Model compatibility and issue reporting

Model-family capabilities do not guarantee that every artifact and backend implements the same inputs, quantization or generation features. Check the exact provider, checkpoint and installed runtime versions.

  • Vision capabilities explains native image delivery, delivery metadata, model-specific limits and caption fallback.
  • Native MLX runtime documents supported Qwen3.8 MTP, concurrency, caching and current-turn image inputs.
  • Native MLX measurements records tested artifacts and measured behavior; it does not certify other models or runtime versions.
  • HuggingFace compatibility covers Transformers/GGUF loading and quantized checkpoint requirements.
  • Troubleshooting provides symptom-oriented checks and fixes.

When reporting an issue, include the exact model/artifact, provider and runtime versions, operating system, generation controls, expected/actual behavior and a minimal example. For vision, include a shareable image and image-delivery metadata. Remove credentials and private prompt/media content before posting.

Use the issue tracker.