Search results

Gemma 4 with thinking enabled produces only reasoning field, content is empty in API response

Status: Open.

#1352 In ml-explore/mlx-lm;

· liuyangxy opened on Jun 5, 2026

Feature Request: Implement Adaptive PFlash (Self-Tuning Prefill) for Apple Silicon (MLX/llama.cpp)

Status: Open.

#1350 In ml-explore/mlx-lm;

· Revive-Curiosity opened on Jun 4, 2026

Post-val-pass hang at iter 1 with LoRA rank 16 + 7 target modules on Qwen3-8B-4bit

Status: Open.

#1348 In ml-explore/mlx-lm;

· mcf-capx opened on Jun 4, 2026

Tool calls dropped: json_tools start marker never matches when the tokenizer merges the <tool_call> closing >

Status: Open.

#1335 In ml-explore/mlx-lm;

· snagnever opened on May 31, 2026

DeepSeek-V4 (Flash/Pro) on Apple Silicon: unbounded Metal residency growth during decode → [metal::malloc] Resource limit (499000) exceeded after ~11K tokens

Status: Open.

#1332 In ml-explore/mlx-lm;

· snagnever opened on May 31, 2026

mlx_lm.server returns 404 "list index out of range" on short prompts (TokenizerWrapper._find doesn't clamp negative start)

Status: Open.

#1326 In ml-explore/mlx-lm;

· snagnever opened on May 29, 2026

Feature Request: Expose ⁠--kv-bits⁠ and thinking parameters in ⁠mlx_lm.server⁠ for large model inference

Status: Open.

#1308 In ml-explore/mlx-lm;

· Wolfbane1 opened on May 24, 2026

Devstral-Small-2-24B-Instruct-2512 (MXFP4): model emits valid [TOOL_CALLS] markup but mlx_lm.server returns empty message

Status: Open.

#1307 In ml-explore/mlx-lm;

· aanjhan opened on May 24, 2026

Mlx qwopus model doesn't works

Status: Open.

#1304 In ml-explore/mlx-lm;

· Wolfbane1 opened on May 23, 2026

why is this mlx shit not maintained ?>??? there is so much stuff going on deepseek v4 mimo 2.5 and suport? nothing .... and people who want to contribute got ignored ...

Status: Open.

#1303 In ml-explore/mlx-lm;

· jonipupet opened on May 23, 2026

converted repos missing some eos ids for mlx-community/Ministral-3-3B-Instruct-2512-4bit, mlx-community/Qwen3.5-9B-MLX-4bit

Status: Open.

#1296 In ml-explore/mlx-lm;

· jrp2014 opened on May 21, 2026

feat: add tool parser support for Qwen 3.5/3.6 (non-Coder) models

Status: Open.

#1293 In ml-explore/mlx-lm;

· TinyWorkshopDesign opened on May 21, 2026