Update app.py

app.py

@@ -22,6 +22,7 @@ MODEL_SUBFOLDER = os.getenv('MODEL_SUBFOLDER', '')
 MODEL_IDENTITY = os.getenv('MODEL_IDENTITY', '')
 DEFAULT_SYSTEM_PROMPT = os.getenv('SYSTEM_MESSAGE', MODEL_IDENTITY or 'You are a helpful assistant. Reasoning: medium')
 DEFAULT_DEVELOPER_PROMPT = os.getenv('DEVELOPER_MESSAGE', '')
+DEFAULT_REASONING_EFFORT = os.getenv('REASONING_EFFORT', 'medium')
 
 # If the LORA_MODEL_ID is the same as BASE_MODEL_ID, this is a merged model, not LoRA
 USE_LORA = LORA_MODEL_ID != BASE_MODEL_ID and not LORA_MODEL_ID.startswith(BASE_MODEL_ID)
@@ -135,7 +136,7 @@ def format_analysis_response(text):
     return cleaned
 
 @spaces.GPU(duration=60)
-def generate_response(input_data, chat_history, max_new_tokens, model_identity, system_prompt, developer_prompt, temperature, top_p, top_k, repetition_penalty):
+def generate_response(input_data, chat_history, max_new_tokens, model_identity, system_prompt, developer_prompt, reasoning_effort, temperature, top_p, top_k, repetition_penalty):
     if not input_data.strip():
         yield "Please enter a prompt."
         return
@@ -146,10 +147,15 @@ def generate_response(input_data, chat_history, max_new_tokens, model_identity,
     
     new_message = {"role": "user", "content": input_data}
     # Combine model identity with system prompt for a single system message
-    combined_system = "".join([
-        (model_identity.strip() + "\n\n") if model_identity and model_identity.strip() else "",
-        system_prompt.strip() if system_prompt else ""
-    ]).strip()
+    combined_parts = []
+    if model_identity and model_identity.strip():
+        combined_parts.append(model_identity.strip())
+    if system_prompt and system_prompt.strip():
+        combined_parts.append(system_prompt.strip())
+    if reasoning_effort and isinstance(reasoning_effort, str) and reasoning_effort.strip():
+        # Append explicit reasoning directive
+        combined_parts.append(f"Reasoning: {reasoning_effort.strip()}")
+    combined_system = "\n\n".join(combined_parts).strip()
     system_message = ([{"role": "system", "content": combined_system}] if combined_system else [])
     developer_message = [{"role": "developer", "content": developer_prompt}] if developer_prompt else []
     processed_history = format_conversation_history(chat_history)
@@ -252,6 +258,12 @@ demo = gr.ChatInterface(
             lines=4,
             placeholder="Optional developer instructions"
         ),
+        gr.Dropdown(
+            label="Reasoning Effort",
+            choices=["low", "medium", "high"],
+            value=DEFAULT_REASONING_EFFORT,
+            interactive=True,
+        ),
         gr.Slider(label="Temperature", minimum=0.1, maximum=2.0, step=0.1, value=0.7),
         gr.Slider(label="Top-p", minimum=0.05, maximum=1.0, step=0.05, value=0.9),
         gr.Slider(label="Top-k", minimum=1, maximum=100, step=1, value=50),