Update Gradio_UI.py

Gradio_UI.py

@@ -36,14 +36,13 @@ def pull_messages_from_step_dict(step_log: MemoryStep):
 
         if hasattr(step_log, "model_output") and step_log.model_output is not None:
             model_output = step_log.model_output.strip()
-            # More robust cleaning for <end_code> potentially wrapped in backticks or with newlines
             model_output = re.sub(r"```\s*<end_code>[\s\S]*|[\s\S]*<end_code>\s*```", "```", model_output, flags=re.DOTALL)
-            model_output = re.sub(r"<end_code>", "", model_output) # Remove standalone tag
+            model_output = re.sub(r"<end_code>", "", model_output) 
             model_output = model_output.strip()
             yield {"role": "assistant", "content": model_output}
 
         if hasattr(step_log, "tool_calls") and step_log.tool_calls:
-            tc = step_log.tool_calls[0] # Process first tool call for simplicity in this format
+            tc = step_log.tool_calls[0] 
             tool_info_md = f"🛠️ **Tool Used: {tc.name}**\n"
             
             args = tc.arguments
@@ -54,7 +53,6 @@ def pull_messages_from_step_dict(step_log: MemoryStep):
             
             if tc.name == "python_interpreter":
                 code_content = args_str
-                # Clean up common wrapping issues
                 code_content = re.sub(r"^```python\s*\n?", "", code_content)
                 code_content = re.sub(r"\n?```\s*$", "", code_content)
                 code_content = re.sub(r"^\s*<end_code>\s*", "", code_content)
@@ -66,34 +64,32 @@ def pull_messages_from_step_dict(step_log: MemoryStep):
 
             if hasattr(step_log, "observations") and step_log.observations and step_log.observations.strip():
                 obs_content = step_log.observations.strip()
-                # Remove "Execution logs:" prefix if present for cleaner display
                 obs_content = re.sub(r"^Execution logs:\s*", "", obs_content).strip()
-                if obs_content: # Only show if there's something after stripping
-                    tool_info_md += f"📝 **Tool Output/Logs:**\n```\n{obs_content}\n```\n"
+                if obs_content: 
+                    tool_info_md += f"📝 **Tool Output/Logs:**\n```text\n{obs_content}\n```\n" # Use text for generic logs
             
             if hasattr(step_log, "error") and step_log.error:
                 tool_info_md += f"💥 **Error:** {str(step_log.error)}\n"
             
             yield {"role": "assistant", "content": tool_info_md.strip()}
 
-        elif hasattr(step_log, "error") and step_log.error: # Standalone error not from a tool call
+        elif hasattr(step_log, "error") and step_log.error: 
             yield {"role": "assistant", "content": f"💥 **Error:** {str(step_log.error)}"}
         
-        # --- Minimal footnote for type="messages" ---
         footnote_parts = []
         if step_log.step_number is not None:
             footnote_parts.append(f"Step {step_log.step_number}")
         if hasattr(step_log, "duration") and step_log.duration is not None:
             footnote_parts.append(f"Duration: {round(float(step_log.duration), 2)}s")
-        if hasattr(step_log, "input_token_count") and step_log.input_token_count is not None: # Check for None
+        if hasattr(step_log, "input_token_count") and step_log.input_token_count is not None:
              footnote_parts.append(f"InTokens: {step_log.input_token_count:,}")
-        if hasattr(step_log, "output_token_count") and step_log.output_token_count is not None: # Check for None
+        if hasattr(step_log, "output_token_count") and step_log.output_token_count is not None:
              footnote_parts.append(f"OutTokens: {step_log.output_token_count:,}")
         
         if footnote_parts:
             footnote_text = " | ".join(footnote_parts)
             yield {"role": "assistant", "content": f"""<p style="color: #999; font-size: 0.8em; margin-top:0; margin-bottom:0;">{footnote_text}</p>"""}
-        yield {"role": "assistant", "content": "---"} # Separator
+        yield {"role": "assistant", "content": "---"} 
 
 
 def stream_to_gradio(
@@ -102,34 +98,41 @@ def stream_to_gradio(
     reset_agent_memory: bool = False,
     additional_args: Optional[dict] = None,
 ):
-    """Runs an agent, yields message dicts for Gradio type='messages' Chatbot."""
     if not _is_package_available("gradio"):
         raise ModuleNotFoundError("Install 'gradio': `pip install 'smolagents[gradio]'`")
 
