Gradio_UI.py

#!/usr/bin/env python
# coding=utf-8
# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import mimetypes
import os
import re
import shutil
from typing import Optional
import tempfile 
from PIL import Image as PILImage 

from smolagents.agent_types import AgentAudio, AgentImage, AgentText, handle_agent_output_types
from smolagents.agents import ActionStep, MultiStepAgent 
from smolagents.memory import MemoryStep
from smolagents.utils import _is_package_available
import gradio as gr 


def pull_messages_from_step_dict(step_log: MemoryStep):
    """Extract messages as dicts for Gradio type='messages' Chatbot"""
    if isinstance(step_log, ActionStep):
        step_number_str = f"Step {step_log.step_number}" if step_log.step_number is not None else "Processing"
        yield {"role": "assistant", "content": f"**{step_number_str}**"}

        if hasattr(step_log, "model_output") and step_log.model_output is not None:
            model_output = step_log.model_output.strip()
            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) 
            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] 
            tool_info_md = f"🛠️ **Tool Used: {tc.name}**\n"
            
            args = tc.arguments
            if isinstance(args, dict):
                args_str = str(args.get("answer", str(args)))
            else:
                args_str = str(args).strip()
            
            if tc.name == "python_interpreter":
                code_content = args_str
                code_content = re.sub(r"^```python\s*\n?", "", code_content)
                code_content = re.sub(r"\n?```\s*$", "", code_content)
                code_content = re.sub(r"^\s*<end_code>\s*", "", code_content)
                code_content = re.sub(r"\s*<end_code>\s*$", "", code_content)
                code_content = code_content.strip()
                tool_info_md += f"Executing Code:\n```python\n{code_content}\n```\n"
            else:
                tool_info_md += f"Arguments: `{args_str}`\n"

            if hasattr(step_log, "observations") and step_log.observations and step_log.observations.strip():
                obs_content = step_log.observations.strip()
                obs_content = re.sub(r"^Execution logs:\s*", "", obs_content).strip()
                if obs_content: 
                    tool_info_md += f"📝 **Tool Output/Logs:**\n```text\n{obs_content}\n```\n"
            
            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: 
            yield {"role": "assistant", "content": f"💥 **Error:** {str(step_log.error)}"}
        
        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:
             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:
             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": "---"} 


def stream_to_gradio(
    agent,
    task: str,
    reset_agent_memory: bool = False,
    additional_args: Optional[dict] = None,
):
    if not _is_package_available("gradio"):
        raise ModuleNotFoundError("Install 'gradio': `pip install 'smolagents[gradio]'`")

    if hasattr(agent, 'interaction_logs'): 
        agent.interaction_logs.clear()
        print("DEBUG Gradio: Cleared agent interaction_logs for new 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) 
        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): 
            yield msg_dict
    
    if not all_step_logs:
        yield {"role": "assistant", "content": "Agent did not produce any output."}
        return

    final_answer_content = all_step_logs[-1] 

    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)}")

    if isinstance(actual_content_for_handling, PILImage.Image):
        print("DEBUG Gradio (stream_to_gradio): Actual 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 PIL image to temp path: {image_path_for_gradio}")
            # MODIFIED: Yield tuple (filepath, alt_text)
            yield {"role": "assistant", "content": (image_path_for_gradio, "Generated Image")} 
            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):** {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): final_answer_processed is AgentImage. Path: {image_path}")
        if image_path and os.path.exists(image_path):
             # MODIFIED: Yield tuple (filepath, alt_text)
             yield {"role": "assistant", "content": (image_path, "Generated Image (from AgentImage)")} 
        else:
            err_msg = f"Error: Image path from AgentImage ('{image_path}') not found or invalid."
            print(f"DEBUG Gradio: {err_msg}")
            yield {"role": "assistant", "content": f"**Final Answer ({err_msg})**"}
    elif isinstance(final_answer_processed, AgentAudio):
        audio_path = final_answer_processed.to_string()
        print(f"DEBUG Gradio (stream_to_gradio): AgentAudio path: {audio_path}")
        if audio_path and os.path.exists(audio_path):
            # MODIFIED: Yield tuple (filepath, alt_text) for consistency, though Gradio might just use path for audio
            yield {"role": "assistant", "content": (audio_path, "Generated Audio")}
        else:
            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: 
        yield {"role": "assistant", "content": f"**Final Answer:**\n{str(final_answer_processed)}"}


