new-qwen/serv/llm.py

105 lines
3.8 KiB
Python

from __future__ import annotations
import json
from typing import Any
from urllib import request
from config import ServerConfig
from oauth import OAuthError, QwenOAuthManager
from tools import ToolError, ToolRegistry
DEFAULT_SYSTEM_PROMPT = """You are new-qwen-serv, a server-side coding agent.
You help the user through a remote client.
Use tools when they are necessary.
Be concise, precise and action-oriented.
When you modify files, explain what changed.
Do not claim to have executed tools unless a tool result confirms it."""
class QwenAgent:
def __init__(self, config: ServerConfig, oauth: QwenOAuthManager, tools: ToolRegistry) -> None:
self.config = config
self.oauth = oauth
self.tools = tools
def _request_completion(self, messages: list[dict[str, Any]]) -> dict[str, Any]:
creds = self.oauth.get_valid_credentials()
base_url = self.oauth.get_openai_base_url(creds)
payload = {
"model": self.config.model,
"messages": messages,
"tools": self.tools.schemas(),
"tool_choice": "auto",
}
data = json.dumps(payload).encode("utf-8")
req = request.Request(
f"{base_url}/chat/completions",
data=data,
headers={
"Content-Type": "application/json",
"Authorization": f"Bearer {creds['access_token']}",
},
method="POST",
)
with request.urlopen(req, timeout=180) as response:
return json.loads(response.read().decode("utf-8"))
def run(self, history: list[dict[str, Any]], user_message: str) -> dict[str, Any]:
system_prompt = self.config.system_prompt or DEFAULT_SYSTEM_PROMPT
messages: list[dict[str, Any]] = [{"role": "system", "content": system_prompt}]
messages.extend(history)
messages.append({"role": "user", "content": user_message})
events: list[dict[str, Any]] = []
for _ in range(self.config.max_tool_rounds):
response = self._request_completion(messages)
choice = response["choices"][0]["message"]
tool_calls = choice.get("tool_calls") or []
content = choice.get("content")
if content:
events.append({"type": "assistant", "content": content})
if not tool_calls:
return {
"answer": content or "",
"events": events,
"usage": response.get("usage"),
"messages": messages + [{"role": "assistant", "content": content or ""}],
}
messages.append(
{
"role": "assistant",
"content": content or "",
"tool_calls": tool_calls,
}
)
for call in tool_calls:
tool_name = call["function"]["name"]
try:
arguments = json.loads(call["function"]["arguments"] or "{}")
except json.JSONDecodeError:
arguments = {}
events.append({"type": "tool_call", "name": tool_name, "arguments": arguments})
try:
result = self.tools.execute(tool_name, arguments)
except ToolError as exc:
result = {"error": str(exc)}
except Exception as exc:
result = {"error": f"Unexpected tool failure: {exc}"}
events.append({"type": "tool_result", "name": tool_name, "result": result})
messages.append(
{
"role": "tool",
"tool_call_id": call["id"],
"content": self.tools.encode_result(result),
}
)
raise OAuthError("Max tool rounds exceeded")