app.py

from fastapi import FastAPI, HTTPException, Request, Response
from pydantic import BaseModel
from typing import List, Optional
from llama_cpp import Llama
from fastapi.responses import PlainTextResponse, JSONResponse
from starlette.middleware.base import BaseHTTPMiddleware


import logging
import json
import os
import time
import uuid

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("api_logger")

class LoggingMiddleware(BaseHTTPMiddleware):
    async def dispatch(self, request: Request, call_next):
        # Read request body (must be buffered manually)
        body = await request.body()
        logger.info(f"REQUEST: {request.method} {request.url}\nBody: {body.decode('utf-8')}")
        
        # Rebuild the request with body for downstream handlers
        request = Request(request.scope, receive=lambda: {'type': 'http.request', 'body': body})

        # Process the response
        response = await call_next(request)
        response_body = b""
        async for chunk in response.body_iterator:
            response_body += chunk

        # Log response body and status code
        logger.info(f"RESPONSE: Status {response.status_code}\nBody: {response_body.decode('utf-8')}")

        # Rebuild response to preserve original functionality
        return Response(
            content=response_body,
            status_code=response.status_code,
            headers=dict(response.headers),
            media_type=response.media_type
        )

# FastAPI app with middleware
app = FastAPI()
app.add_middleware(LoggingMiddleware)

llm = None

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

class ChatRequest(BaseModel):
    model: str
    messages: List[Message]
    temperature: Optional[float] = 0.7
    max_tokens: Optional[int] = 256

class GenerateRequest(BaseModel):
    model: str
    prompt: str
    max_tokens: Optional[int] = 256
    temperature: Optional[float] = 0.7


class ModelInfo(BaseModel):
      id: str
      object: str
      type: str
      publisher: str
      arch: str
      compatibility_type: str
      quantization: str
      state: str
      max_context_length: int

AVAILABLE_MODELS = [
    ModelInfo(
        id="codellama-7b-instruct",
        object="model",
        type="llm",
        publisher="lmstudio-community",
        arch="llama",
        compatibility_type="gguf",
        quantization="Q4_K_M",
        state="loaded",
        max_context_length=32768
    )
]


@app.on_event("startup")
def load_model():
    global llm
    model_path_file = "/tmp/model_path.txt"
    if not os.path.exists(model_path_file):
        raise RuntimeError(f"Model path file not found: {model_path_file}")
    with open(model_path_file, "r") as f:
        model_path = f.read().strip()
    if not os.path.exists(model_path):
        raise RuntimeError(f"Model not found at path: {model_path}")
    llm = Llama(model_path=model_path)

@app.get("/", response_class=PlainTextResponse)
async def root():
    return "Ollama is running"

@app.get("/health")
async def health_check():
    return {"status": "ok"}

@app.get("/api/tags")
async def api_tags():
    return JSONResponse(content={
        "data": [model.dict() for model in AVAILABLE_MODELS]
    })

@app.get("/models")
async def list_models():
    # Return available models info
    return [model.dict() for model in AVAILABLE_MODELS]

@app.get("/api/v0/models")
async def api_models():
    return {"data": [model.dict() for model in AVAILABLE_MODELS]}

@app.get("/models/{model_id}")
async def get_model(model_id: str):
    for model in AVAILABLE_MODELS:
        if model.id == model_id:
            return model.dict()
    raise HTTPException(status_code=404, detail="Model not found")

@app.post("/chat")
async def chat(req: ChatRequest):
    global llm
    if llm is None:
        return {"error": "Model not initialized."}

    # Validate model - simple check
    if req.model not in [m.id for m in AVAILABLE_MODELS]:
        raise HTTPException(status_code=400, detail="Unsupported model")

    # Construct prompt from messages
    prompt = ""
    for m in req.messages:
        prompt += f"{m.role}: {m.content}\n"
    prompt += "assistant:"

    output = llm(
        prompt,
        max_tokens=req.max_tokens,
        temperature=req.temperature,
        stop=["user:", "assistant:"]
    )
    text = output.get("choices", [{}])[0].get("text", "").strip()

    response = {
        "id": str(uuid.uuid4()),
        "model": req.model,
        "choices": [
            {
                "message": {"role": "assistant", "content": text},
                "finish_reason": "stop"
            }
        ]
    }
    return response

@app.post("/api/v0/generate")
async def api_generate(req: GenerateRequest):
    global llm
    if llm is None:
        raise HTTPException(status_code=503, detail="Model not initialized")

    if req.model not in [m.id for m in AVAILABLE_MODELS]:
        raise HTTPException(status_code=400, detail="Unsupported model")

    output = llm(
        req.prompt,
        max_tokens=req.max_tokens,
        temperature=req.temperature,
        stop=["\n\n"]  # Or any stop sequence you want
    )
    text = output.get("choices", [{}])[0].get("text", "").strip()

    return {
        "id": str(uuid.uuid4()),
        "model": req.model,
        "choices": [
            {
                "text": text,
                "index": 0,
                "finish_reason": "stop"
            }
        ]
    }