-    if hasattr(agent, 'interaction_logs'): # Clear logs for this new agent run
+    if hasattr(agent, 'interaction_logs'): 
         agent.interaction_logs.clear()
         print("DEBUG Gradio: Cleared agent interaction_logs for new run.")
 
+    # This will collect all step_log objects from the agent run
+    all_step_logs = []
     for step_log in agent.run(task, stream=True, reset=reset_agent_memory, additional_args=additional_args):
+        all_step_logs.append(step_log) # Store the log
         if hasattr(agent.model, "last_input_token_count") and agent.model.last_input_token_count is not None:
             if isinstance(step_log, ActionStep):
                 step_log.input_token_count = agent.model.last_input_token_count
                 step_log.output_token_count = agent.model.last_output_token_count
         
-        for msg_dict in pull_messages_from_step_dict(step_log): # Use new dict-yielding function
+        for msg_dict in pull_messages_from_step_dict(step_log): 
             yield msg_dict
+    
+    # After the loop, the last item in all_step_logs is the final output/state from agent.run
+    if not all_step_logs: # Should not happen if agent.run yields at least one thing
+        yield {"role": "assistant", "content": "Agent did not produce any output."}
+        return
 
-    final_answer_content = step_log # Last step_log is the final output/state
+    final_answer_content = all_step_logs[-1] # This is what final_answer tool returns or the last ActionStep.final_answer
 
     # --- Handle final answer for type="messages" ---
     if isinstance(final_answer_content, PILImage.Image):
-        print("DEBUG Gradio (stream_to_gradio): Final answer is raw PIL Image.")
+        print("DEBUG Gradio (stream_to_gradio): Final answer content IS a raw PIL Image.")
         try:
+            # delete=False is crucial for Gradio to access the file before it's cleaned up
             with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp_file:
                 final_answer_content.save(tmp_file, format="PNG")
                 image_path_for_gradio = tmp_file.name
-            print(f"DEBUG Gradio: Saved PIL image to temp path: {image_path_for_gradio}")
-            # For Gradio type="messages", image content is just the path string
+            print(f"DEBUG Gradio: Saved PIL image to temp path for display: {image_path_for_gradio}")
             yield {"role": "assistant", "content": image_path_for_gradio} 
             return 
         except Exception as e:
@@ -137,17 +140,43 @@ def stream_to_gradio(
             yield {"role": "assistant", "content": f"**Final Answer (Error displaying image):** {e}"}
             return
 
-    final_answer_processed = handle_agent_output_types(final_answer_content)
+    # If not a raw PIL Image, then try smolagents processing from handle_agent_output_types
+    # The 'final_answer_content' here could be a FinalAnswerStep object or similar
+    # We need to extract the actual content from it if it's a wrapper.
+    actual_content_for_handling = final_answer_content
+    if hasattr(final_answer_content, 'final_answer') and not isinstance(final_answer_content, (str, PILImage.Image)):
+         actual_content_for_handling = final_answer_content.final_answer
+         print(f"DEBUG Gradio: Extracted actual_content_for_handling from FinalAnswerStep: {type(actual_content_for_handling)}")
+
+
+    # Re-check if the extracted content is a PIL Image
+    if isinstance(actual_content_for_handling, PILImage.Image):
+        print("DEBUG Gradio (stream_to_gradio): Extracted content IS a raw PIL Image.")
+        try:
+            with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as tmp_file:
+                actual_content_for_handling.save(tmp_file, format="PNG")
+                image_path_for_gradio = tmp_file.name
+            print(f"DEBUG Gradio: Saved extracted PIL image to temp path: {image_path_for_gradio}")
+            yield {"role": "assistant", "content": image_path_for_gradio} 
+            return
+        except Exception as e:
+            print(f"DEBUG Gradio: Error saving extracted PIL image: {e}")
+            yield {"role": "assistant", "content": f"**Final Answer (Error displaying image from extracted content):** {e}"}
+            return
+
+    final_answer_processed = handle_agent_output_types(actual_content_for_handling)
+    print(f"DEBUG Gradio: final_answer_processed type after handle_agent_output_types: {type(final_answer_processed)}")
+
 
