RemVdH/Test_based_on_agent_template

1#!/usr/bin/env python2# coding=utf-83# Copyright 2024 The HuggingFace Inc. team. All rights reserved.4#5# Licensed under the Apache License, Version 2.0 (the "License");6# you may not use this file except in compliance with the License.7# You may obtain a copy of the License at8#9#     http://www.apache.org/licenses/LICENSE-2.010#11# Unless required by applicable law or agreed to in writing, software12# distributed under the License is distributed on an "AS IS" BASIS,13# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.14# See the License for the specific language governing permissions and15# limitations under the License.16import mimetypes17import os18import re19import shutil20from typing import Optional21 22from smolagents.agent_types import AgentAudio, AgentImage, AgentText, handle_agent_output_types23from smolagents.agents import ActionStep, MultiStepAgent24from smolagents.memory import MemoryStep25from smolagents.utils import _is_package_available26 27def pull_messages_from_step(28    step_log: MemoryStep,29):30    """Extract ChatMessage objects from agent steps with proper nesting"""31    import gradio as gr32 33    if isinstance(step_log, ActionStep):34        # Output the step number35        step_number = f"Step {step_log.step_number}" if step_log.step_number is not None else ""36        yield gr.ChatMessage(role="assistant", content=f"**{step_number}**")37 38        # First yield the thought/reasoning from the LLM39        if hasattr(step_log, "model_output") and step_log.model_output is not None:40            # Clean up the LLM output41            model_output = step_log.model_output.strip()42            # Remove any trailing <end_code> and extra backticks, handling multiple possible formats43            model_output = re.sub(r"```\s*<end_code>", "```", model_output)  # handles ```<end_code>44            model_output = re.sub(r"<end_code>\s*```", "```", model_output)  # handles <end_code>```45            model_output = re.sub(r"```\s*\n\s*<end_code>", "```", model_output)  # handles ```\n<end_code>46            model_output = model_output.strip()47            yield gr.ChatMessage(role="assistant", content=model_output)48 49        # For tool calls, create a parent message50        if hasattr(step_log, "tool_calls") and step_log.tool_calls is not None:51            first_tool_call = step_log.tool_calls[0]52            used_code = first_tool_call.name == "python_interpreter"53            parent_id = f"call_{len(step_log.tool_calls)}"54 55            # Tool call becomes the parent message with timing info56            # First we will handle arguments based on type57            args = first_tool_call.arguments58            if isinstance(args, dict):59                content = str(args.get("answer", str(args)))60            else:61                content = str(args).strip()62 63            if used_code:64                # Clean up the content by removing any end code tags65                content = re.sub(r"```.*?\n", "", content)  # Remove existing code blocks66                content = re.sub(r"\s*<end_code>\s*", "", content)  # Remove end_code tags67                content = content.strip()68                if not content.startswith("```python"):69                    content = f"```python\n{content}\n```"70 71            parent_message_tool = gr.ChatMessage(72                role="assistant",73                content=content,74                metadata={75                    "title": f"🛠️ Used tool {first_tool_call.name}",76                    "id": parent_id,77                    "status": "pending",78                },79            )80            yield parent_message_tool81 82            # Nesting execution logs under the tool call if they exist83            if hasattr(step_log, "observations") and (84                step_log.observations is not None and step_log.observations.strip()85            ):  # Only yield execution logs if there's actual content86                log_content = step_log.observations.strip()87                if log_content:88                    log_content = re.sub(r"^Execution logs:\s*", "", log_content)89                    yield gr.ChatMessage(90                        role="assistant",91                        content=f"{log_content}",92                        metadata={"title": "📝 Execution Logs", "parent_id": parent_id, "status": "done"},93                    )94 95            # Nesting any errors under the tool call96            if hasattr(step_log, "error") and step_log.error is not None:97                yield gr.ChatMessage(98                    role="assistant",99                    content=str(step_log.error),100                    metadata={"title": "💥 Error", "parent_id": parent_id, "status": "done"},101                )102 103            # Update parent message metadata to done status without yielding a new message104            parent_message_tool.metadata["status"] = "done"105 106        # Handle standalone errors but not from tool calls107        elif hasattr(step_log, "error") and step_log.error is not None:108            yield gr.ChatMessage(role="assistant", content=str(step_log.error), metadata={"title": "💥 Error"})109 110        # Calculate duration and token information111        step_footnote = f"{step_number}"112        if hasattr(step_log, "input_token_count") and hasattr(step_log, "output_token_count"):113            token_str = (114                f" | Input-tokens:{step_log.input_token_count:,} | Output-tokens:{step_log.output_token_count:,}"115            )116            step_footnote += token_str117        if hasattr(step_log, "duration"):118            step_duration = f" | Duration: {round(float(step_log.duration), 2)}" if step_log.duration else None119            step_footnote += step_duration120        step_footnote = f"""<span style="color: #bbbbc2; font-size: 12px;">{step_footnote}</span> """121        yield gr.ChatMessage(role="assistant", content=f"{step_footnote}")122        yield gr.ChatMessage(role="assistant", content="-----")123 124 125def stream_to_gradio(126    agent,127    task: str,128    reset_agent_memory: bool = False,129    additional_args: Optional[dict] = None,130):131    """Runs an agent with the given task and streams the messages from the agent as gradio ChatMessages."""132    if not _is_package_available("gradio"):133        raise ModuleNotFoundError(134            "Please install 'gradio' extra to use the