All issues

Search results

Missing support for meituan-longcat/LongCat-Flash-Lite-Sparse

Status: Open.

#1794 In ml-explore/mlx-lm;

· MattBash17 opened on Aug 27, 2026

feat: MXFP8 / fp8 mixed-precision training path (LoRA + full finetune)

Status: Open.

#1793 In ml-explore/mlx-lm;

· dahai80 opened on Aug 27, 2026

gated_delta: vectorized Metal kernel reads dt_bias per-head; callers (kimi_linear) pass per-element [H, head_dim] — GPU/CPU gate mismatch

Status: Open.

#1792 In ml-explore/mlx-lm;

· aidiffuser opened on Aug 27, 2026

[BUG] Qwen3 MoE LoRA backward fails through discrete expert-routing indices

Status: Open.

#1786 In ml-explore/mlx-lm;

· nicejazzburger opened on Aug 26, 2026

mlx_lm "pythonic" tool-call parser truncates arguments containing single-quoted strings with embedded commas

Status: Open.

#1785 In ml-explore/mlx-lm;

· Thomas-Boutin opened on Aug 25, 2026

Serving a sliding-window model alongside another model corrupts the other model's BatchKVCache: silent degenerate outputs on cache hits, then _idx=-1 crash (deterministic repro, 3/3; negative control 200 rounds clean)

Status: Open.

#1776 In ml-explore/mlx-lm;

· hofmanfr opened on Aug 23, 2026

mlx_lm.generate.generate_step() omits prompt-prefill tokens from logits-processor history.

Status: Open.

#1773 In ml-explore/mlx-lm;

· Sayitobar opened on Aug 22, 2026

Pipeline parallelism support for qwen3_moe

enhancement

Status: Open.

#1763 In ml-explore/mlx-lm;

· fretnoyz opened on Aug 19, 2026

mlx_lm.server never clears MLX's buffer pool on model swap, retaining every model it has loaded

bug

Status: Open.

#1757 In ml-explore/mlx-lm;

· j7t8t9gyw6-tech opened on Aug 18, 2026

Distributed JACCL server crashes when two requests arrive concurrently

Status: Open.

#1737 In ml-explore/mlx-lm;

· nachogala opened on Aug 15, 2026

Code in the first example for Python SDK do not pass type check on load()'s returned values

Status: Open.

#1732 In ml-explore/mlx-lm;

· rija opened on Aug 14, 2026

make_shards prepends an empty shard when the first tensor exceeds the size limit

Status: Open.

#1720 In ml-explore/mlx-lm;

· FU-max-boop opened on Aug 13, 2026