lark_llg
Runnable Python SDK example lark_llg.
from mistralrs import Runner, Which, ChatCompletionRequest
runner = Runner( which=Which.Plain( model_id="microsoft/Phi-3.5-mini-instruct", ), num_device_layers=["500"],)
# see https://github.com/guidance-ai/llguidance/blob/main/docs/syntax.md for docs on syntax
top_lark = r"""start: "Reasoning: " /.+/ "\nJSON: " answeranswer: %json { "type": "object", "properties": { "answer": {"type": "string", "enum": ["Yes", "No"]} }, "required": ["answer"], "additionalProperties": false}"""
res = runner.send_chat_completion_request( ChatCompletionRequest( model="default", messages=[ { "role": "user", "content": "If all dogs are mammals, and all mammals are animals, are dogs animals?", } ], max_tokens=100, temperature=0.1, grammar_type="lark", grammar=top_lark, ))print(res.choices[0].message.content)print(res.usage)Source: examples/python/lark_llg.py