hpyapali/tinyllama-workout

1import os2import gradio as gr3from fastapi import FastAPI, HTTPException4from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM5import uvicorn6 7# ✅ Load Model Configuration8MODEL_NAME = "hpyapali/tinyllama-workout"9HF_TOKEN = os.getenv("HF_TOKEN", "your_huggingface_api_key")  # Replace with your actual Hugging Face API key10 11app = FastAPI()12 13try:14    print("🔄 Loading Model...")15    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=HF_TOKEN)16    model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, token=HF_TOKEN)17    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)18    print("✅ Model Loaded Successfully!")19except Exception as e:20    print(f"❌ Error loading model: {e}")21    pipe = None22 23 24# ✅ AI Function - Generates Structured Workout Recommendations25def recommend_next_workout(last_workouts: str):26    """27    Analyzes and ranks workouts based on intensity and heart rate drop.28    Provides a recommendation for the next workout.29    """30    if pipe is None:31        return "❌ AI model not loaded."32 33    instruction = (34        "You are a fitness AI assistant specializing in analyzing workout effectiveness. "35        "Based on the last 7 workouts, rank them from the most to least effective based on:\n"36        "- Heart rate drop after workout (faster drop = better recovery)\n"37        "- Workout intensity (higher effort = more impact)\n"38        "- Duration (longer workouts generally contribute more)\n"39        "- Calories burned (higher calories = higher impact)\n"40        "- Variability (mixing workout types is important)\n\n"41        "### Last 7 Workouts:\n"42    )43 44    full_prompt = instruction + last_workouts + "\n\n### Ranking (Best to Least Effective):\n"45 46    try:47        print(f"🧐 AI Processing: {full_prompt}")48        result = pipe(49            full_prompt, 50            max_new_tokens=150,  # 🔼 Increased token limit for full ranking51            do_sample=True,  # 🔼 Enabled sampling for variability52            temperature=0.7,  # 🔼 Slight randomness for better insights53            top_p=0.9  # 🔼 Limits unlikely outputs while keeping diversity54        )55        print(f"🔍 Raw AI Output: {result}")56 57        if not result or not result[0]["generated_text"].strip():58            return "❌ AI did not generate any output."59 60        response_text = result[0]["generated_text"].strip()61 62        # ✅ Remove repeated prompt if AI echoes it63        if full_prompt in response_text:64            response_text = response_text.replace(full_prompt, "").strip()65 66        print(f"✅ AI Recommendation: {response_text}")67        return response_text68    except Exception as e:69        print(f"❌ AI Processing Error: {e}")70        return "❌ Error generating workout recommendation."71 72 73# ✅ FastAPI Route - Returns AI Response Directly74@app.post("/gradio_api/call/predict")75async def predict(data: dict):76    try:77        last_workouts = data.get("data", [""])[0]78        if not last_workouts:79            raise HTTPException(status_code=400, detail="Invalid input")80 81        ai_response = recommend_next_workout(last_workouts)82 83        return {"data": [ai_response]}  # ✅ Directly returning structured response84    except Exception as e:85        return {"error": str(e)}86 87 88# ✅ Gradio UI (Optional for Testing)89iface = gr.Interface(90    fn=recommend_next_workout,91    inputs="text",92    outputs="text",93    title="TinyLlama Workout Recommendations",94    description="Enter workout data to receive AI-powered recommendations."95)96 97# ✅ Ensure Proper Gradio Launch98iface.launch(server_name="0.0.0.0", server_port=7860)99 100# ✅ FastAPI Server Execution101if __name__ == "__main__":102    uvicorn.run(app, host="0.0.0.0", port=7860)103