app.py

import json
import time
from contextlib import asynccontextmanager
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel

llm = None

@asynccontextmanager
async def lifespan(app: FastAPI):
    global llm
    from llama_cpp import Llama

    print("Descargando modelo Qwen2.5-Coder-7B... puede tardar varios minutos.")
    start = time.time()

    llm = Llama.from_pretrained(
        repo_id="bartowski/Qwen2.5-Coder-7B-Instruct-GGUF",
        filename="Qwen2.5-Coder-7B-Instruct-Q4_K_M.gguf",
        n_ctx=4096,
        n_threads=2,
        n_batch=256,
        verbose=False,
    )

    elapsed = time.time() - start
    print(f"Modelo listo en {elapsed:.0f} segundos")
    yield

app = FastAPI(title="Code Agent LLM", lifespan=lifespan)

class Message(BaseModel):
    role: str
    content: str

class Tool(BaseModel):
    type: str = "function"
    function: dict

class ChatRequest(BaseModel):
    model: str = "qwen2.5-coder-7b"
    messages: list[Message]
    tools: list[Tool] = None
    tool_choice: str = "auto"
    temperature: float = 0.7
    max_tokens: int = 2048
    stream: bool = False

def build_system_with_tools(tools):
    tools_str = json.dumps(
        [t.model_dump() if hasattr(t, "model_dump") else t for t in tools],
        indent=2,
    )
    return (
        "You are a helpful coding assistant with access to tools.\n"
        "When a tool can help, call it by responding ONLY with a JSON block:\n"
        "```tool_call\n"
        '{"name": "<tool_name>", "arguments": {<args>}}\n'
        "```\n"
        "Do NOT include any other text when making a tool call.\n"
        "When answering directly, respond normally in markdown.\n\n"
        f"Available tools:\n{tools_str}"
    )

@app.get("/v1/chat/completions")
async def chat_completion_get():
    return {"status": "ok", "model": "qwen2.5-coder-7b"}

@app.post("/v1/chat/completions")
async def chat_completion(request: ChatRequest):
    if llm is None:
        raise HTTPException(status_code=503, detail="Model still loading")
    try:
        messages = []
        if request.tools:
            messages.append({
                "role": "system",
                "content": build_system_with_tools(request.tools),
            })
        for m in request.messages:
            messages.append({"role": m.role, "content": m.content})
        response = llm.create_chat_completion(
            messages=messages,
            temperature=request.temperature,
            max_tokens=request.max_tokens,
        )
        return response
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.get("/v1/models")
async def list_models():
    return {
        "data": [{
            "id": "qwen2.5-coder-7b",
            "object": "model",
            "created": int(time.time()),
            "owned_by": "local",
        }]
    }

@app.get("/health")
async def health():
    return {"status": "ok", "model_loaded": llm is not None}