     if isinstance(final_answer_processed, AgentText):
         yield {"role": "assistant", "content": f"**Final Answer:**\n{final_answer_processed.to_string()}"}
     elif isinstance(final_answer_processed, AgentImage):
         image_path = final_answer_processed.to_string()
-        print(f"DEBUG Gradio (stream_to_gradio): AgentImage path: {image_path}")
+        print(f"DEBUG Gradio (stream_to_gradio): final_answer_processed is AgentImage. Path: {image_path}")
         if image_path and os.path.exists(image_path):
              yield {"role": "assistant", "content": image_path} 
         else:
-            err_msg = f"Error: Image path from AgentImage not found or invalid ('{image_path}')"
+            err_msg = f"Error: Image path from AgentImage ('{image_path}') not found or invalid after smolagents processing."
             print(f"DEBUG Gradio: {err_msg}")
             yield {"role": "assistant", "content": f"**Final Answer ({err_msg})**"}
     elif isinstance(final_answer_processed, AgentAudio):
@@ -156,16 +185,15 @@ def stream_to_gradio(
         if audio_path and os.path.exists(audio_path):
             yield {"role": "assistant", "content": audio_path}
         else:
-            err_msg = f"Error: Audio path from AgentAudio not found ('{audio_path}')"
+            err_msg = f"Error: Audio path from AgentAudio ('{audio_path}') not found"
             print(f"DEBUG Gradio: {err_msg}")
             yield {"role": "assistant", "content": f"**Final Answer ({err_msg})**"}
     else: 
+        # This will display the string representation of FinalAnswerStep if not handled above
         yield {"role": "assistant", "content": f"**Final Answer:**\n{str(final_answer_processed)}"}
 
 
 class GradioUI:
-    """A one-line interface to launch your agent in Gradio"""
-
     def __init__(self, agent: MultiStepAgent, file_upload_folder: str | None = None):
         if not _is_package_available("gradio"):
             raise ModuleNotFoundError("Install 'gradio': `pip install 'smolagents[gradio]'`")
@@ -180,82 +208,65 @@ class GradioUI:
         self._latest_file_path_for_download = None 
         if hasattr(self.agent, 'interaction_logs') and self.agent.interaction_logs:
             print(f"DEBUG Gradio UI: Checking {len(self.agent.interaction_logs)} interaction log entries for created files.")
-            for log_entry in reversed(self.agent.interaction_logs): # Check recent logs first
-                if isinstance(log_entry, ActionStep) and hasattr(log_entry, 'tool_calls') and log_entry.tool_calls:
-                    for tool_call in log_entry.tool_calls:
-                        if tool_call.name == "create_document":
-                            tool_output_value = getattr(log_entry, 'observations', None)
-                            print(f"DEBUG Gradio UI: Log for 'create_document' call, observed output: {tool_output_value}")
-                            if tool_output_value and isinstance(tool_output_value, str):
-                                # Try to extract path if it's wrapped, e.g. by "Execution logs:"
-                                cleaned_output = re.sub(r"^Execution logs:\s*", "", tool_output_value).strip()
-                                path_match = re.search(r"(/tmp/[a-zA-Z0-9_]+/generated_document\.(?:docx|pdf|txt))", cleaned_output)
-                                extracted_path = path_match.group(1) if path_match else cleaned_output
-
-                                if not extracted_path.lower().startswith("error:"):
-                                    normalized_path = os.path.normpath(extracted_path)
-                                    if os.path.exists(normalized_path):
-                                        self._latest_file_path_for_download = normalized_path
-                                        print(f"DEBUG Gradio UI: File path for download set: {self._latest_file_path_for_download}")
-                                        return True 
-                                    else:
-                                        print(f"DEBUG Gradio UI: Path from 'create_document' log ('{normalized_path}') does not exist.")
-                                else:
-                                    print(f"DEBUG Gradio UI: 'create_document' tool reported error in observations: {extracted_path}")
-        print("DEBUG Gradio UI: No valid 'create_document' output found for download.")
+            for log_entry in reversed(self.agent.interaction_logs): 
+                if isinstance(log_entry, ActionStep):
+                    observations = getattr(log_entry, 'observations', None)
+                    tool_calls = getattr(log_entry, 'tool_calls', [])
+
+                    # Check if python_interpreter was used AND its code involved create_document
+                    # For simplicity, we'll primarily rely on parsing observations for the path pattern
+                    if observations and isinstance(observations, str):
+                        # This regex should match paths printed by your create_document tool
+                        path_match = re.search(r"(/tmp/[a-zA-Z0-9_]+/generated_document\.(?:docx|pdf|txt))", observations)
+                        if path_match:
+                            extracted_path = path_match.group(1)
+                            normalized_path = os.path.normpath(extracted_path)
+                            if os.path.exists(normalized_path):
+                                self._latest_file_path_for_download = normalized_path
+                                print(f"DEBUG Gradio UI: File path for download set (from observations): {self._latest_file_path_for_download}")
+                                return True 
+                            else:
+                                print(f"DEBUG Gradio UI: Path from observations ('{normalized_path}') does not exist.")
+        print("DEBUG Gradio UI: No valid generated file path found in agent logs for download.")
         return False
 
