token_source

Runnable Python SDK example token_source.

from mistralrs import Runner, Which, ChatCompletionRequest

runner = Runner(
    which=Which.GGUF(
        tok_model_id="mistralai/Mistral-7B-Instruct-v0.1",
        quantized_model_id="TheBloke/Mistral-7B-Instruct-v0.1-GGUF",
        quantized_filename="mistral-7b-instruct-v0.1.Q4_K_M.gguf",
    ),
    token_source="literal: ...",  # One of: "literal:<value>", "env:<value>", "path:<value>", "cache", "none"
)

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,
        presence_penalty=1.0,
        top_p=0.1,
        temperature=0.1,
    )
)
print(res.choices[0].message.content)
print(res.usage)

Source: examples/python/token_source.py