app.py

import os
import re
os.environ["TRANSFORMERS_CACHE"] = "/tmp/huggingface"
from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Initialize FastAPI app
app = FastAPI()

# Enable CORS (Allow all origins)
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Allow all origins
    allow_credentials=True,
    allow_methods=["*"],  # Allow all methods
    allow_headers=["*"],  # Allow all headers
)

# Load the model
MODEL_ID = "mzman123/musa-chef-gpt"
tokenizer = AutoTokenizer.from_pretrained("auhide/chef-gpt-en")
chef_gpt = AutoModelForCausalLM.from_pretrained(MODEL_ID)
print("Model Loaded")


# Define request body structure
class IngredientsRequest(BaseModel):
    ingredients: str


@app.get("/")
def home():
    return {"message": "Hello World"}


@app.post("/generate")
def generate_from_model(request: IngredientsRequest):
    """Generate a recipe from the given ingredients."""
    
    ingredients = request.ingredients.split(", ")
    print("at backend:", ingredients)

    if isinstance(ingredients, list):
        prompt_text = f"ingredients>> {', '.join(ingredients)} ; recipe>>"
        prompt_tokens = tokenizer(prompt_text, return_tensors="pt")

        print("Prompt =", prompt_text)

        output_test = chef_gpt.generate(
            prompt_tokens.input_ids,
            do_sample=True,
            max_length=1000,
            top_p=0.95,
            attention_mask=prompt_tokens.attention_mask
        )

        recipe = tokenizer.batch_decode(output_test)[0]
        print("Recipe before regex =", recipe)

        # Extract recipe part
        pattern = r"recipe>>([\s\S]*?)"
        match = re.search(pattern, recipe)
        if match:
            recipe = recipe[match.end():].strip()
        else:
            print("Recipe section not found.")

        # Clean unwanted text
        unwanted_phrases = ['<|endoftext|>', '']
        for phrase in unwanted_phrases:
            recipe = recipe.replace(phrase, '')

        recipe_lines = recipe.split("\n")

        return {"recipe": recipe_lines}

    else:
        raise HTTPException(status_code=400, detail="Input data should be a comma-separated string of ingredients")


# Run the FastAPI app (only when executing locally)
if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=int(os.getenv("PORT", 7860)))