-    def interact_with_agent(self, prompt_text: str, current_chat_tuples: list):
-        # current_chat_tuples is the history from the chatbot (list of lists/tuples)
-        # Convert to 'messages' format if needed, or adapt stream_to_gradio if chatbot is not type="messages"
-        # For type="messages", current_chat_tuples is already list of dicts.
-        
+    def interact_with_agent(self, prompt_text: str, current_chat_history: list):
         print(f"DEBUG Gradio: interact_with_agent called with prompt: '{prompt_text}'")
-        print(f"DEBUG Gradio: Current chat history (input): {current_chat_tuples}")
+        print(f"DEBUG Gradio: Current chat history (input type {type(current_chat_history)}): {current_chat_history}")
 
-        # Add user's new message to the chat history list
-        current_chat_messages = current_chat_tuples + [{"role": "user", "content": prompt_text}]
+        # current_chat_history from gr.Chatbot(type="messages") is already a list of dicts
+        updated_chat_history = current_chat_history + [{"role": "user", "content": prompt_text}]
         
-        # Initial yield to show user message immediately and hide download items
-        yield current_chat_messages, gr.update(visible=False), gr.update(value=None, visible=False)
+        yield updated_chat_history, gr.update(visible=False), gr.update(value=None, visible=False)
 
-        # Stream agent messages
         agent_responses_for_history = []
         for msg_dict in stream_to_gradio(self.agent, task=prompt_text, reset_agent_memory=False):
             agent_responses_for_history.append(msg_dict)
-            # Yield progressively: current user message + all agent messages so far
-            yield current_chat_messages + agent_responses_for_history, gr.update(visible=False), gr.update(value=None, visible=False) 
+            yield updated_chat_history + agent_responses_for_history, gr.update(visible=False), gr.update(value=None, visible=False) 
 
-        # After streaming all agent messages, check for created file
         file_found = self._check_for_created_file()
         
-        # Final state for UI components
-        final_chat_display = current_chat_messages + agent_responses_for_history
-        print(f"DEBUG Gradio: Final chat history for display: {final_chat_display}")
+        final_chat_display = updated_chat_history + agent_responses_for_history
+        print(f"DEBUG Gradio: Final chat history for display: {len(final_chat_display)} messages.")
         yield final_chat_display, gr.update(visible=file_found), gr.update(value=None, visible=False)
 
-
     def upload_file(self, file, file_uploads_log_state):
-        if file is None: # No file selected
+        if file is None: 
             return gr.update(value="No file uploaded.", visible=True), file_uploads_log_state
 
