Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions docs/BENCHMARKS.md
Original file line number Diff line number Diff line change
Expand Up @@ -435,6 +435,11 @@ built on it rather than keeping the flattering one.

**CPU elementwise GEMM, transpose-free `[K,N]` path (2026-08-07).** On dgx aarch64 the `[K,N]` path beats `[N,K]` by 1.16x to 1.30x, byte-identically. The x86 arm is INDICATIVE ONLY, not binding: that box is VOID for timing per `CLAIM-KERNEL-CPU-ELEM-GEMM-1`. `VT_CPU_MATMUL_STEAL` ships default OFF and is NOT measured; it must justify itself by measurement and may measure neutral.

**Darwin Qwen3.5 build repair (2026-08-16).** Benchmarking is NOT APPLICABLE.
The change removes a redundant namespace-scope lambda capture that Apple Clang
rejects under `-Werror`; it does not change generated refusal text, model math,
or any runtime path. The binding gate is the Apple Clang build.

## Open gaps

| Track | Status | Next gate |
Expand Down
4 changes: 4 additions & 0 deletions docs/FEATURES.md
Original file line number Diff line number Diff line change
Expand Up @@ -148,6 +148,10 @@ speed-pending, which [BENCHMARKS.md](BENCHMARKS.md) tracks.
| `CohereForCausalLM` | Command-R / Cohere (and Cohere2) | scaffold: W0 tiny-random oracle run-verified; real-checkpoint gate blocked | no run |
<!-- supported-arch-table:end -->

The Qwen3.5 MoE loader also builds under Apple Clang with project warnings
promoted to errors. Its layout-refusal path uses the same messages and behavior
on every platform.

### Standalone and non-registered lanes

These run through dedicated forwards, not the `REGISTER_VLLM_MODEL` registry, so
Expand Down
6 changes: 6 additions & 0 deletions docs/STATUS.md
Original file line number Diff line number Diff line change
Expand Up @@ -2529,6 +2529,12 @@ The next run also guarded Voxtral's GCC-only `-Wstringop-overflow` suppression
out of Clang, where it was fatal. Its Go `go-m1cpu` diagnostics were nonfatal
and outside this repo.

**Darwin Qwen3.5 build repair (2026-08-16).** The MoE layout refusal lambda no
longer captures its namespace-scope help string. Apple Clang diagnosed that
redundant capture as `-Wunused-lambda-capture`, and the project promotes the
warning to an error. The lambda can still read the namespace-scope name
directly. Runtime behavior and the Qwen3.5 lifecycle state are unchanged.

**Agent onboarding:** [session](../.agents/specs/session-onboarding.md) +
[entry](../.agents/specs/developer-agent-protocol-entrypoint.md) implemented;
documentation-only.
2 changes: 1 addition & 1 deletion src/vllm/model_executor/models/qwen3_5_weights.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -950,7 +950,7 @@ void CheckMoeQuantLayoutSupported(const std::vector<std::string>& names,
// ...and the three NON-routed components, refused by the dtype the probe
// RESOLVED rather than discovered as a complaint from inside a reader (#490).
// Each of these already failed before #864; naming it is the whole change.
const auto refuse = [&kRequired](const char* what, MoeProjDtype got,
const auto refuse = [](const char* what, MoeProjDtype got,
const char* supported) {
VT_CHECK(false, std::string("qwen3_5 weights: a ") +
MoeProjDtypeName(got) + " " + what +
Expand Down
Loading