GGUF¶
vllm-metal supports dense decoder GGUF checkpoints through the MLX runtime. A GGUF file carries weights only, so it needs a Hugging Face config and tokenizer source.
Local weights¶
Use a local .gguf file and point --tokenizer at the matching config and
tokenizer source:
If config.json is next to the .gguf, --tokenizer is optional.
Remote weights¶
Remote references use the same repo_id:quant shape as vLLM's GGUF plugin:
The source priority is:
vllm-metal downloads exactly one matching .gguf file from the remote
repository. Missing, ambiguous, or sharded matches fail before model load.
Current scope¶
- Supported model families: Qwen2, Qwen3, Llama, and Mistral.
- Supported qtypes: Q8_0, Q4_0, and Q4_1.
- Unsupported: K-quants, MoE, SSM or hybrid models, vision models, fused-QKV GGUFs, and sharded GGUF files.