app.py

from flask import Flask, request, jsonify
from flask_cors import CORS
import os
import requests

app = Flask(__name__)
CORS(app)  # Keeps your Acode terminal connected securely!

# Pulls your hidden WRITE token from Space secrets
HF_TOKEN = os.getenv("HF_TOKEN")
MODEL_REPO = "unsloth/mistral-7b-instruct-v0.3-bnb-4bit"
API_URL = f"https://api-inference.huggingface.co/models/{MODEL_REPO}"

@app.route("/", methods=["GET"])
def home():
    return "Krypton-1 Core Unsloth Node Active."

@app.route("/api/chat", methods=["POST"])
def chat():
    try:
        data = request.json
        user_prompt = data.get("prompt", "")
        
        headers = {
            "Authorization": f"Bearer {HF_TOKEN}",
            "Content-Type": "application/json"
        }
        
        # Structure the payload exactly how the serverless API wants it
        payload = {
            "inputs": f"<s>[INST] {user_prompt} [/INST]",
            "parameters": {
                "max_new_tokens": 250,
                "temperature": 0.7,
                "return_full_text": False
            }
        }
        
        response = requests.post(API_URL, headers=headers, json=payload)
        res_data = response.json()
        
        # Check if the API tells us the model is still loading weights on their servers
        if isinstance(res_data, dict) and "error" in res_data:
            err_msg = res_data.get("error", "")
            if "loading" in err_msg.lower():
                est_time = round(res_data.get('estimated_time', 20))
                return jsonify({"reply": f"SYS_BOOT: Krypton-1 is loading on Hugging Face clusters. Activation wait time: {est_time}s. Try hitting RUN again in a moment!"})
            return jsonify({"reply": f"SYS_ALERT: API Error. Details: {err_msg}"})
            
        # Extract the text cleanly
        if isinstance(res_data, list) and len(res_data) > 0:
            reply = res_data[0].get("generated_text", "Data packet empty.")
            return jsonify({"reply": reply})
        else:
            return jsonify({"reply": f"SYS_ALERT: Unexpected API data layout format. Received: {str(res_data)}"})
        
    except Exception as e:
        return jsonify({"reply": f"SYS_ERR: Proxy pipeline broken. {str(e)}"})

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