from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch

# Initialize FastAPI app
app = FastAPI()

# Load the Falcon-7B model with 8-bit quantization (if CUDA is available)
model_id = "tiiuae/falcon-7b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Check if CUDA is available
if torch.cuda.is_available():
    # Load the model with 8-bit quantization for GPU
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        load_in_8bit=True,
        device_map="auto",
        trust_remote_code=True
    )
else:
    # Fallback to CPU or full precision
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        device_map="auto",
        trust_remote_code=True
    )

# Create a text generation pipeline
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Define request body schema
class TextGenerationRequest(BaseModel):
    prompt: str
    max_new_tokens: int = 50
    temperature: float = 0.7
    top_k: int = 50
    top_p: float = 0.9
    do_sample: bool = True

# Define API endpoint
@app.post("/generate-text")
async def generate_text(request: TextGenerationRequest):
    try:
        # Generate text using the pipeline
        outputs = pipe(
            request.prompt,
            max_new_tokens=request.max_new_tokens,
            temperature=request.temperature,
            top_k=request.top_k,
            top_p=request.top_p,
            do_sample=request.do_sample
        )
        return {"generated_text": outputs[0]["generated_text"]}
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

# Add a root endpoint for health checks
@app.get("/test")
async def root():
    return {"message": "API is running!"}