Gradio_UI.py

import mimetypes
import os
import re
import shutil
from typing import Optional

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


def pull_messages_from_step(step_log: MemoryStep):
    """Extract ChatMessage objects from agent steps with proper nesting"""
    import gradio as gr

    if isinstance(step_log, ActionStep):
        step_number = f"Step {step_log.step_number}" if step_log.step_number is not None else ""
        yield gr.ChatMessage(role="assistant", content=f"**{step_number}**")

        if hasattr(step_log, "model_output") and step_log.model_output:
            model_output = step_log.model_output.strip()
            model_output = re.sub(r"```\s*<end_code>", "```", model_output)
            model_output = re.sub(r"<end_code>\s*```", "```", model_output)
            model_output = re.sub(r"```\s*\n\s*<end_code>", "```", model_output)
            yield gr.ChatMessage(role="assistant", content=model_output)

        if hasattr(step_log, "tool_calls") and step_log.tool_calls:
            first_tool_call = step_log.tool_calls[0]
            used_code = first_tool_call.name == "python_interpreter"
            parent_id = f"call_{len(step_log.tool_calls)}"

            args = first_tool_call.arguments
            content = str(args.get("answer", str(args))) if isinstance(args, dict) else str(args).strip()

            if used_code:
                content = re.sub(r"```.*?\n", "", content)
                content = re.sub(r"\s*<end_code>\s*", "", content)
                content = content.strip()
                if not content.startswith("```python"):
                    content = f"```python\n{content}\n```"

            parent_message_tool = gr.ChatMessage(
                role="assistant",
                content=content,
                metadata={"title": f"🛠️ Used tool {first_tool_call.name}", "id": parent_id, "status": "pending"},
            )
            yield parent_message_tool

            if hasattr(step_log, "observations") and step_log.observations:
                log_content = re.sub(r"^Execution logs:\s*", "", step_log.observations.strip())
                yield gr.ChatMessage(
                    role="assistant",
                    content=log_content,
                    metadata={"title": "📝 Execution Logs", "parent_id": parent_id, "status": "done"},
                )

            if hasattr(step_log, "error") and step_log.error:
                yield gr.ChatMessage(
                    role="assistant",
                    content=str(step_log.error),
                    metadata={"title": "💥 Error", "parent_id": parent_id, "status": "done"},
                )

            parent_message_tool.metadata["status"] = "done"

        elif hasattr(step_log, "error") and step_log.error:
            yield gr.ChatMessage(role="assistant", content=str(step_log.error), metadata={"title": "💥 Error"})

        step_footnote = f"{step_number}"
        if hasattr(step_log, "input_token_count") and hasattr(step_log, "output_token_count"):
            step_footnote += f" | Input-tokens:{step_log.input_token_count:,} | Output-tokens:{step_log.output_token_count:,}"
        if hasattr(step_log, "duration"):
            step_footnote += f" | Duration: {round(float(step_log.duration), 2)}"
        step_footnote = f"""<span style="color: #bbbbc2; font-size: 12px;">{step_footnote}</span> """
        yield gr.ChatMessage(role="assistant", content=step_footnote)
        yield gr.ChatMessage(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("Please install 'gradio': `pip install 'smolagents[gradio]'`")

    import gradio as gr
    from smolagents.agents import FinalAnswerStep

    total_input_tokens = 0
    total_output_tokens = 0

    for step_log in agent.run(task, stream=True, reset=reset_agent_memory, additional_args=additional_args):
        if hasattr(agent.model, "last_input_token_count"):
            total_input_tokens += agent.model.last_input_token_count
            total_output_tokens += agent.model.last_output_token_count
            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 message in pull_messages_from_step(step_log):
            yield message

    final_output = None
    if isinstance(step_log, FinalAnswerStep):
        final_output = step_log.final_answer
    else:
        final_output = getattr(step_log, "tool_output", None)
        if not final_output and hasattr(step_log, "tool_calls"):
            for call in step_log.tool_calls:
                if call.name == "final_answer":
                    final_output = call.arguments.get("answer")
                    break

    final_output = handle_agent_output_types(final_output)

