side sub-FFN telemetry contract that documents and pins the SHIP-007 layer-3 ffn_swigl bisection cascade.

Correction during authoring: I initially claimed M-FFN-GGUF-1 and M-FFN-GGUF-2 were PENDING. They're actually SHIPPED via prior PRs #1081 (PR A scaffold) and #1082 (PR B sub-FFN populate). The contract has been amended to reflect this. M-FFN-GGUF-3 (heavy comparison harness) and M-FFN-GGUF-4 (SHIP-007 fix PR) remain OPEN.

The blocker this contract documents

SHIP-007 §21 (aprender PR #1072) narrowed the bug to "(layer=3, ffn_swigl element-wise multiply)" on the APR forward path:

APR layer-3 ffn_swigl std = 1.222 (17.2× layer-2 baseline 0.071)

§21 cannot distinguish two competing hypotheses:

Hypothesis	What it means
H1	Token-position-dependent correlation — NORMAL model behavior; SHIP-007 root cause is ELSEWHERE
H2	APR-side bug — fix at `inference.rs:160-164` swigl elementwise multiply

Implementation stages (CORRECTED)

Stage	Status	What
M-FFN-GGUF-0	SHIPPED (this PR)	Contract scaffold
M-FFN-GGUF-1	SHIPPED	LayerActivation shared APR↔GGUF (re-export verified)
M-FFN-GGUF-2	SHIPPED	`OwnedQuantizedModel::forward_traced` for dense SwiGLU path (PRs #1081 + #1082)
M-FFN-GGUF-3	PENDING	Heavy comparison harness APR vs GGUF layer-3 ffn_swigl
M-FFN-GGUF-4	PENDING	SHIP-007 fix PR cites H1 or H2

Known limitation in M-FFN-GGUF-2: the MoE path's forward_qwen3_moe_traced does NOT populate the 4 per-layer LayerActivation sub-FFN fields. Not blocking H1/H2 for the dense 7B Qwen2.5-Coder; future enhancement if MoE-side SHIP-007 bisection ever needed.

Why ship a contract that retro-records prior work

The contract still serves a load-bearing purpose:

Pins the architecture explicitly so M-FFN-GGUF-3 + M-FFN-GGUF-4 have a clear discharge path
Cross-references the prior-work PRs (feat(p3-pra): SHIP-007 GGUF forward_traced scaffold — 6 non-FFN LayerActivation fields per layer #1081, feat(p3-prb): SHIP-007 GGUF forward_traced sub-FFN populate — 4 sub-FFN ActivationStats slots filled #1082) for anyone reading the contract surface
Adds 4 falsification tests + 2 proof obligations + 1 Kani harness obligation that didn't have a contract surface before

This is the parallel-track autonomous work that should have been authored during M-GPU-MOE CI dead time. With M-GPU-MOE-1.x cascade closed, the SHIP-007 layer-3 cascade is now contract-pinned and only M-FFN-GGUF-3/4 remain.

Test plan

pv validate 0/0
No production code touched (YAML-only)
References cite memory project_ship_007_layer_3_swiglu_bisection.md, evidence dir, sibling APR contract trace-ffn-sub-block-v1, prior-work PRs feat(p3-pra): SHIP-007 GGUF forward_traced scaffold — 6 non-FFN LayerActivation fields per layer #1081/feat(p3-prb): SHIP-007 GGUF forward_traced sub-FFN populate — 4 sub-FFN ActivationStats slots filled #1082, and proven precedent trace-moe-gpu-sub-stages-v1

