Issues

Support Nemotron Puzzle 75B / nemotron_h_puzzle heterogeneous MoE models

#1506

· JuizSpeaking opened

on Jul 9, 2026

server: any uncaught exception in _generate leaves HTTP threads serving while every completion hangs forever

#1505

· pengying opened

on Jul 9, 2026

`mlx_lm.server` pins a CPU core at ~100% while idle (generation worker busy-polls the request queue with `get_nowait()`)

#1500

· itpro2792-beep opened

on Jul 8, 2026

N-gram self-speculative decoding for hybrid GDN/SA models

#1497

· dahai80 opened

on Jul 8, 2026

LRUPromptCache: one-token prefixes never match in PromptTrie.search; eviction ignores fetch recency (FIFO, not LRU)

#1495

· T0mSIlver opened

on Jul 7, 2026

Prompt-cache reuse can return KV state that doesn't match the keyed prefix (windowed-cache trim contract)

#1494

· T0mSIlver opened

on Jul 7, 2026

mlx_lm.server: generation hangs at 0% CPU right after prompt processing on ~22-26k-token streaming requests (gemma-4-26b-a4b, 0.31.3)

#1493

· TadeuszWolfGang opened

on Jul 7, 2026

Model type gemma4_unified not supported

#1481

· MattBash17 opened

on Jul 6, 2026

MLX/Metal OOM during long-context prefill for Qwen3.6/Qwen3.5 hybrid MoE despite only 10 KV-cache layers

#1480

· Water-Run opened

on Jul 6, 2026

Project status: what should contributors expect going forward?

#1475

· weklund opened

on Jul 5, 2026

mlx_lm.server: generation thread dies with TypeError ('NoneType' object is not iterable) when a batch mixes requests with and without logits processors — server then hangs forever

#1472

· Robaloman opened

on Jul 4, 2026

speculative_generate_step diverges from plain generation at temp=0 with num_draft_tokens=4 (dense model, no hybrid layers)

#1470

· h9q2cyxvgm-ui opened

on Jul 4, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Support Nemotron Puzzle 75B / nemotron_h_puzzle heterogeneous MoE models

server: any uncaught exception in _generate leaves HTTP threads serving while every completion hangs forever

`mlx_lm.server` pins a CPU core at ~100% while idle (generation worker busy-polls the request queue with `get_nowait()`)

N-gram self-speculative decoding for hybrid GDN/SA models

LRUPromptCache: one-token prefixes never match in PromptTrie.search; eviction ignores fetch recency (FIFO, not LRU)

Prompt-cache reuse can return KV state that doesn't match the keyed prefix (windowed-cache trim contract)

mlx_lm.server: generation hangs at 0% CPU right after prompt processing on ~22-26k-token streaming requests (gemma-4-26b-a4b, 0.31.3)

Model type gemma4_unified not supported

MLX/Metal OOM during long-context prefill for Qwen3.6/Qwen3.5 hybrid MoE despite only 10 KV-cache layers

Project status: what should contributors expect going forward?

mlx_lm.server: generation thread dies with TypeError ('NoneType' object is not iterable) when a batch mixes requests with and without logits processors — server then hangs forever

speculative_generate_step diverges from plain generation at temp=0 with num_draft_tokens=4 (dense model, no hybrid layers)

Uh oh!

Search results