class GradioUI:
    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]'`")
        self.agent = agent
        self.file_upload_folder = file_upload_folder
        if self.file_upload_folder is not None:
            if not os.path.exists(self.file_upload_folder):
                os.makedirs(self.file_upload_folder, exist_ok=True)
        self._latest_file_path_for_download = None

    def _check_for_created_file(self):
        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): 
                if isinstance(log_entry, ActionStep):
                    observations = getattr(log_entry, 'observations', None)
                    if observations and isinstance(observations, str):
                        print(f"DEBUG Gradio UI: Checking observations: {observations[:200]}") 
                        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_history: list):
        print(f"DEBUG Gradio: interact_with_agent called with prompt: '{prompt_text}'")
        print(f"DEBUG Gradio: Current chat history (input type {type(current_chat_history)}): {current_chat_history}")

        updated_chat_history = current_chat_history + [{"role": "user", "content": prompt_text}]
        
        yield updated_chat_history, gr.update(visible=False), gr.update(value=None, visible=False)

        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 updated_chat_history + agent_responses_for_history, gr.update(visible=False), gr.update(value=None, visible=False) 

        file_found = self._check_for_created_file()
        
        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: 
            return gr.update(value="No file uploaded.", visible=True), file_uploads_log_state

        if not self.file_upload_folder or not os.path.exists(self.file_upload_folder):
            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",
        ]
        
        original_name = file.orig_name if hasattr(file, 'orig_name') and file.orig_name else os.path.basename(file.name)
        
        mime_type, _ = mimetypes.guess_type(file.name) 
        if mime_type is None: 
            mime_type, _ = mimetypes.guess_type(original_name)

        if mime_type not in allowed_file_types:
            return gr.update(value=f"File type '{mime_type or 'unknown'}' for '{original_name}' is disallowed.", visible=True), file_uploads_log_state

        sanitized_name = re.sub(r"[^\w\-.]", "_", original_name)
        base_name, current_ext = os.path.splitext(sanitized_name)
        
        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 = common_mime_to_ext.get(mime_type)

        if expected_ext and current_ext.lower() != expected_ext.lower():
            sanitized_name = base_name + expected_ext
        
        destination_path = os.path.join(self.file_upload_folder, sanitized_name)
        
        try:
            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}", 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[:300]}...") 
        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):
            print(f"DEBUG Gradio UI: Preparing download for UI component: {self._latest_file_path_for_download}")
            return gr.File.update(value=self._latest_file_path_for_download, 
                                  label=os.path.basename(self._latest_file_path_for_download), 
                                  visible=True)
        else:
            print("DEBUG Gradio UI: No valid file path to prepare for download component.")
            return gr.File.update(visible=False, value=None) 

    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("## Smol Talk with your Agent") 

            with gr.Row(equal_height=False): 
                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=700, 
                        show_copy_button=True,
                        bubble_full_width=False,
                        show_label=False 
                    )
                    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...",
                        show_label=False 
                    )
                
                with gr.Column(scale=1):
                    with gr.Accordion("File Upload", open=False): 
                        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) 
                        file_uploader.upload( 
                            self.upload_file,
                            [file_uploader, file_uploads_log_state],
                            [upload_status_text, file_uploads_log_state],
                        )
                    
                    with gr.Accordion("Generated File", open=True): 
                        download_action_button = gr.Button("Download Generated File", visible=False) 
                        file_download_display_component = gr.File(label="Downloadable Document", visible=False, interactive=False) 

            text_message_input.submit(
                self.log_user_message, 
                [text_message_input, file_uploads_log_state],
                [prepared_prompt_for_agent, text_message_input] 
            ).then(
                self.interact_with_agent, 
                [prepared_prompt_for_agent, chatbot_display], 
                [chatbot_display, download_action_button, file_download_display_component] 
            )

            download_action_button.click(
                self.prepare_and_show_download_file, 
                [], 
                [file_download_display_component] 
            )
        demo.launch(debug=True, share=kwargs.get("share", False), **kwargs)

__all__ = ["stream_to_gradio", "GradioUI"]