Update app.py

app.py

@@ -133,7 +133,8 @@ if not IS_MOCK_MODE:
         fury_model = AutoModelForCausalLM.from_pretrained(
             FURY_MODEL_ID,
             device_map="auto",
-            torch_dtype=torch.bfloat16,
+            # GPT-OSS provides its own MXFP4 quantization and dtype metadata.
+            torch_dtype="auto",
             trust_remote_code=True
         )
 
@@ -272,6 +273,7 @@ def generate_response_real(history_msgs, engine_choice):
     if "Fury" in engine_choice:
 
         print("🔥 FURY ROUTE SELECTED")
+        yield "🔥 **RiShre Fury is processing...**\n\nLoading the GPT-OSS reasoning core and preparing its response."
 
         # ----------------------------------------------------
         # NEVER FALL BACK TO FLASH
@@ -414,6 +416,10 @@ def generate_response_real(history_msgs, engine_choice):
 
             "max_new_tokens": 1024,
 
+            "use_cache": True,
+
+            "pad_token_id": fury_tokenizer.eos_token_id,
+
             "temperature": 0.4,
 
             "top_p": 0.9,