Search results

Qwen3.6-35B-A3B MTP builds (MXFP4-MTP, MXFP8-MTP) do not load on vMLX 1.5.61

Status: Open.

#213 In jjang-ai/vmlx;

· electricfx opened on Jun 19, 2026

Can't run Nemotron-Cascade-2-30B-A3B-JANG_2L -- fails to load.

Status: Open.

#212 In jjang-ai/vmlx;

· rickeames opened on Jun 18, 2026

flash-moe: synchronous expert-routing host sync (inds.tolist()) crashes under MLLM async chunked prefill — "There is no Stream(gpu,N) in current thread"

Status: Open.

#211 In jjang-ai/vmlx;

· st-adam opened on Jun 18, 2026

Claude code breakage

Status: Open.

#209 In jjang-ai/vmlx;

· CobraSoftware opened on Jun 17, 2026

AttributeError: 'create_ssm_mask' missing in qwen3_5/language.py during stream_chat

Status: Open.

#208 In jjang-ai/vmlx;

· Hinsael opened on Jun 16, 2026

Update Electron framework due to deprecated/old macOS Corner Mask API behavior

Status: Open.

#206 In jjang-ai/vmlx;

· trigger2k20 opened on Jun 15, 2026

qwen3_vl models: every request fails 500 "int += NoneType" — cache with offset/_idx = None reaches mlx_vlm

Status: Open.

#197 In jjang-ai/vmlx;

· ptrpspeen opened on Jun 12, 2026

Decode strips grounding-token spans (<|box_start|>(x,y)<|box_end|>) to empty string — coordinate-emitting VLMs like UI-TARS unusable via the OpenAI API

Status: Open.

#196 In jjang-ai/vmlx;

· emptinesscreature opened on Jun 11, 2026

Image Server (z-image turbo) fails to start in v1.5.54: --lora-scales requires --lora-paths

Status: Open.

#185 In jjang-ai/vmlx;

· WaveF opened on Jun 4, 2026

MiniMax-M2.7-JANGTQ_K produces gabage

Status: Open.

#179 In jjang-ai/vmlx;

· d1slike opened on May 27, 2026

swift2 rc1 无法稳定运行

Status: Open.

#167 In jjang-ai/vmlx;

· seekfutureone-lgtm opened on May 18, 2026

Outlines natively into vmlx?

Status: Open.

#160 In jjang-ai/vmlx;

· Red-fuchs opened on May 12, 2026