chore: bump mlx-rs pin to the qwen3 quantize-before-load fix #1

Merged

peterswimm merged 1 commit from chore/bump-mlx-rs-pin-qwen3-fix into main

2026-07-16 20:01:38 +00:00

peterswimm commented

2026-07-16 18:46:20 +00:00

Owner

Blocked on Toilville/mlx-rs-mirror#2 — points at that PR's branch tip (fix/qwen3-quantize-before-load) rather than main, since main still crashes on the first RMSNorm call for quantized Qwen3 checkpoints ((*weight) must have the same size as the last dimension of x).

Verified locally against a real mlx-community/Qwen3-0.6B-4bit snapshot via the downstream ritual_engine integration tests — both mlx_concat_smoke and mlx_smoke pass with this pin, and both fail identically without it.

Do not merge until mlx-rs-mirror#2 merges — re-pin this branch to main once it does, rather than merging as-is against a side branch.

Blocked on Toilville/mlx-rs-mirror#2 — points at that PR's branch tip (`fix/qwen3-quantize-before-load`) rather than `main`, since `main` still crashes on the first `RMSNorm` call for quantized Qwen3 checkpoints (`(*weight) must have the same size as the last dimension of x`). Verified locally against a real `mlx-community/Qwen3-0.6B-4bit` snapshot via the downstream `ritual_engine` integration tests — both `mlx_concat_smoke` and `mlx_smoke` pass with this pin, and both fail identically without it. Do not merge until mlx-rs-mirror#2 merges — re-pin this branch to `main` once it does, rather than merging as-is against a side branch.

peterswimm added 1 commit

2026-07-16 18:46:20 +00:00

chore: bump mlx-rs pin to the qwen3 quantize-before-load fix

CI / Core (default features, no MLX) (pull_request) Failing after 3s

Details

CI / MLX feature (macOS + Apple Silicon) (pull_request) Failing after 2s

Details

CI / lint-workflows (pull_request) Successful in 10s

Details

1c7d5d1728

Points at Toilville/mlx-rs-mirror#fix/qwen3-quantize-before-load
(pending its own PR review) instead of main, since main is still
missing the fix for the rms_norm shape-mismatch crash on quantized
Qwen3 checkpoints. Re-pin to main once that PR merges.

peterswimm merged commit b8283cc0f2 into main

2026-07-16 20:01:38 +00:00

peterswimm deleted branch chore/bump-mlx-rs-pin-qwen3-fix

2026-07-16 20:01:38 +00:00

peterswimm referenced this pull request from a commit

2026-07-16 20:01:39 +00:00

Merge pull request 'chore: bump mlx-rs pin to the qwen3 quantize-before-load fix' (#1) from chore/bump-mlx-rs-pin-qwen3-fix into main

peterswimm referenced this pull request

2026-07-16 20:07:03 +00:00

chore: point mlx-rs pin at mlx-rs-mirror main tip #2