chore: bump mlx-swift-lm to pick up MTP head quantization fix - #154
Merged
Conversation
Fixes #153. mlx-swift-lm#56 fixes a per-layer quantization lookup bug where MTP prediction-head modules (indexed as mtp.<depth>.layers... in the Swift module tree, but keyed as mtp.layers... in checkpoint quantization overrides) fell through to the top-level default quantization instead of their own override — crashing quantized_matmul on checkpoints with a separately-quantized MTP head, e.g. a uniform 8-bit MTP head over a mixed 4/5/6-bit qwen3_5_moe main model, under --stream-experts --mtp.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
mlx-swift-lmsubmodule to pick up mlx-swift-lm#56.qwen3_5_moecheckpoint with a mixed-precision MTP head (a standardmlx_lm.quantizeoutput pattern) crashedquantized_matmulon the first inference request under--stream-experts --mtp, because MTP-head modules never matched their per-layer quantization override (mtp.<depth>.layers...in the Swift module tree vs.mtp.layers...in the checkpoint's override keys) and silently fell back to the wrong default bit-width.Test plan