🤖 Generated with Claude Code

…P-007 layer-3 H1/H2 unblock Authors the GGUF-side sub-FFN telemetry contract that unblocks the SHIP-007 layer-3 ffn_swigl bisection (memory: project_ship_007_layer_3_swiglu_bisection.md). BACKGROUND: SHIP-007 §21 (aprender PR #1072) narrowed the bug to "(layer=3, ffn_swigl element-wise multiply)" on the APR forward path. APR layer-3 ffn_swigl std = 1.222 (17.2× layer-2 baseline). But §21 cannot distinguish: H1: Token-position-dependent correlation (NORMAL model behavior) H2: APR-side bug (forward path produces wrong VALUES vs GGUF) without GGUF-side per-layer sub-FFN telemetry. Currently NO forward_traced method exists on OwnedQuantizedModel. THIS CONTRACT: Pins the architecture for adding GGUF-side traced forward mirroring `trace-ffn-sub-block-v1` (APR sibling). Same 5 sub-FFN fields (gate_proj_out, up_proj_out, silu_gate, swiglu_inner, ffn_down_out). Same equation. Cross-comparison enabled by schema parity (PO-TRACE-FFN-GGUF-002). Implementation stages (multi-PR cascade, deliberate session): - M-FFN-GGUF-0: contract scaffold (this PR) - M-FFN-GGUF-1: LayerActivation struct on GGUF side - M-FFN-GGUF-2: NEW forward_traced on OwnedQuantizedModel - M-FFN-GGUF-3: heavy comparison harness APR vs GGUF layer-3 - M-FFN-GGUF-4: SHIP-007 fix PR cites H1 or H2 4 falsification tests defined: - FFN-GGUF-001: forward_traced exists on OwnedQuantizedModel - FFN-GGUF-002: byte-identity vs production forward - FFN-GGUF-003: APR-vs-GGUF layer-3 ffn_swigl std distinguishes H1/H2 - FFN-GGUF-004: SHIP-007 fix PR cites bisected hypothesis PATTERN PRECEDENT: Mirrors the proven trace-moe-gpu-sub-stages-v1 cascade closure (M50→M87, 2026-05-04→06): contract scaffold first, then implementation stages, then heavy harness, then fix. Bug class is similar: bisection identified a SURFACE but cannot distinguish hypotheses without sibling telemetry. This contract is the parallel-track autonomous work that should have been authored during M-GPU-MOE CI dead time but wasn't. With M-GPU-MOE-1.x cascade now closed, the SHIP-007 layer-3 cascade can proceed in parallel-or-serial fashion. Discharges 5 transitively-blocked MODEL-1 PARTIALs per ship-two- models-spec.md §17.5: SHIP-002, SHIP-005, SHIP-006, SHIP-007, SHIP-008. YAML-only — production hot paths byte-unchanged. `pv validate` 0/0. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…— SHIP-007 H1/H2 bisection Authors the heavy comparison harness pinned in contract trace-ffn-sub-block-gguf-v1 v1.0.0 step M-FFN-GGUF-3 (M88 PR #1532 squash ca03361). Mirrors the M80 skip-if-not-present pattern from `qwen3_moe_gpu_per_stage_diff::falsify_moe_sub_002_*`. What it does: - Loads APR 7B teacher via AprTransformer::from_apr_file - Loads same model GGUF via OwnedQuantizedModel::from_mapped - Runs forward_traced on both with the same canonical SHIP-007 prompt ("What is 2+2?") - Extracts per-layer ffn_swiglu_inner_stats.std_dev for each - Computes ratio at layer 3 (the §21 anomaly site) - Reports verdict: - H1 (ratio in [0.5, 2.0]): NORMAL model behavior — SHIP-007 root cause is ELSEWHERE (lm_head / post-FFN residual / token- position correlation) - H2 (ratio outside band): APR-side bug — fix at `inference.rs:160-164` swigl elementwise multiply Discharges FALSIFY-FFN-GGUF-003 at the algorithm level: the harness EXISTS and produces a single H1/H2 verdict on operator dispatch. Full DISCHARGED requires the operator to run with canonical files present and capture the verdict in evidence. `#[ignore]`-gated; skips cleanly if either canonical 7B .apr or .gguf is missing on the host (verified locally — host has the .gguf at /mnt/nvme-raid0/models/ship-two-001/ but no .apr; test ran clean and skipped). Contract amendment: M-FFN-GGUF-3 status PENDING → ALGORITHM_LEVEL_DISCHARGED. Operator workflow documented inline. No production hot path touched — additive test-only file. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…— SHIP-007 H1/H2 bisection (#1533) Authors the heavy comparison harness pinned in contract trace-ffn-sub-block-gguf-v1 v1.0.0 step M-FFN-GGUF-3 (M88 PR #1532 squash ca03361). Mirrors the M80 skip-if-not-present pattern from `qwen3_moe_gpu_per_stage_diff::falsify_moe_sub_002_*`. What it does: - Loads APR 7B teacher via AprTransformer::from_apr_file - Loads same model GGUF via OwnedQuantizedModel::from_mapped - Runs forward_traced on both with the same canonical SHIP-007 prompt ("What is 2+2?") - Extracts per-layer ffn_swiglu_inner_stats.std_dev for each - Computes ratio at layer 3 (the §21 anomaly site) - Reports verdict: - H1 (ratio in [0.5, 2.0]): NORMAL model behavior — SHIP-007 root cause is ELSEWHERE (lm_head / post-FFN residual / token- position correlation) - H2 (ratio outside band): APR-side bug — fix at `inference.rs:160-164` swigl elementwise multiply Discharges FALSIFY-FFN-GGUF-003 at the algorithm level: the harness EXISTS and produces a single H1/H2 verdict on operator dispatch. Full DISCHARGED requires the operator to run with canonical files present and capture the verdict in evidence. `#[ignore]`-gated; skips cleanly if either canonical 7B .apr or .gguf is missing on the host (verified locally — host has the .gguf at /mnt/nvme-raid0/models/ship-two-001/ but no .apr; test ran clean and skipped). Contract amendment: M-FFN-GGUF-3 status PENDING → ALGORITHM_LEVEL_DISCHARGED. Operator workflow documented inline. No production hot path touched — additive test-only file. Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>

noahgift enabled auto-merge (squash) May 6, 2026 10:57

noahgift force-pushed the contract/trace-ffn-sub-block-gguf-v1.0.0-scaffold branch from 18a76aa to b3928d8 Compare May 6, 2026 11:15

noahgift merged commit ca03361 into main May 6, 2026
10 checks passed

noahgift deleted the contract/trace-ffn-sub-block-gguf-v1.0.0-scaffold branch May 6, 2026 11:32

noahgift mentioned this pull request May 6, 2026

feat(M-FFN-GGUF-3): heavy APR-vs-GGUF layer-3 ffn_swigl diff harness — SHIP-007 H1/H2 bisection #1533

Merged

5 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

contract(trace-ffn-sub-block-gguf-v1): v1.0.0 PROPOSED scaffold — SHIP-007 layer-3 H1/H2 unblock#1532

contract(trace-ffn-sub-block-gguf-v1): v1.0.0 PROPOSED scaffold — SHIP-007 layer-3 H1/H2 unblock#1532
noahgift merged 1 commit into
mainfrom
contract/trace-ffn-sub-block-gguf-v1.0.0-scaffold

noahgift commented May 6, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

noahgift commented May 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary (CORRECTED)

The blocker this contract documents

Implementation stages (CORRECTED)

Why ship a contract that retro-records prior work

Test plan

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

noahgift commented May 6, 2026 •

edited

Loading