lmstudio-ai/mlx-engine

View on GitHub

LM Studio Apple MLX engine

Stars: 1.2k
Forks: 133
Open beginner issues: 0
Indexed issues: 87
Avg merge: 21h 6m
Merged PRs (30d): 1

Dominant language: Python
License: MIT
Last GitHub push: Aug 21, 2026
Latest indexed: Sep 19, 2026
Contributing guide: Contributing guide
Code of conduct: No code of conduct
Beginner labels: good first issue

Beginner friendly Everything

Open All

87 open issues indexed

BatchedModelKit does not respect multiple EOS token IDs from model config Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

lmstudio-ai/mlx-engine#374 · Opened Sep 5, 2026
MLX server (#353) as an Engine Protocol runtime: use case + offer to help Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

lmstudio-ai/mlx-engine#373 · 3 comments · Opened Sep 2, 2026
glm5_next (GLM-5.3-Flash): prompt cache save always fails — save_safetensors on zero-size array in batched-vision blob store; plus latent hybrid-cache bug in CacheWrapper._num_tokens_in_cache Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

lmstudio-ai/mlx-engine#372 · Opened Aug 31, 2026
Draft-only (MTP) repos indexed as loadable models, then crash at generation: 'Qwen3_5MTPDraftModel' object has no attribute 'get_input_embeddings' Open

Difficulty 3/5 1-2 days Newbie friendliness 65/100

lmstudio-ai/mlx-engine#371 · 1 comment · 1 reaction · Opened Aug 30, 2026
[Feature] MLX engine ignores built-in MTP (nextn) heads — native MTP speculative decoding missing on the MLX path Open

Difficulty 5/5 Over a week Newbie friendliness 42/100

lmstudio-ai/mlx-engine#370 · 1 comment · 2 reactions · Opened Aug 21, 2026
Batched vision models cannot use a draft model: is_draft_model_compatible() returns False unconditionally Open

Difficulty 4/5 3-5 days Newbie friendliness 38/100

lmstudio-ai/mlx-engine#368 · 3 reactions · Opened Aug 19, 2026
MLX auto-fit replaces the configured context length and disagrees with the prompt cache budget Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

lmstudio-ai/mlx-engine#366 · 1 comment · 5 reactions · Opened Aug 15, 2026
Partial GPU Offloading support for MLX Qwen3.6-35b-a3b 8bit on M4 Mac 48GB? Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#365 · Opened Aug 13, 2026
Backend scheduler crashes with broadcast_shapes ValueError in BatchRotatingKVCache.merge under concurrent requests with shared prefix Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lmstudio-ai/mlx-engine#363 · Opened Aug 7, 2026
repeat_penalty causes fatal backend-scheduler crashes in multi-turn conversations (32-bit slice error / KV cache merge broadcast error) Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

lmstudio-ai/mlx-engine#359 · Opened Aug 5, 2026
Text-only architectures implemented only in mlx_vlm (e.g. deepseek_v4) fail to load because dispatch checks only for vision_config Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

lmstudio-ai/mlx-engine#357 · 1 reaction · Opened Aug 1, 2026
Disk prefix KV caching Open

Difficulty 5/5 Over a week Newbie friendliness 42/100

lmstudio-ai/mlx-engine#354 · 1 comment · 2 reactions · Opened Jul 24, 2026
Please support for Ternary Bonsai 27b with vision an reasoning Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

lmstudio-ai/mlx-engine#347 · 2 comments · 5 reactions · Opened Jul 15, 2026
Bug: Batched VLM loader treats mtp.safetensors as base weights for Qwen3.5-4B-OptiQ-4bit Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

lmstudio-ai/mlx-engine#343 · 3 comments · Opened Jul 8, 2026
Feature: fused batched decode for concurrent requests (close the gap to mlx_lm.batch_generate) Open

Difficulty 5/5 Over a week Newbie friendliness 30/100

lmstudio-ai/mlx-engine#342 · Opened Jun 30, 2026
[Feature Request] Add configuration toggle to disable prompt disk caching to prevent excessive SSD wear Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lmstudio-ai/mlx-engine#341 · 6 reactions · Opened Jun 25, 2026
We hope to upgrade the mlx-vlm version to 0.6.2 or above to resolve the issue of VLM lacking the qwen3_5_moe_vision type, which prevents the Holo 3.1 model from being used. Open

Difficulty 2/5 1-3 hours Newbie friendliness 68/100

lmstudio-ai/mlx-engine#339 · 3 comments · 3 reactions · Opened Jun 16, 2026
Gemma 4 (26B-A4B) reasoning never terminates on MLX — fills max_tokens with reasoning_content, empty content (GGUF terminates fine) Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

lmstudio-ai/mlx-engine#337 · 1 reaction · Opened Jun 14, 2026
DiffusionGemma Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#336 · 1 comment · 4 reactions · Opened Jun 12, 2026
Qwen3.6 35B A3B Mxfp8 - failed to load Open

