Search results

mlx_lm fuse produces non-functional model with gpt-oss-20b MXFP4-Q8 (adapter-loaded serving works)

Status: Open.

#1172 In ml-explore/mlx-lm;

· proflouishyman opened on Apr 20, 2026

[Bug] Qwen3-Next concatenate shape mismatch under mixed concurrent requests (mlx-lm 0.31.2)

Status: Open.

#1166 In ml-explore/mlx-lm;

· zouchengfang opened on Apr 19, 2026

NameError: tree_reduce is not defined in mlx_lm/models/cache.py _BaseCache.nbytes

Status: Open.

#1164 In ml-explore/mlx-lm;

· siiea-ai opened on Apr 19, 2026

Bug report — Qwen3-Next hybrid cache silently fails - Prompt cache silently broken for Qwen3-Next (hybrid Gated-DeltaNet + SWA Gated-Attention) — recomputes every turn

Status: Open.

#1162 In ml-explore/mlx-lm;

· bmaxim77 opened on Apr 17, 2026

mlx_lm.dwq silently skips target computation when --target-dir is empty but exists

Status: Open.

#1159 In ml-explore/mlx-lm;

· laradji opened on Apr 16, 2026

Stateful logits processors see stale tokens due to lazy evaluation in stream context

Status: Open.

#1156 In ml-explore/mlx-lm;

· Thump604 opened on Apr 15, 2026

Dynamic quantization support

Status: Open.

#1154 In ml-explore/mlx-lm;

· dsrenesanse opened on Apr 15, 2026

Optimize Mamba/SSM decode and prefill kernels on Apple Silicon

Status: Open.

#1152 In ml-explore/mlx-lm;

· Gal-bloch opened on Apr 15, 2026

Quanting gemma-4

Status: Open.

#1149 In ml-explore/mlx-lm;

· nightmedia opened on Apr 14, 2026

mlx_lm server do not end tcp connection after prompt

Status: Open.

#1148 In ml-explore/mlx-lm;

· clarkttfu opened on Apr 14, 2026

Is vLLM's StreamingInput on the roadmap?

Status: Open.

#1147 In ml-explore/mlx-lm;

· stefanwebb opened on Apr 14, 2026

MiniMax-M2.7: Tool call XML output missing angle brackets on M3 Ultra cluster

Status: Open.

#1145 In ml-explore/mlx-lm;

· michalkomar opened on Apr 13, 2026