Search results

DFlash + concurrent requests: TypeError in rollback_speculative_cache (qwen3_5/language.py:523)

Status: Open.

#1086 In Blaizzy/mlx-vlm;

· nnorris7 opened on Apr 28, 2026

module 'mlx.core' has no attribute 'new_thread_local_stream'

Status: Open.

#1085 In Blaizzy/mlx-vlm;

· zhangcz-ai opened on Apr 28, 2026

OpenAI-compat /v1/chat/completions: only the LAST image_url block is used; multi-image content silently truncated to N=1

Status: Open.

#1084 In Blaizzy/mlx-vlm;

· jescoti opened on Apr 27, 2026

batch_generate produces token-level loops on Gemma 4 models at batch_size>=4 (works at B=1)

Status: Open.

#1083 In Blaizzy/mlx-vlm;

· emmilco opened on Apr 27, 2026

Feature request: /v1/cache/warm-prefix endpoint for first-turn prefix cache warming

Status: Open.

#1080 In Blaizzy/mlx-vlm;

· danielkhalilov opened on Apr 26, 2026

Unable to use new Intel/Qwen3.6-27B-4.5b-mlx-AutoRound

Status: Open.

#1079 In Blaizzy/mlx-vlm;

· ivanfioravanti opened on Apr 26, 2026

Qwen3.5 MoE models use dense MLP instead of SparseMoeBlock

Status: Open.

#1073 In Blaizzy/mlx-vlm;

· Thump604 opened on Apr 26, 2026

feature: add missing metadata in server endpoints

Status: Open.

#1072 In Blaizzy/mlx-vlm;

· goniz opened on Apr 25, 2026

Hardcoded 60s queue timeout in token_iterator silently truncates long-prompt prefills

Status: Open.

#1068 In Blaizzy/mlx-vlm;

· eloe opened on Apr 25, 2026

20260424: Diagnostics Report — 2 failure(s), 4 harness issue(s) (mlx-vlm 0.4.5)

Status: Open.

#1066 In Blaizzy/mlx-vlm;

· jrp2014 opened on Apr 24, 2026

Qwen/Qwen3-VL-2B-Instruct: GPU Hang Error (00000003:kIOGPUCommandBufferCallbackErrorHang)

Status: Open.

#1064 In Blaizzy/mlx-vlm;

· jrp2014 opened on Apr 24, 2026

Improve graceful degradation and error handling

Status: Open.

#1063 In Blaizzy/mlx-vlm;

· VanTam-CN opened on Apr 24, 2026