-        # Ensure file_upload_folder exists (it should from __init__)
         if not self.file_upload_folder or not os.path.exists(self.file_upload_folder):
-            os.makedirs(self.file_upload_folder, exist_ok=True) # Defensive check
+            os.makedirs(self.file_upload_folder, exist_ok=True)
 
         allowed_file_types = [
             "application/pdf",
             "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
-            "text/plain",
-            "image/jpeg", "image/png", # Added image types
+            "text/plain", "image/jpeg", "image/png",
         ]
         
-        # Gradio File object has 'name' (temp path) and 'orig_name'
-        original_name = file.orig_name if hasattr(file, 'orig_name') else os.path.basename(file.name)
+        original_name = file.orig_name if hasattr(file, 'orig_name') and file.orig_name else os.path.basename(file.name)
         
-        # Try to guess mime type from temp file name first, then from original name if needed
         mime_type, _ = mimetypes.guess_type(file.name) 
-        if mime_type is None: # Fallback
+        if mime_type is None: 
             mime_type, _ = mimetypes.guess_type(original_name)
 
         if mime_type not in allowed_file_types:
@@ -264,13 +275,13 @@ class GradioUI:
         sanitized_name = re.sub(r"[^\w\-.]", "_", original_name)
         base_name, current_ext = os.path.splitext(sanitized_name)
         
-        type_to_ext_map = {v: k for k, v_list in mimetypes. প্রেফারেন্সেস.items() for v in v_list} # More robust ext map
-        type_to_ext_map.update({ # Manual overrides / common types
+        # Updated mimetypes to extension mapping
+        common_mime_to_ext = {
             "application/pdf": ".pdf",
             "application/vnd.openxmlformats-officedocument.wordprocessingml.document": ".docx",
             "text/plain": ".txt", "image/jpeg": ".jpg", "image/png": ".png"
-        })
-        expected_ext = type_to_ext_map.get(mime_type)
+        }
+        expected_ext = common_mime_to_ext.get(mime_type)
 
         if expected_ext and current_ext.lower() != expected_ext.lower():
             sanitized_name = base_name + expected_ext
@@ -278,22 +289,21 @@ class GradioUI:
         destination_path = os.path.join(self.file_upload_folder, sanitized_name)
         
         try:
-            shutil.copy(file.name, destination_path) # file.name is the temp path from Gradio
+            shutil.copy(file.name, destination_path) 
             print(f"DEBUG Gradio: File '{original_name}' copied to '{destination_path}'")
             updated_log = file_uploads_log_state + [destination_path]
-            return gr.update(value=f"Uploaded: {original_name} (as {sanitized_name})", visible=True), updated_log
+            return gr.update(value=f"Uploaded: {original_name}", visible=True), updated_log
         except Exception as e:
             print(f"DEBUG Gradio: Error copying uploaded file: {e}")
             return gr.update(value=f"Error uploading {original_name}: {e}", visible=True), file_uploads_log_state
 
-
     def log_user_message(self, text_input_value: str, current_file_uploads: list):
         full_prompt = text_input_value
         if current_file_uploads:
             files_str = ", ".join([os.path.basename(f) for f in current_file_uploads])
             full_prompt += f"\n\n[Uploaded files for context: {files_str}]"
-        print(f"DEBUG Gradio: Prepared prompt for agent: {full_prompt}")
-        return full_prompt, "" # Clears the text input box
+        print(f"DEBUG Gradio: Prepared prompt for agent: {full_prompt[:300]}...") # Log snippet
+        return full_prompt, "" 
 
     def prepare_and_show_download_file(self):
         if self._latest_file_path_for_download and os.path.exists(self._latest_file_path_for_download):
@@ -303,56 +313,57 @@ class GradioUI:
                                   visible=True)
         else:
             print("DEBUG Gradio UI: No valid file path to prepare for download component.")
-            gr.Warning("No file available for download or path is invalid.")
-            return gr.File.update(visible=False)
+            # gr.Warning("No file available for download or path is invalid.") # Causes JS error if used as return
+            return gr.File.update(visible=False, value=None) # Ensure value is None if not visible
 
     def launch(self, **kwargs):
         with gr.Blocks(fill_height=True, theme=gr.themes.Soft(primary_hue=gr.themes.colors.blue)) as demo:
             file_uploads_log_state = gr.State([]) 
             prepared_prompt_for_agent = gr.State("")
 
