ai: Allow reasoning/non on audio/image generation.

* Keep using non-reasoning mode if the response fails.

src/client/chat_handler.py

@@ -77,7 +77,13 @@ async def respond(
             new_history,  # Conversation history
             session_id,  # Session ID
             selected_model,  # Selected model
-            jarvis  # AI backend function
+            jarvis,  # AI backend function
+            mode,  # Mode for AI response
+            temperature,  # temperature parameter
+            top_k,  # top_k parameter
+            min_p,  # min_p parameter
+            top_p,  # top_p parameter
+            repetition_penalty  # repetition_penalty parameter
         ):
             yield audio_response  # Yield audio response
         return  # Exit function after handling audio
@@ -89,7 +95,13 @@ async def respond(
             new_history,  # Conversation history
             session_id,  # Session ID
             selected_model,  # Selected model
-            jarvis  # AI backend function
+            jarvis,  # AI backend function
+            mode,  # Mode for AI response
+            temperature,  # temperature parameter
+            top_k,  # top_k parameter
+            min_p,  # min_p parameter
+            top_p,  # top_p parameter
+            repetition_penalty  # repetition_penalty parameter
         ):
             yield image_response  # Yield image response
         return  # Exit function after handling image

src/client/responses/audio.py

@@ -12,7 +12,13 @@ async def audio_integration(
     new_history,  # Conversation history in message format
     session_id,  # Session ID for conversation context
     selected_model,  # Selected AI model for generation
-    jarvis  # AI backend function for generating responses
+    jarvis,  # AI backend function for generating responses
+    mode,  # Mode for AI response generation
+    temperature,  # Temperature parameter for AI
+    top_k,  # Top-k parameter for AI
+    min_p,  # Min-p parameter for AI
+    top_p,  # Top-p parameter for AI
+    repetition_penalty  # Repetition penalty for AI
 ):
     # Extract the audio instruction text after the '/audio' command prefix and strip whitespace
     audio_instruction = input[6:].strip()  # Get instruction after /audio
@@ -58,12 +64,12 @@ async def audio_integration(
             model=selected_model,  # Selected model
             history=audio_generation_result,  # Updated history with audio result
             user_message=input,  # User input
-            mode="/no_think",  # Use non-reasoning mode to avoid extra processing overhead
-            temperature=0.7,  # Fixed temperature for consistent audio description generation
-            top_k=20,  # Limit token sampling to top 20 most probable tokens
-            min_p=0,  # Minimum probability threshold set to zero
-            top_p=0.8,  # Nucleus sampling threshold for quality control
-            repetition_penalty=1  # No repetition penalty for this step
+            mode=mode,  # Mode for AI response
+            temperature=temperature,  # temperature parameter
+            top_k=top_k,  # top_k parameter
+            min_p=min_p,  # min_p parameter
+            top_p=top_p,  # top_p parameter
+            repetition_penalty=repetition_penalty  # repetition_penalty parameter
         ):
             yield [{"role": "tool", "content": audio_description}]  # Yield audio description in tool role
         return  # Exit after handling audio

src/client/responses/image.py

@@ -12,7 +12,13 @@ async def image_integration(
     new_history,  # Conversation history in message format
     session_id,  # Session ID for conversation context
     selected_model,  # Selected AI model for generation
-    jarvis  # AI backend function for generating responses
+    jarvis,  # AI backend function for generating responses
+    mode,  # Mode for AI response generation
+    temperature,  # Temperature parameter for AI
+    top_k,  # Top-k parameter for AI
+    min_p,  # Min-p parameter for AI
+    top_p,  # Top-p parameter for AI
+    repetition_penalty  # Repetition penalty for AI
 ):
     # Extract the image generation instruction after the '/image' command prefix and strip whitespace
     generate_image_instruction = input[6:].strip()  # Get instruction after /image
@@ -58,12 +64,12 @@ async def image_integration(
             model=selected_model,  # Selected model
             history=image_generation_result,  # Updated history with image result
             user_message=input,  # User input
-            mode="/no_think",  # Use non-reasoning mode to avoid extra processing overhead
-            temperature=0.7,  # Fixed temperature for consistent image description generation
-            top_k=20,  # Limit token sampling to top 20 most probable tokens
-            min_p=0,  # Minimum probability threshold set to zero
-            top_p=0.8,  # Nucleus sampling threshold for quality control
-            repetition_penalty=1  # No repetition penalty for this step
+            mode=mode,  # Mode for AI response
+            temperature=temperature,  # temperature parameter
+            top_k=top_k,  # top_k parameter
+            min_p=min_p,  # min_p parameter
+            top_p=top_p,  # top_p parameter
+            repetition_penalty=repetition_penalty  # repetition_penalty parameter
         ):
             yield [{"role": "tool", "content": image_description}]  # Yield image description in tool role
         return  # Exit after handling image