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

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

from PIL import Image as PILImage
from io import BytesIO
import base64


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):
        # Output the step number
        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}**")

        # First yield the thought/reasoning from the LLM
        if hasattr(step_log, "model_output") and step_log.model_output is not None:
            model_output = step_log.model_output.strip()
            # Clean up code block endings
            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)
            model_output = model_output.strip()
            yield gr.ChatMessage(role="assistant", content=model_output)

        # Tool calls handling
        if hasattr(step_log, "tool_calls") and step_log.tool_calls is not None:
            first_tool_call = step_log.tool_calls[0]
            used_code = first_tool_call.name == "python_interpreter"
            parent_id = f"call_{id(step_log)}"

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

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

            # Errors from tool
            if hasattr(step_log, "error") and step_log.error is not None:
                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"

        # Standalone errors
        elif hasattr(step_log, "error") and step_log.error is not None:
            yield gr.ChatMessage(role="assistant", content=str(step_log.error), metadata={"title": "Error"})

        # Footnote with tokens and duration
        step_footnote = step_number
        if hasattr(step_log, "input_token_count") and step_log.input_token_count is not None:
            if hasattr(step_log, "output_token_count") and step_log.output_token_count is not None:
                token_str = f" | Input-tokens:{step_log.input_token_count:,} | Output-tokens:{step_log.output_token_count:,}"
                step_footnote += token_str
        if hasattr(step_log, "duration") and step_log.duration:
            step_footnote += f" | Duration: {round(float(step_log.duration), 2)}s"

        if step_footnote != step_number:
            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,
):
    """Runs an agent with the given task and streams messages as Gradio ChatMessages."""
    if not _is_package_available("gradio"):
        raise ModuleNotFoundError("Please install 'gradio' extra: `pip install 'smolagents[gradio]'`")

    import gradio as gr

    total_input_tokens = 0
    total_output_tokens = 0

    for step_log in agent.run(
        task=task,
        stream=True,
        reset=reset_agent_memory,
        additional_args=additional_args or {},
    ):
        # Safely accumulate token counts
        if hasattr(agent.model, "last_input_token_count"):
            total_input_tokens += agent.model.last_input_token_count or 0
        if hasattr(agent.model, "last_output_token_count"):
            total_output_tokens += agent.model.last_output_token_count or 0

        if isinstance(step_log, ActionStep):
            step_log.input_token_count = getattr(agent.model, "last_input_token_count", None)
            step_log.output_token_count = getattr(agent.model, "last_output_token_count", None)

        for message in pull_messages_from_step(step_log):
            yield message

    # Final answer handling
    final_answer = handle_agent_output_types(step_log)

    if isinstance(final_answer, AgentText):
        yield gr.ChatMessage(role="assistant", content=f"**Final answer:**\n{final_answer.to_string()}")

    elif isinstance(final_answer, AgentImage):
        # Direct support for AgentImage (returns path or PIL)
        image_path = final_answer.to_string()
        yield gr.ChatMessage(
            role="assistant",
            content="**Final answer (Image generated):**\n",
        )
        yield gr.ChatMessage(
            role="assistant",
            content={"path": image_path, "mime_type": "image/png"},
        )

    elif isinstance(final_answer, AgentAudio):
        yield gr.ChatMessage(
            role="assistant",
            content={"path": final_answer.to_string(), "mime_type": "audio/wav"},
        )

    else:
        # Fallback: handle raw PIL Image, URL, or path
        if isinstance(final_answer, PILImage.Image):
            # Save temporarily if needed
            temp_path = "temp_generated_image.png"
            final_answer.save(temp_path)
            yield gr.ChatMessage(role="assistant", content={"path": temp_path, "mime_type": "image/png"})
        elif isinstance(final_answer, str) and (
            final_answer.startswith("http") or os.path.exists(final_answer)
        ):
            yield gr.ChatMessage(role="assistant", content={"path": final_answer, "mime_type": "image/png"})
        else:
            yield gr.ChatMessage(role="assistant", content=f"**Final answer:** {str(final_answer)}")


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("Please install 'gradio' extra: `pip install 'smolagents[gradio]'`")
        self.agent = agent
        self.file_upload_folder = file_upload_folder
        if self.file_upload_folder is not None:
            os.makedirs(file_upload_folder, exist_ok=True)

    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

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

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

        original_name = os.path.basename(file.name)
        sanitized_name = re.sub(r"[^\w\-.]", "_", original_name)
        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):
        extra = (
            f"\nYou have been provided with these files, which might be helpful or not: {file_uploads_log}"
            if file_uploads_log
            else ""
        )
        return text_input + extra, ""

    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",
                ),
                height=800,
            )

            if self.file_upload_folder is not None:
                upload_file = gr.File(label="Upload a file (PDF, DOCX, TXT)")
                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, placeholder="Type your message here...", label="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],
            )

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


__all__ = ["stream_to_gradio", "GradioUI"]