-            gr.Markdown("#  agente inteligente")
+            gr.Markdown("## Smol Talk with your Agent") # Changed title slightly
 
-            with gr.Row():
+            with gr.Row(equal_height=False): # Allow columns to size independently
                 with gr.Column(scale=3):
                     chatbot_display = gr.Chatbot(
                         label="Agent Interaction",
                         type="messages", 
                         avatar_images=(None, "https://huggingface.co/datasets/huggingface/brand-assets/resolve/main/hf-logo-round.png"),
-                        height=600,
+                        height=700, # Increased height
                         show_copy_button=True,
-                        bubble_full_width=False
+                        bubble_full_width=False,
+                        show_label=False # Hide the "Agent Interaction" label above chatbot
                     )
                     text_message_input = gr.Textbox(
                         lines=1, 
                         label="Your Message to the Agent", 
-                        placeholder="Type your message and press Enter, or Shift+Enter for new line..."
+                        placeholder="Type your message and press Enter, or Shift+Enter for new line...",
+                        show_label=False # Hide label for text input
                     )
                 
                 with gr.Column(scale=1):
                     if self.file_upload_folder is not None:
-                        gr.Markdown("### File Upload")
-                        file_uploader = gr.File(label="Upload a supporting file (PDF, DOCX, TXT, JPG, PNG)")
-                        upload_status_text = gr.Textbox(label="Upload Status", interactive=False, lines=2, max_lines=4)
-                        file_uploader.upload(
-                            self.upload_file,
-                            [file_uploader, file_uploads_log_state],
-                            [upload_status_text, file_uploads_log_state],
-                        )
+                        with gr.Accordion("File Upload", open=False): # Collapsible section
+                            file_uploader = gr.File(label="Upload a supporting file (PDF, DOCX, TXT, JPG, PNG)")
+                            upload_status_text = gr.Textbox(label="Upload Status", interactive=False, lines=1) # single line
+                            file_uploader.upload( # Changed from .change to .upload for gr.File
+                                self.upload_file,
+                                [file_uploader, file_uploads_log_state],
+                                [upload_status_text, file_uploads_log_state],
+                            )
                     
-                    gr.Markdown("### Generated File")
-                    download_action_button = gr.Button("Download Generated File", visible=False) 
-                    file_download_display_component = gr.File(label="Downloadable Document", visible=False, interactive=False) 
+                    with gr.Accordion("Generated File", open=True): # Collapsible, open by default
+                        download_action_button = gr.Button("Download Generated File", visible=False) 
+                        file_download_display_component = gr.File(label="Downloadable Document", visible=False, interactive=False) 
 
-            # Event Handling Chain for Text Submission
             text_message_input.submit(
-                self.log_user_message, # Step 1: Prepare prompt, clear input
+                self.log_user_message, 
                 [text_message_input, file_uploads_log_state],
                 [prepared_prompt_for_agent, text_message_input] 
             ).then(
-                self.interact_with_agent, # Step 2: Run agent, stream to chatbot, update download button
-                [prepared_prompt_for_agent, chatbot_display], 
-                [chatbot_display, download_action_button, file_download_display_component] 
+                self.interact_with_agent, 
+                [prepared_prompt_for_agent, chatbot_display], # chatbot_display is input here
+                [chatbot_display, download_action_button, file_download_display_component] # chatbot_display is output here
             )
 
             download_action_button.click(
@@ -360,7 +371,7 @@ class GradioUI:
                 [], 
                 [file_download_display_component] 
             )
-
+        # Default share=False, can be overridden by kwargs
         demo.launch(debug=True, share=kwargs.get("share", False), **kwargs)
 
 __all__ = ["stream_to_gradio", "GradioUI"]