Update app.py

app.py

@@ -11,8 +11,8 @@ class XylariaChat:
 
         # Initialize the inference client
         self.client = InferenceClient(
-            model="Qwen/QwQ-32B-Preview",
-            api_key=self.hf_token
+            model="Qwen/QwQ-32B-Preview",  # Changed model name to a placeholder
+            token=self.hf_token
         )
 
         # Initialize conversation history and persistent memory
@@ -74,19 +74,34 @@ Capabilities:
 
         # Generate response with streaming
         try:
-            stream = self.client.chat.completions.create(
-                messages=messages,
+            response_stream = self.client.text_generation(
+                prompt=self.messages_to_prompt(messages),  # Convert messages to prompt format
+                max_new_tokens=1024,
                 temperature=0.5,
-                max_tokens=10240,
                 top_p=0.7,
                 stream=True
             )
 
-            return stream
+            return response_stream
 
         except Exception as e:
             return f"Error generating response: {str(e)}"
 
+    def messages_to_prompt(self, messages):
+        """
+        Converts a list of messages in OpenAI format to a prompt string.
+        """
+        prompt = ""
+        for message in messages:
+            if message["role"] == "system":
+                prompt += f"<|im_start|>system\n{message['content']}<|im_end|>\n"
+            elif message["role"] == "user":
+                prompt += f"<|im_start|>user\n{message['content']}<|im_end|>\n"
+            elif message["role"] == "assistant":
+                prompt += f"<|im_start|>assistant\n{message['content']}<|im_end|>\n"
+        prompt += "<|im_start|>assistant\n"
+        return prompt
+
     def create_interface(self):
         # Local storage JavaScript functions
         local_storage_js = """
@@ -117,14 +132,12 @@ Capabilities:
             updated_history = chat_history + [[message, ""]]
 
             # Streaming output
-            for chunk in response_stream:
-                if chunk.choices[0].delta.content:
-                    chunk_content = chunk.choices[0].delta.content
-                    full_response += chunk_content
+            for response_text in response_stream:
+                full_response += response_text
 
-                    # Update the last message in chat history with partial response
-                    updated_history[-1][1] = full_response
-                    yield "", updated_history
+                # Update the last message in chat history with partial response
+                updated_history[-1][1] = full_response
+                yield "", updated_history
 
             # Update conversation history
             self.conversation_history.append(
@@ -166,7 +179,7 @@ Capabilities:
                     label="Xylaria 1.4 Senoa",
                     height=500,
                     show_copy_button=True,
-                    # value=lambda: gr.load("json", "loadFromLocalStorage()")  # Remove this line
+                    type="messages"  # Use the 'messages' format
                 )
 
                 # Input row with improved layout
@@ -185,11 +198,11 @@ Capabilities:
 
             # Load chat history from local storage on page load
             demo.load(
-                fn=None,  # No Python function needed, just JavaScript
-                _js="() => loadFromLocalStorage()"  # Call your JavaScript function
+                fn=None,
+                _js="() => loadFromLocalStorage()"
             ).then(
                 fn=lambda x: x,
-                inputs=chatbot,  # Use chatbot as input to get the loaded value
+                inputs=chatbot,
                 outputs=chatbot
             )
 
@@ -199,7 +212,9 @@ Capabilities:
                 inputs=[txt, chatbot],
                 outputs=[txt, chatbot]
             ).then(
-                fn=None,  # JavaScript callback
+                fn=None,
+                inputs=[chatbot],  # Pass chatbot history to JavaScript
+                outputs=None,
                 _js='(chatHistory) => saveToLocalStorage(chatHistory)'
             )
             txt.submit(
@@ -207,7 +222,9 @@ Capabilities:
                 inputs=[txt, chatbot],
                 outputs=[txt, chatbot]
             ).then(
-                fn=None,  # JavaScript callback
+                fn=None,
+                inputs=[chatbot],  # Pass chatbot history to JavaScript
+                outputs=None,
                 _js='(chatHistory) => saveToLocalStorage(chatHistory)'
             )
 
@@ -217,17 +234,17 @@ Capabilities:
                 inputs=None,
                 outputs=[chatbot]
             ).then(
-                fn=None,  # JavaScript callback
+                fn=None,
                 _js='() => clearLocalStorage()'
             )
 
             # Clear persistent memory and reset conversation with local storage clear
             clear_memory.click(
-                fn=lambda: [],
+                fn=self.reset_conversation,
                 inputs=None,
                 outputs=[chatbot]
             ).then(
-                fn=None,  # JavaScript callback
+                fn=None,
                 _js='() => clearLocalStorage()'
             )