What problem does it solve? LM Studio's model listing API can respond normally while chat completions time out indefinitely, leaving you unsure whether the model is stuck loading or just busy. This Skill provides a systematic diagnostic workflow to distinguish a stuck JIT load from normal processing and recover without guesswork. ## Core Features & Use Cases - Symptom-Based Diagnosis: Compare /v1/models response speed, worker process CPU patterns, and memory pressure to pinpoint whether the JIT loader is stuck in a Metal kernel wait. - Stuck vs. Busy Differentiation: A decision table separates a hung loader (flat ~45% CPU for 5+ minutes) from a busy inference queue (pulsing CPU), preventing counterproductive restarts or duplicate requests. - Known Pitfall Coverage: Documents Qwen3.5 thinking models returning empty content at low max_tokens, and OptiQ mixed-precision MLX builds failing to load in LM Studio. - Use Case: Your agent's LLM calls to LM Studio suddenly time out. Run the inspection commands to check worker CPU and memory, confirm a stuck JIT load, then fully quit and relaunch LM Studio with only the needed model loaded. ## Quick Start Ask the agent to diagnose why LM Studio chat completions are timing out while the models endpoint still responds.
