33 lines
1.4 KiB
Python
33 lines
1.4 KiB
Python
from typing import Any, Dict
|
|
from ollama import Client
|
|
from langchain_core.runnables.base import Runnable
|
|
|
|
class OllamaWrapper(Runnable):
|
|
def __init__(self, host: str, model: str = "llama3.2:latest"):
|
|
self.client = Client(host=host)
|
|
self.model = model
|
|
|
|
def invoke(self, prompt: Any, config: Dict[str, Any] = None, **kwargs: Any) -> str:
|
|
# Handle different prompt types (StringPromptValue, str, etc.)
|
|
if hasattr(prompt, 'to_string'):
|
|
prompt = prompt.to_string()
|
|
elif not isinstance(prompt, str):
|
|
prompt = str(prompt)
|
|
|
|
# Use the model from constructor, but allow override via kwargs
|
|
model_name = kwargs.get("model", self.model)
|
|
options = {
|
|
"temperature": kwargs.get("temperature"),
|
|
"top_p": kwargs.get("top_p"),
|
|
"max_tokens": kwargs.get("max_tokens"),
|
|
}
|
|
messages = [{"role": "user", "content": prompt}]
|
|
response = self.client.chat(model=model_name, messages=messages, options=options)
|
|
if response and "message" in response and "content" in response["message"]:
|
|
return response["message"]["content"]
|
|
else:
|
|
raise ValueError("Invalid response structure from model")
|
|
|
|
async def ainvoke(self, prompt: Any, config: Dict[str, Any] = None, **kwargs: Any) -> str:
|
|
return self.invoke(prompt, config, **kwargs)
|