Update app.py

app.py

@@ -2,7 +2,7 @@ import os
 import torch
 import numpy as np
 from fastapi import FastAPI, Request
-from chronos import ChronosPipeline
+from chronos import ChronosPipeline  # This works ONLY if chronos-forecasting is installed
 import uvicorn
 
 app = FastAPI()
@@ -14,12 +14,12 @@ pipeline = None
 def load_model():
     global pipeline
     print("Loading Chronos model...")
-    # Using 'tiny' first to ensure it fits in RAM. 
-    # Swap to 'amazon/chronos-t5-small' if you have confirmed 16GB+ RAM
+    # Using 'tiny' for stability on HF basic instances
+    # Ensure dtype is torch.bfloat16 for efficiency
     pipeline = ChronosPipeline.from_pretrained(
         "amazon/chronos-t5-tiny", 
-        device_map="auto",
-        dtype=torch.bfloat16
+        device_map="cpu",  # Use "cpu" if no GPU, or "auto" if you have a T4
+        torch_dtype=torch.bfloat16
     )
     print("Model loaded successfully.")
 
@@ -29,37 +29,55 @@ def home():
 
 @app.post("/predict")
 async def get_forecast(request: Request):
-    body = await request.json()
-    history = body.get("history", [])
-    user_rr = float(body.get("rr_ratio", 2.0))
-    
-    if not history:
-        return {"error": "History array is empty"}
+    global pipeline
+    try:
+        body = await request.json()
+        history = body.get("history", [])
+        user_rr = float(body.get("rr_ratio", 2.0))
+        
+        if not history:
+            return {"error": "History array is empty"}
 
-    # Chronos prediction logic
-    context = torch.tensor(history, dtype=torch.float32)
-    forecast = pipeline.predict(context, prediction_length=12)
-    
-    forecast_np = forecast.numpy()[0]
-    low_bound = np.percentile(forecast_np, 10, axis=0)
-    median_pred = np.percentile(forecast_np, 50, axis=0)
-    high_bound = np.percentile(forecast_np, 90, axis=0)
+        # 1. Convert to tensor and add batch dimension [1, seq_len]
+        # Chronos expects a batch dimension
+        context = torch.tensor(history, dtype=torch.float32).unsqueeze(0)
+        
+        # 2. Prediction logic
+        # num_samples=20 gives enough for quantiles without killing RAM
+        forecast = pipeline.predict(context, prediction_length=12, num_samples=20)
+        
+        # 3. Extract results (remove batch dimension for processing)
+        # forecast shape is [batch, samples, horizon] -> [samples, horizon]
+        forecast_np = forecast[0].numpy()
+        
+        low_bound = np.percentile(forecast_np, 10, axis=0)
+        median_pred = np.percentile(forecast_np, 50, axis=0)
+        high_bound = np.percentile(forecast_np, 90, axis=0)
 
-    entry_price = float(history[-1])
-    sl = float(low_bound[0])
-    if sl >= entry_price: sl = entry_price * 0.98
-    
-    tp = entry_price + ((entry_price - sl) * user_rr)
+        # 4. Trading Logic
+        entry_price = float(history[-1])
+        first_pred = float(median_pred[0])
+        
+        # Stop Loss (SL) at the 10th percentile
+        sl = float(low_bound[0])
+        if sl >= entry_price: 
+            sl = entry_price * 0.98 # Safety fallback
+        
+        # Take Profit (TP)
+        risk = entry_price - sl
+        tp = entry_price + (risk * user_rr)
 
-    return {
-        "prediction": median_pred.tolist(),
-        "upper_bound": high_bound.tolist(),
-        "lower_bound": low_bound.tolist(),
-        "suggested_sl": round(sl, 4),
-        "suggested_tp": round(tp, 4),
-        "verdict": "Entry Confirmed" if median_pred[0] > entry_price else "Wait"
-    }
+        return {
+            "prediction": median_pred.tolist(),
+            "upper_bound": high_bound.tolist(),
+            "lower_bound": low_bound.tolist(),
+            "suggested_sl": round(sl, 4),
+            "suggested_tp": round(tp, 4),
+            "verdict": "Entry Confirmed" if first_pred > entry_price else "Wait for better entry"
+        }
+    except Exception as e:
+        return {"error": str(e)}
 
 if __name__ == "__main__":
-    # CRITICAL: Hugging Face requires port 7860
+    # Port 7860 is required for Hugging Face Spaces
     uvicorn.run(app, host="0.0.0.0", port=7860)