Assign stream_task cpu only

pipeline.py

@@ -242,7 +242,7 @@ def _handle_clinical_answer(session_id: str, answer_text: str):
     return {"type": "question", "prompt": prompt}
 
 
-@spaces.GPU(max_duration=120)
+# @spaces.GPU(max_duration=120)
 def stream_chat(
     message: str,
     history: list,

requirements.txt

@@ -12,6 +12,7 @@ sentencepiece>=0.1.99
 google-genai
 langdetect
 gradio
+gradio[mcp]
 # MCP dependencies (required for Gemini MCP)
 # Install MCP SDK: pip install mcp
 # The MCP package provides Model Context Protocol server and client functionality

ui.py

@@ -140,24 +140,24 @@ def create_demo():
                             label="Disable agentic reasoning",
                             info="Use MedSwin model alone without agentic reasoning, RAG, or web search"
                         )
-                        show_agentic_thought = gr.Button(
-                            "Show agentic thought",
-                            size="sm"
-                        )
+                        # show_agentic_thought = gr.Button(
+                        #     "Show agentic thought",
+                        #     size="sm"
+                        # )
                     enable_clinical_intake = gr.Checkbox(
                         value=True,
                         label="Enable clinical intake (max 5 Q&A)",
                         info="Ask focused follow-up questions before breaking down the case"
                     )
-                    agentic_thoughts_box = gr.Textbox(
-                        label="Agentic Thoughts",
-                        placeholder="Internal thoughts from MedSwin and supervisor will appear here...",
-                        lines=8,
-                        max_lines=15,
-                        interactive=False,
-                        visible=False,
-                        elem_classes="agentic-thoughts"
-                    )
+                    # agentic_thoughts_box = gr.Textbox(
+                    #     label="Agentic Thoughts",
+                    #     placeholder="Internal thoughts from MedSwin and supervisor will appear here...",
+                    #     lines=8,
+                    #     max_lines=15,
+                    #     interactive=False,
+                    #     visible=False,
+                    #     elem_classes="agentic-thoughts"
+                    # )
                     with gr.Row():
                         use_rag = gr.Checkbox(
                             value=False,