app.py

from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import uvicorn
import os

os.environ['TRANSFORMERS_CACHE'] = '/tmp/transformers_cache'
os.environ['TORCH_HOME'] = '/tmp/torch_cache'

app = FastAPI(title="DIANA - Diet And Nutrition Assistant")

app.add_middleware(CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"])

DEVICE = torch.device('cpu')
torch.set_num_threads(4)
torch.set_grad_enabled(False)

model = None
tokenizer = None
MODEL_LOADED = False

def load_model():
    global model, tokenizer, MODEL_LOADED
    try:
        print("Starting model load...")
        model_name = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
        
        tokenizer = AutoTokenizer.from_pretrained(
            model_name,
            cache_dir='/tmp/transformers_cache',
            use_fast=True
        )
        
        model = AutoModelForCausalLM.from_pretrained(
            model_name,
            torch_dtype=torch.float32,
            low_cpu_mem_usage=True,
            device_map=None,
            cache_dir='/tmp/transformers_cache'
        ).to(DEVICE)
        
        model.eval()
        MODEL_LOADED = True
        return True
    except Exception as e:
        print(f"Error loading model: {str(e)}")
        MODEL_LOADED = False
        return False

print("Initiating DIANA...")
load_model()

class Query(BaseModel):
    prompt: str
    max_length: int = 150
    temperature: float = 0.7

def get_structured_response(topic):
    return f"""Here's what you need to know about {topic}:

1. Start with the basics:
   • Begin gradually
   • Focus on proper form
   • Stay consistent

2. Key points to remember:
   • Set realistic goals
   • Track your progress
   • Listen to your body

3. Tips for success:
   • Start today, not tomorrow
   • Keep it simple
   • Stay motivated

Need more specific advice about any of these points?

- DIANA 💪"""

def is_greeting(text):
    return any(g in text.lower() for g in ['hi', 'hello', 'hey'])

@app.post("/chat")
async def chat(query: Query):
    if not MODEL_LOADED:
        raise HTTPException(status_code=503, detail="DIANA is initializing. Please try again.")
    
    try:
        # Handle greetings
        if is_greeting(query.prompt):
            return {"response": "Hi! I'm DIANA, your fitness assistant. How can I help you today?\n\n- DIANA 💪"}
        
        # Optimized but complete prompt template
        system_prompt = f"""You are DIANA, a fitness assistant. Give clear, complete advice about {query.prompt}.
        Structure your response like this:
        1. Brief welcome and intro
        2. 3 main points with bullets
        3. Encouraging conclusion
        4. Sign with '- DIANA 💪'
        IMPORTANT: Never end mid-sentence. Always complete your thoughts."""

        formatted_prompt = f"<|system|>{system_prompt}</s><|user|>Give structured fitness advice about: {query.prompt}</s><|assistant|>Let me help you with that!\n\n"
        
        inputs = tokenizer(
            formatted_prompt,
            return_tensors="pt",
            truncation=True,
            max_length=200,
            padding=False
        ).to(DEVICE)
        
        with torch.inference_mode():
            outputs = model.generate(
                inputs["input_ids"],
                max_new_tokens=150,
                min_new_tokens=100,  # Ensure minimum length
                temperature=0.7,
                top_p=0.9,
                do_sample=True,
                pad_token_id=tokenizer.eos_token_id,
                repetition_penalty=1.2,
                no_repeat_ngram_size=3,
                eos_token_id=tokenizer.eos_token_id,  # Proper ending
                num_beams=1,
                early_stopping=True,
                use_cache=True
            )
        
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
        response = response.split("Let me help you with that!")[-1].strip()
        
        # Validate response completeness
        sentences = [s.strip() for s in response.split('.') if s.strip()]
        words = response.split()
        
        # If response might be incomplete, use structured format
        if len(sentences) < 4 or len(words) < 50 or not response.endswith(('!', '.', '?', '💪')):
            return {"response": get_structured_response(query.prompt)}
        
        # Ensure proper signature
        if "- DIANA 💪" not in response:
            response += "\n\n- DIANA 💪"
        
        return {"response": response}
        
    except Exception as e:
        print(f"Error: {str(e)}")
        return {"response": get_structured_response(query.prompt)}

@app.get("/")
def read_root():
    return {"status": "DIANA is ready!", "model_loaded": MODEL_LOADED}

if __name__ == "__main__":
    uvicorn.run("app:app", host="0.0.0.0", port=7860)