fix(gguf): Q5_0/Q5_1 dequant layout matches GGML reference (closes #1623)#1656
Merged
Conversation
…w-major (was [K,N]); MODEL-1 → 100% (PMAT-CODE-SHIP-007-F32-GEMV-LAYOUT-FIX)
§74 localized the SHIP-007 PARITY-GATE bug to f32_gemv_into via PR-B's
stage-bisection scaffold (CPU vs GPU per-stage statistics analysis).
The F32 GEMV PTX kernel was reading weights with TRANSPOSED layout
interpretation:
Bug: kernel assumed A is K-rows × N-cols row-major (A[i,j] at i*N+j),
but actual ML weights are stored [output_dim=N, input_dim=K]
row-major (A[i,j] at i*K+j per PyTorch/SafeTensors/GGUF convention
and PMAT-333 F32 dequantization output).
Symptom: GPU read transposed weights → computed y = A^T @ x instead
of y = A @ x → systematically anti-correlated logits
(cos=-0.005190 vs CPU, top-10 divergences all sign-flipped,
CPU mean=-2.42 vs GPU mean=0.013).
Fix: rewrite the inner loop to iterate along the K dimension within
row block_id:
row_base = a_ptr + block_id * K * 4
thread reads A[block_id, t], A[block_id, t+32], ...
instead of:
col_base = a_ptr + block_id * 4
thread reads A[t, block_id], A[t+32, block_id], ...
Empirical discharge (canonical 7B teacher, lambda-vector RTX 4090,
default graphed path):
PARITY-GATE: PASS (no error from forward_gpu_resident)
Throughput @ 128-tok 5-iter decode: 124.6 tok/s
AC-SHIP1-007 floor: 30 tok/s
Headroom: 4.15× over floor
TTFT: 8.39 ms
p50 latency: 1016 ms
Before PR-E:
PARITY-GATE FAILED cos=-0.005190
Throughput (with SKIP_PARITY_GATE=1 + SKIP_FP8_WARMUP=1): 5.6 tok/s (§63) / 54.5 tok/s (§73)
GPU CANNOT serve this model
After PR-E:
PARITY-GATE PASS, default path, NO workarounds
124.6 tok/s, 4.15× over floor
Ship-% impact:
MODEL-1 ship %: **99% → 100%**
10 of 10 AC-SHIP1-* LIVE-DISCHARGED:
SHIP-001 (§72) SHIP-002 (§61) SHIP-003 (§72)
SHIP-004 (§72) SHIP-005 (§71) SHIP-006 (§61.8)
SHIP-007 (this PR) SHIP-008 (§61) SHIP-009 (§72)
SHIP-010 (§72)
MODEL-2 ship %: unchanged at 57% (independent track).
Cascade arc closeout: §63 → §73 → PR-A (#1648) → PR-B (#1649)
→ §74 (#1650) → PR-E (this). One PR shipped in 1 day after §73's
'3-5 PR / 3-5 day' estimate.
Auxiliary change: logits.rs adds APR_LM_HEAD_FORCE_QTYPE env-var
probe kept as a diagnostic tool (zero behavior change when unset).
Test plan:
- [x] cargo build --release -p apr-cli --bin apr --features cuda → clean
- [x] apr bench (default path, 128-tok 5-iter) → 124.6 tok/s, passed: true
- [x] apr parity → PARITY-GATE PASS
- [ ] CI tests (workspace-test on per-PR runner)
Refs:
- §74 SHIP-007 bug localized (PR #1650)
- §73 SHIP-007 cascade reduction (PR #1647)
- contracts/apr-ship-007-gpu-stage-bisection-v1.yaml (PR-A #1648 contract)
- PR #1649 (PR-B GPU stage dump scaffold)
- AC-SHIP1-007 (spec §5)
- evidence/section-75-ship-007-discharged-2026-05-13/
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…07 contract violation (PMAT-CODE-SHIP-007-PR-E-FALSIFY-007-CLEAN)
The env-var bisection probe added in PR-E (this branch) introduced a
`_ =>` catch-all inside a `match` expression that referenced
`WeightQuantType` in its arm values. The `falsify_007_no_catch_all_
in_dispatch_sites` contract test's 30-line walk-back heuristic flagged
this as a violation, even though the match was on `&str` (env var
value), not on `WeightQuantType`.
The probe was a bisection tool used to identify the bug location
during §74. Now that §75 has shipped the actual fix and the probe is
no longer needed, removing it cleans up the contract violation.
The remaining PR-E change is solely the F32 GEMV PTX kernel layout
fix in `crates/aprender-gpu/src/kernels/gemv/mod.rs` — that's the
actual bug fix.
Test verified:
cargo test -p aprender-serve --lib \
quantize::contract_tests::tests::falsify_007_no_catch_all_in_dispatch_sites
→ 1 passed
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
) The Q5_0 and Q5_1 dequantizers in aprender-core were emitting values in interleaved order [v0, v1, v0, v1, ...] and using wrong high-bit indices (i*2 / i*2+1). GGML / llama.cpp layout is: for j in 0..16: y[j] = low0 (qs[j] & 0x0F) | (qh bit j << 4) y[j + 16] = low1 (qs[j] >> 4) | (qh bit j+16 << 4) Two halves, NOT interleaved. High bit for element j uses bit j; for element j+16 uses bit j+16. Existing tests only checked length and finite-ness — never the layout. Adds two GGML-reference layout tests (`test_dequantize_q5_0_ggml_layout`, `test_dequantize_q5_1_ggml_layout`) that fail under the buggy code and pass under the fix. Reported in #1623 from a Coursera capstone using mixed-quant GGUF.
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
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
Closes #1623. Fixes a real Q5_0/Q5_1 dequantization bug in `crates/aprender-core/src/format/gguf/dequant.rs`.
The bug
For each block (32 elements), the code emitted values in interleaved order `[v0, v1, v0, v1, ...]` and indexed the high-bits register with `i2` / `i2+1`. GGML / llama.cpp instead use two halves:
```
for j in 0..16:
y[j] = low0 (qs[j] & 0x0F) | (qh bit j << 4)
y[j + 16] = low1 (qs[j] >> 4) | (qh bit j+16 << 4)
```
Why tests didn't catch it
The pre-existing `test_dequantize_q5_0_basic` / `_q5_1_basic` tests used all-zero inputs and only asserted result length and `is_finite()`. The interleaved emit produced 32 zeros either way.
Fix
block_start + jandblock_start + j + 16), corrected high-bit indices (qh >> jandqh >> (j + 16))Both follow llama.cpp's `dequantize_row_q5_0` / `dequantize_row_q5_1` exactly.
Tests
Two new layout-verifying tests:
```
test_dequantize_q5_0_ggml_layout
test_dequantize_q5_1_ggml_layout
```
Both set qh bits at positions 0 and 16, set qs[0] = 0x0F (low0=15, low1=0), and assert that:
These tests fail under the buggy code and pass under the fix.
Reporter: a Coursera capstone investigation that found the bug while running mixed-quant GGUF through
apr run.Test plan
cargo test -p aprender-core --lib dequantize_q5— 14/14 pass locallyNote
The issue also called out `prepare_tokens_apr` in aprender-serve. That is a separate prompt-templating bug — out of scope for this PR.
🤖 Generated with Claude Code