Update app.py

app.py

@@ -59,8 +59,8 @@ llm = Llama.from_pretrained(
     repo_id="amir22010/fine_tuned_product_marketing_email_gemma_2_9b_q4_k_m", #custom fine tuned model
     filename="unsloth.Q4_K_M.gguf", #model file name
     cache_dir=os.path.abspath(os.getcwd()),
-    n_ctx=2048, 
-    n_batch=126,
+    # n_ctx=2048, 
+    # n_batch=126,
     verbose=False
 )
 
@@ -109,6 +109,7 @@ async def greet(product,description):
         partial_message = ""
         audio_list = []
         for chunk in output:
+            print(chunk)
             delta = chunk['choices'][0]['delta']
             if 'content' in delta:
                 # audio_list.append([text_to_speech(delta.get('content', ''))])