Difficulty 3/5 1-2 days Newbie friendliness 52/100

lmstudio-ai/mlx-engine#330 · 2 comments · Opened May 24, 2026
kv cache starts back from 0 every other turn or so in all models and in all harnesses Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lmstudio-ai/mlx-engine#327 · 1 comment · Opened May 15, 2026
Image tokens not included when running Qwen3.6-35B-A3B MLX (tokens: 0, features N) Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lmstudio-ai/mlx-engine#325 · 1 comment · 1 reaction · Opened May 11, 2026
Support for model type Zaya? Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lmstudio-ai/mlx-engine#324 · 1 reaction · Opened May 10, 2026
New MTP models not supported Open

enhancement

lmstudio-ai/mlx-engine#323 · 25 comments · 60 reactions · 1 assignee · Opened May 8, 2026
Double RMSNorm in PatchedQwen3_5TextModel Open

Difficulty 1/5 Under an hour Newbie friendliness 88/100

lmstudio-ai/mlx-engine#322 · Opened May 5, 2026
[Feature Request] Support weightless RMSNorm (for FlashNorm weight folding trick) Open

Difficulty 3/5 1-2 days Newbie friendliness 58/100

lmstudio-ai/mlx-engine#321 · Opened May 1, 2026
Laguna-XS.2 Support Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#320 · Opened Apr 30, 2026
[Feature]: Support encoder models Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

lmstudio-ai/mlx-engine#318 · 2 reactions · Opened Apr 22, 2026
Nemotron 3 Nano MLX no longer loading as of 1.5.0 (and 1.6.0) Open

Difficulty 3/5 1-2 days Newbie friendliness 48/100

lmstudio-ai/mlx-engine#316 · 1 comment · Opened Apr 18, 2026
Multi-round inference leads to insufficient memory crash Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

lmstudio-ai/mlx-engine#314 · 4 comments · Opened Apr 18, 2026
[feature] Support for TriAttention Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

lmstudio-ai/mlx-engine#313 · 1 reaction · Opened Apr 17, 2026
mlx1.6.0仍然无法加载gemma4 Open

Difficulty 3/5 1-2 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#312 · 2 comments · 1 reaction · Opened Apr 15, 2026
'list' object has no attribute 'swapaxes' — deterministic cache-rebuild crash on GLM-4.7-Flash (glm4_moe_lite MLA) under Parallel>1 Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

lmstudio-ai/mlx-engine#310 · Opened Apr 11, 2026
LRU text cache and image checkpointing for VisionAddOn Open

Difficulty 4/5 3-5 days Newbie friendliness 48/100

lmstudio-ai/mlx-engine#307 · 5 comments · 1 reaction · Opened Apr 8, 2026
it doesn't seem to support 1bit Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

lmstudio-ai/mlx-engine#306 · 3 reactions · Opened Apr 7, 2026
Support for 1bit models Open

Difficulty 4/5 3-5 days Newbie friendliness 52/100

lmstudio-ai/mlx-engine#299 · 5 comments · 57 reactions · Opened Mar 31, 2026
[Feature]: turboquant: KV cache Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

lmstudio-ai/mlx-engine#296 · 2 comments · 43 reactions · Opened Mar 28, 2026
Support Flash MoE Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

lmstudio-ai/mlx-engine#294 · 1 reaction · Opened Mar 23, 2026
oMLX Support Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#290 · 13 reactions · Opened Mar 17, 2026
MLX vision models do not currently support KV cache quantization for qwen3.5 Open

Difficulty 3/5 1-2 days Newbie friendliness 52/100

lmstudio-ai/mlx-engine#286 · 5 comments · 20 reactions · Opened Mar 7, 2026
Wide performance discrepancy between LM Studio MLX and `mlx_vlm.generate` with Qwen3.5-35B-A3B Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#285 · 1 comment · Opened Mar 6, 2026
Error when loading model: ValueError: Model type qwen3_5 not supported. Open

Difficulty 4/5 3-5 days Newbie friendliness 32/100

lmstudio-ai/mlx-engine#284 · 1 comment · 9 reactions · Opened Mar 3, 2026
QK-Normalization Support Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#283 · 1 comment · 1 reaction · Opened Feb 28, 2026
`mode` was ignored in `model_kit.vision_add_ons.load_utils.maybe_apply_quantization`, causing `mxfp8` and such quants to fail. Open

Difficulty 3/5 1-2 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#282 · Opened Feb 22, 2026
Speculative decoding fails (almost) silently instead of clearly notifying users about draft model loading issues Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

lmstudio-ai/mlx-engine#280 · 1 comment · Opened Feb 20, 2026
Support for ouro Architecture Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

lmstudio-ai/mlx-engine#277 · 1 comment · 3 reactions · Opened Feb 16, 2026
🥲 Failed to load the model:- lfm2-vl-3b Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

lmstudio-ai/mlx-engine#275 · Opened Feb 15, 2026
SpeculativeDecodingNotSupportedError: Speculative decoding is not supported for batched MLX models. Open

Difficulty 3/5 1-2 days Newbie friendliness 48/100

lmstudio-ai/mlx-engine#269 · 11 comments · 2 reactions · Opened Feb 7, 2026
Error in iterating prediction stream: RuntimeError: [metal::malloc] Resource limit (499000) exceeded. Open

Difficulty 4/5 3-5 days Newbie friendliness 38/100

lmstudio-ai/mlx-engine#264 · 5 comments · 2 reactions · Opened Jan 28, 2026
Feature: Support for OpenResponses Spec Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

lmstudio-ai/mlx-engine#263 · 1 comment · Opened Jan 28, 2026

lmstudio-ai/mlx-engine

Get new issues in your inbox