Update app.py

app.py

@@ -91,15 +91,20 @@ def optimize_embedding(texts, precision='uint8'):
     # Step 1: Generate embeddings with 384 dimensions
     embeddings = model.encode(texts)
 
-    # Step 2: Quantize embeddings to chosen precision (e.g., uint8)
+    # Step 2: Normalize embeddings to [0, 1] range
+    embeddings_min = embeddings.min(axis=1, keepdims=True)
+    embeddings_max = embeddings.max(axis=1, keepdims=True)
+    normalized_embeddings = (embeddings - embeddings_min) / (embeddings_max - embeddings_min + 1e-8)
+
+    # Step 3: Scale normalized embeddings to fit within the range of uint8 or uint16
     if precision == 'uint8':
-        quantized_embeddings = np.array(embeddings, dtype='float32').astype('uint8')
+        scaled_embeddings = (normalized_embeddings * 255).astype('uint8')
     elif precision == 'uint16':
-        quantized_embeddings = np.array(embeddings, dtype='float32').astype('uint16')
+        scaled_embeddings = (normalized_embeddings * 65535).astype('uint16')
     else:
         raise ValueError("Unsupported precision. Use 'uint8' or 'uint16'.")
 
-    return quantized_embeddings
+    return scaled_embeddings
 
 if __name__ == "__main__":
     import uvicorn