Search results

ERROR:vmlx_engine.models.mllm:VLM stream_generate error: RuntimeError: There is no Stream(gpu, 0) in current thread

Status: Open.

#139 In jjang-ai/vmlx;

· ghl116 opened on May 3, 2026

Bug: --kv-cache-quantization {q4|q8|none} silently overridden by unconditional TurboQuant patching on hybrid models

Status: Open.

#138 In jjang-ai/vmlx;

· st-adam opened on May 2, 2026

Docs: prompt_lookup.py docstring still references pre-#26 design ("Phase 2 future", "blocked by mlx-lm")

Status: Open.

#137 In jjang-ai/vmlx;

· st-adam opened on May 2, 2026

Feature: native PFlash-style speculative prefill (importance-scored sparse prefill) for long-context cold TTFT

Status: Open.

#136 In jjang-ai/vmlx;

· st-adam opened on May 2, 2026

Perf: enable draft-model speculative decoding under continuous batching (lift speculative.py:184 gate)

Status: Open.

#135 In jjang-ai/vmlx;

· st-adam opened on May 2, 2026

Perf: PLD verify-cost on hybrid SSM models — proposal for SSM checkpoint/replay (extends #26)

Status: Open.

#134 In jjang-ai/vmlx;

· st-adam opened on May 2, 2026

Can't open directory in Swift Version

Status: Open.

#133 In jjang-ai/vmlx;

· hareluya opened on May 2, 2026

AttributeError: 'Model' object has no attribute 'model' when converting Gemma 4 MoE with AWQ

Status: Open.

#132 In jjang-ai/vmlx;

· Zeafer opened on May 2, 2026

MCPServerConfig rejects headers; can't connect to remote MCP servers (Exa, GitHub, etc.)

Status: Open.

#131 In jjang-ai/vmlx;

· BlueEmb3r opened on May 1, 2026

Expose logprobs / top_logprobs in /v1/completions and /v1/chat/completions

Status: Open.

#127 In jjang-ai/vmlx;

· xukai92 opened on May 1, 2026

M1 Max just shut off 2 times

Status: Open.

#126 In jjang-ai/vmlx;

· Kulimeya opened on Apr 30, 2026

flash-moe

Status: Open.

#125 In jjang-ai/vmlx;

· 4external opened on Apr 30, 2026