gguf.py

"""Load a GGUF model from Hugging Face.

Configuration and tokenizer assets are discovered automatically. Set `tok_model_id` only to
override that choice or when the source cannot be identified.
"""

from mistralrs import Runner, Which, ChatCompletionRequest

runner = Runner(
    which=Which.GGUF(
        quantized_model_id="unsloth/Qwen3-0.6B-GGUF",
        quantized_filename="Qwen3-0.6B-Q4_K_M.gguf",
    )
)

res = runner.send_chat_completion_request(
    ChatCompletionRequest(
        model="default",
        messages=[
            {"role": "user", "content": "Tell me a story about the Rust type system."}
        ],
        max_tokens=256,
    )
)
print(res.choices[0].message.content)
print(res.usage)