    if isinstance(final_output, AgentText):
        yield gr.ChatMessage(role="assistant", content=final_output.to_string().strip())
    elif isinstance(final_output, AgentImage):
        yield gr.ChatMessage(role="assistant", content={"path": final_output.to_string(), "mime_type": "image/png"})
    elif isinstance(final_output, AgentAudio):
        yield gr.ChatMessage(role="assistant", content={"path": final_output.to_string(), "mime_type": "audio/wav"})
    else:
        yield gr.ChatMessage(role="assistant", content=str(final_output).strip())


class GradioUI:
    def __init__(self, agent: MultiStepAgent, file_upload_folder: Optional[str] = None, submit_fn=None):
        if not _is_package_available("gradio"):
            raise ModuleNotFoundError("Please install 'gradio': `pip install 'smolagents[gradio]'`")
        self.agent = agent
        self.file_upload_folder = file_upload_folder
        self.submit_fn = submit_fn
        if self.file_upload_folder and not os.path.exists(file_upload_folder):
            os.mkdir(file_upload_folder)

    def interact_with_agent(self, prompt, messages):
        import gradio as gr
        messages.append(gr.ChatMessage(role="user", content=prompt))
        yield messages
        for msg in stream_to_gradio(self.agent, task=prompt, reset_agent_memory=False):
            messages.append(msg)
            yield messages
        yield messages

    def upload_file(self, file, file_uploads_log, allowed_file_types=None):
        import gradio as gr
        if allowed_file_types is None:
            allowed_file_types = [
                "application/pdf",
                "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
                "text/plain",
            ]
        if file is None:
            return gr.Textbox("No file uploaded", visible=True), file_uploads_log

        try:
            mime_type, _ = mimetypes.guess_type(file.name)
        except Exception as e:
            return gr.Textbox(f"Error: {e}", visible=True), file_uploads_log

        if mime_type not in allowed_file_types:
            return gr.Textbox("File type disallowed", visible=True), file_uploads_log

        sanitized_name = re.sub(r"[^\w\-.]", "_", os.path.basename(file.name))
        ext = mimetypes.guess_extension(mime_type)
        sanitized_name = f"{os.path.splitext(sanitized_name)[0]}{ext or ''}"

        file_path = os.path.join(self.file_upload_folder, sanitized_name)
        shutil.copy(file.name, file_path)

        return gr.Textbox(f"File uploaded: {file_path}", visible=True), file_uploads_log + [file_path]

    def log_user_message(self, text_input, file_uploads_log):
        return (
            text_input
            + (f"\nYou have been provided with these files, which might be helpful or not: {file_uploads_log}"
               if file_uploads_log else ""),
            "",
        )

    def launch(self, **kwargs):
        import gradio as gr

        with gr.Blocks(fill_height=True) as demo:
            stored_messages = gr.State([])
            file_uploads_log = gr.State([])

            chatbot = gr.Chatbot(
                label="Agent",
                type="messages",
                avatar_images=(None, "https://huggingface.co/datasets/agents-course/course-images/resolve/main/en/communication/Alfred.png"),
            )

            if self.file_upload_folder:
                upload_file = gr.File(label="Upload a file")
                upload_status = gr.Textbox(label="Upload Status", interactive=False, visible=False)
                upload_file.change(self.upload_file, [upload_file, file_uploads_log], [upload_status, file_uploads_log])

            text_input = gr.Textbox(lines=1, label="Chat Message")
            text_input.submit(
                self.log_user_message,
                [text_input, file_uploads_log],
                [stored_messages, text_input],
            ).then(self.interact_with_agent, [stored_messages, chatbot], [chatbot])

            submit_btn = gr.Button("📤 Submit All GAIA Answers")
            submit_status = gr.Textbox(label="Submission Status", interactive=False)

            if self.submit_fn:
                submit_btn.click(
                    fn=lambda: self.submit_fn() or "✅ Submitted to GAIA scoring API",
                    inputs=[],
                    outputs=[submit_status],
                )

        demo.launch(debug=True, share=True, show_error=True, **kwargs)


__all__ = ["stream_to_gradio", "GradioUI"]