Update server.py

server.py

@@ -8,8 +8,9 @@ import time
 from flask import Flask, request, jsonify, Response
 from flask_cors import CORS
 
-HF_REPO    = "litert-community/gemma-4-E2B-it-litert-lm"
-HF_FILE    = "gemma-4-E2B-it.litertlm"
+# --- KEY CHANGE: Updated to 12B Model ---
+HF_REPO    = "litert-community/gemma-4-12B-it-litert-lm"
+HF_FILE    = "gemma-4-12B-it.litertlm"
 
 _SERVER_DIR   = os.path.dirname(os.path.abspath(__file__))
 _DEFAULT_PATH = os.path.join(_SERVER_DIR, "models", "gemma", HF_FILE)
@@ -157,7 +158,7 @@ def extract_request() -> tuple[str, bytes | None, bool]:
 @app.route("/")
 def index():
     return jsonify({
-        "service": "Gemma 4 API",
+        "service": "Gemma 4 API (12B Model)",
         "model_status": model_status,
         "guide": {
             "1. Text Only Chat": {
@@ -212,7 +213,6 @@ def gemma():
         def generate_stream():
             try:
                 for text_chunk in _run_real_model_generator(ask, image_bytes):
-                    # Standard SSE format for frontend parsing
                     yield f"data: {json.dumps({'text': text_chunk})}\n\n"
                 yield "data: [DONE]\n\n"
             except Exception as e:
@@ -220,7 +220,6 @@ def gemma():
                 
         return Response(generate_stream(), mimetype="text/event-stream")
     else:
-        # Wait for all chunks and join them into one string
         try:
             full_text = "".join(list(_run_real_model_generator(ask, image_bytes)))
             return jsonify({