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 os
import re
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, tuple)):
         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}")
            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):
             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):
            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 = None 
        self._latest_file_path_for_download = None

    def _get_created_document_path(self):
        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 document paths.")
            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', [])
                    
                    is_python_interpreter_step = any(tc.name == "python_interpreter" for tc in tool_calls)

                    if is_python_interpreter_step and observations and isinstance(observations, str):
                        # CRITICAL DEBUG LINE: Print the exact observations string
                        print(f"DEBUG Gradio UI (_get_created_document_path): Python Interpreter Observations: '''{observations}'''") 
                        
                        match = re.search(
                            r"(?:Document created \((?:docx|pdf|txt)\):|Document converted to PDF:)\s*(/tmp/[a-zA-Z0-9_]+/generated_document\.(?:docx|pdf|txt))",
                            observations,
                            re.MULTILINE 
                        )
                        
                        if match:
                            extracted_path = match.group(1) 
                            print(f"DEBUG Gradio UI: Regex matched. Extracted path: '{extracted_path}'")
                            normalized_path = os.path.normpath(extracted_path)
                            if os.path.exists(normalized_path):
                                print(f"DEBUG Gradio UI: Validated path for download: {normalized_path}")
                                return normalized_path 
                            else:
                                print(f"DEBUG Gradio UI: Path from create_document output ('{normalized_path}') does not exist.")
        print("DEBUG Gradio UI: No valid generated document path found in agent logs.")
        return None 

    def interact_with_agent(self, prompt_text: str, current_chat_history: list):
        print(f"DEBUG Gradio: interact_with_agent called with prompt: '{prompt_text}'")

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

        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(value=None, visible=False) # For file_download_display_component

        final_chat_display_content = updated_chat_history + agent_responses_for_history
        
        document_path_to_display = self._get_created_document_path() 
        
        if document_path_to_display:
            print(f"DEBUG Gradio: Document found for display: {document_path_to_display}")
            # CORRECTED: Use gr.update() for the File component
            yield final_chat_display_content, gr.update(value=document_path_to_display, 
                                                             label=os.path.basename(document_path_to_display), 
                                                             visible=True)
        else:
            print(f"DEBUG Gradio: No document found for display.")
            # CORRECTED: Use gr.update() for the File component
            yield final_chat_display_content, gr.update(value=None, visible=False)


    def log_user_message(self, text_input_value: str):
        full_prompt = text_input_value
        print(f"DEBUG Gradio: Prepared prompt for agent: {full_prompt[:300]}...") 
        return full_prompt, "" 

    # prepare_and_show_download_file is not needed if we directly update the gr.File component

    def launch(self, **kwargs):
        with gr.Blocks(fill_height=True, theme=gr.themes.Soft(primary_hue=gr.themes.colors.blue)) as demo:
            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(
                        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, 
                        placeholder="Type your message and press Enter, or Shift+Enter for new line...",
                        show_label=False 
                    )
                
                with gr.Column(scale=1): 
                    # "Generated File" section directly shows the gr.File component
                    gr.Markdown("### Generated Document") 
                    file_download_display_component = gr.File(
                        label="Downloadable Document", 
                        visible=False, 
                        interactive=False 
                    )

            text_message_input.submit(
                self.log_user_message, 
                [text_message_input],
                [prepared_prompt_for_agent, text_message_input] 
            ).then(
                self.interact_with_agent, 
                [prepared_prompt_for_agent, chatbot_display], 
                [chatbot_display, file_download_display_component] # Outputs update chatbot and file component
            )
            
        demo.launch(debug=True, share=kwargs.get("share", False), **kwargs)

__all__ = ["stream_to_gradio", "GradioUI"]