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 base64
import mimetypes
import os
import re
import shutil
import uuid
from typing import Optional

import requests
from smolagents.agent_types import AgentAudio, AgentImage, AgentText, handle_agent_output_types
from smolagents.agents import ActionStep, MultiStepAgent
from smolagents.memory import FinalAnswerStep, 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}**")

        # Show model output
        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>", "```", 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 call display
        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_{len(step_log.tool_calls)}"

            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

            # Tool observations (logs)
            if hasattr(step_log, "observations") and (
                step_log.observations is not None and step_log.observations.strip()
            ):
                log_content = step_log.observations.strip()
                if log_content:
                    log_content = re.sub(r"^Execution logs:\s*", "", log_content)
                    yield gr.ChatMessage(
                        role="assistant",
                        content=f"{log_content}",
                        metadata={"title": "📝 Execution Logs", "parent_id": parent_id, "status": "done"},
                    )

            # Tool error
            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 error
        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 timing
        step_footnote = f"{step_number}"
        if hasattr(step_log, "input_token_count") and hasattr(step_log, "output_token_count"):
            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"):
            step_duration = f" | Duration: {round(float(step_log.duration), 2)}" if step_log.duration else None
            step_footnote += step_duration

        step_footnote = f"""<span style="color: #bbbbc2; font-size: 12px;">{step_footnote}</span> """
        yield gr.ChatMessage(role="assistant", content=f"{step_footnote}")
        yield gr.ChatMessage(role="assistant", content="-----")


def _save_agent_image(agent_img: AgentImage) -> str:
    """
    Convert AgentImage into a real PNG file path so Gradio can render it.
    Supports:
      - existing file paths
      - PIL images in different attrs
      - raw bytes
      - base64
      - URLs
    """
    os.makedirs("generated_images", exist_ok=True)
    img_path = os.path.join("generated_images", f"image_{uuid.uuid4().hex[:8]}.png")

    img_str = agent_img.to_string()

    # 1) If to_string() is a valid local file path
    if isinstance(img_str, str) and os.path.exists(img_str):
        return img_str

    # 2) If to_string() looks like a URL
    if isinstance(img_str, str) and img_str.startswith("http"):
        try:
            r = requests.get(img_str, timeout=30)
            r.raise_for_status()
            with open(img_path, "wb") as f:
                f.write(r.content)
            return img_path
        except Exception:
            pass

    # 3) If to_string() looks like base64
    if isinstance(img_str, str) and "base64" in img_str[:50].lower():
        try:
            b64data = img_str.split("base64,")[-1]
            img_bytes = base64.b64decode(b64data)
            with open(img_path, "wb") as f:
                f.write(img_bytes)
            return img_path
        except Exception:
            pass

    # 4) Try extracting PIL image from common fields
    for attr in ["value", "image", "data", "pil_image"]:
        if hasattr(agent_img, attr):
            candidate = getattr(agent_img, attr)
            if candidate is None:
                continue

            # PIL image
            if hasattr(candidate, "save"):
                try:
                    candidate.save(img_path)
                    return img_path
                except Exception:
                    pass

            # bytes
            if isinstance(candidate, (bytes, bytearray)):
                try:
                    with open(img_path, "wb") as f:
                        f.write(candidate)
                    return img_path
                except Exception:
                    pass

    # 5) Try agent_img.to_pil()
    if hasattr(agent_img, "to_pil"):
        try:
            pil_img = agent_img.to_pil()
            if pil_img is not None and hasattr(pil_img, "save"):
                pil_img.save(img_path)
                return img_path
        except Exception:
            pass

    # If nothing worked, still return path (won't crash)
    return img_path


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 the messages from the agent as gradio ChatMessages."""
    if not _is_package_available("gradio"):
        raise ModuleNotFoundError(
            "Please install 'gradio' extra to use the GradioUI: `pip install 'smolagents[gradio]'`"
        )
    import gradio as gr

    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"):
            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

    raw_final_answer = step_log.final_answer if isinstance(step_log, FinalAnswerStep) else step_log

    # If a tool returns a local image path (e.g. via `save_image`), render it inline in the chat.
    if isinstance(raw_final_answer, str):
        candidate_path = raw_final_answer.strip()
        if candidate_path and os.path.exists(candidate_path):
            mime_type, _ = mimetypes.guess_type(candidate_path)
            if mime_type and mime_type.startswith("image/"):
                yield gr.ChatMessage(role="assistant", content={"path": candidate_path, "mime_type": mime_type})
                return

    final_answer = handle_agent_output_types(raw_final_answer)

    if isinstance(final_answer, AgentText):
        # If the text is actually a local image path, render the image.
        text = final_answer.to_string().strip()
        if text and os.path.exists(text):
            mime_type, _ = mimetypes.guess_type(text)
            if mime_type and mime_type.startswith("image/"):
                yield gr.ChatMessage(role="assistant", content={"path": text, "mime_type": mime_type})
                return
        yield gr.ChatMessage(role="assistant", content=f"**Final answer:**\n{text}\n")

    elif isinstance(final_answer, AgentImage):
        img_path = _save_agent_image(final_answer)
        yield gr.ChatMessage(role="assistant", content={"path": img_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:
        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 to use the GradioUI: `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(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=[
            "application/pdf",
            "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
            "text/plain",
        ],
    ):
        """Handle file uploads, default allowed types are .pdf, .docx, and .txt"""
        import gradio as gr

        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

        original_name = os.path.basename(file.name)
        sanitized_name = re.sub(r"[^\w\-.]", "_", original_name)

        type_to_ext = {}
        for ext, t in mimetypes.types_map.items():
            if t not in type_to_ext:
                type_to_ext[t] = ext

        sanitized_name = sanitized_name.split(".")[:-1]
        sanitized_name.append("" + type_to_ext[mime_type])
        sanitized_name = "".join(sanitized_name)

        file_path = os.path.join(self.file_upload_folder, os.path.basename(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 len(file_uploads_log) > 0
                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",
                avatar_images=(
                    None,
                    "https://huggingface.co/datasets/agents-course/course-images/resolve/main/en/communication/Alfred.png",
                ),
                resizable=True,
                scale=1,
            )

            if self.file_upload_folder is not None:
                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])

        # Disable share on Spaces automatically
        is_spaces = os.environ.get("SPACE_ID") is not None
        demo.launch(debug=True, share=not is_spaces, **kwargs)


__all__ = ["stream_to_gradio", "GradioUI"]