GradioUI: `pip install 'smolagents[gradio]'`"135        )136    import gradio as gr137 138    total_input_tokens = 0139    total_output_tokens = 0140 141    for step_log in agent.run(task, stream=True, reset=reset_agent_memory, additional_args=additional_args):142        # Track tokens if model provides them143        if hasattr(agent.model, "last_input_token_count"):144            total_input_tokens += agent.model.last_input_token_count145            total_output_tokens += agent.model.last_output_token_count146            if isinstance(step_log, ActionStep):147                step_log.input_token_count = agent.model.last_input_token_count148                step_log.output_token_count = agent.model.last_output_token_count149 150        for message in pull_messages_from_step(151            step_log,152        ):153            yield message154 155    final_answer = step_log  # Last log is the run's final_answer156    final_answer = handle_agent_output_types(final_answer)157 158    if isinstance(final_answer, AgentText):159        yield gr.ChatMessage(160            role="assistant",161            content=f"**Final answer:**\n{final_answer.to_string()}\n",162        )163    elif isinstance(final_answer, AgentImage):164        yield gr.ChatMessage(165            role="assistant",166            content={"path": final_answer.to_string(), "mime_type": "image/png"},167        )168    elif isinstance(final_answer, AgentAudio):169        yield gr.ChatMessage(170            role="assistant",171            content={"path": final_answer.to_string(), "mime_type": "audio/wav"},172        )173    else:174        yield gr.ChatMessage(role="assistant", content=f"**Final answer:** {str(final_answer)}")175 176 177class GradioUI:178    """A one-line interface to launch your agent in Gradio"""179 180    def __init__(self, agent: MultiStepAgent, file_upload_folder: str | None = None):181        if not _is_package_available("gradio"):182            raise ModuleNotFoundError(183                "Please install 'gradio' extra to use the GradioUI: `pip install 'smolagents[gradio]'`"184            )185        self.agent = agent186        self.file_upload_folder = file_upload_folder187        if self.file_upload_folder is not None:188            if not os.path.exists(file_upload_folder):189                os.mkdir(file_upload_folder)190 191    def interact_with_agent(self, prompt, messages):192        import gradio as gr193 194        messages.append(gr.ChatMessage(role="user", content=prompt))195        yield messages196        for msg in stream_to_gradio(self.agent, task=prompt, reset_agent_memory=False):197            messages.append(msg)198            yield messages199        yield messages200 201    def upload_file(202        self,203        file,204        file_uploads_log,205        allowed_file_types=[206            "application/pdf",207            "application/vnd.openxmlformats-officedocument.wordprocessingml.document",208            "text/plain",209        ],210    ):211        """212        Handle file uploads, default allowed types are .pdf, .docx, and .txt213        """214        import gradio as gr215 216        if file is None:217            return gr.Textbox("No file uploaded", visible=True), file_uploads_log218 219        try:220            mime_type, _ = mimetypes.guess_type(file.name)221        except Exception as e:222            return gr.Textbox(f"Error: {e}", visible=True), file_uploads_log223 224        if mime_type not in allowed_file_types:225            return gr.Textbox("File type disallowed", visible=True), file_uploads_log226 227        # Sanitize file name228        original_name = os.path.basename(file.name)229        sanitized_name = re.sub(230            r"[^\w\-.]", "_", original_name231        )  # Replace any non-alphanumeric, non-dash, or non-dot characters with underscores232 233        type_to_ext = {}234        for ext, t in mimetypes.types_map.items():235            if t not in type_to_ext:236                type_to_ext[t] = ext237 238        # Ensure the extension correlates to the mime type239        sanitized_name = sanitized_name.split(".")[:-1]240        sanitized_name.append("" + type_to_ext[mime_type])241        sanitized_name = "".join(sanitized_name)242 243        # Save the uploaded file to the specified folder244        file_path = os.path.join(self.file_upload_folder, os.path.basename(sanitized_name))245        shutil.copy(file.name, file_path)246 247        return gr.Textbox(f"File uploaded: {file_path}", visible=True), file_uploads_log + [file_path]248 249    def log_user_message(self, text_input, file_uploads_log):250        return (251            text_input252            + (253                f"\nYou have been provided with these files, which might be helpful or not: {file_uploads_log}"254                if len(file_uploads_log) > 0255                else ""256            ),257            "",258        )259 260    def launch(self, **kwargs):261        import gradio as gr262 263        with gr.Blocks(fill_height=True) as demo:264            stored_messages = gr.State([])265            file_uploads_log = gr.State([])266            chatbot = gr.Chatbot(267                label="Agent",268                type="messages",269                avatar_images=(270                    None,271                    "https://huggingface.co/datasets/agents-course/course-images/resolve/main/en/communication/Alfred.png",272                ),273                resizeable=True,274                scale=1,275            )276            # If an upload folder is provided, enable the upload feature277            if self.file_upload_folder is not None:278                upload_file = gr.File(label="Upload a file")279                upload_status = gr.Textbox(label="Upload Status", interactive=False, visible=False)280                upload_file.change(281                    self.upload_file,282                    [upload_file, file_uploads_log],283                    [upload_status, file_uploads_log],284                )285            text_input = gr.Textbox(lines=1, label="Chat Message")286            text_input.submit(287                self.log_user_message,288                [text_input, file_uploads_log],289                [stored_messages, text_input],290            ).then(self.interact_with_agent, [stored_messages, chatbot], [chatbot])291 292        demo.launch(debug=True, share=True, **kwargs)293 294 295__all__ = ["stream_to_gradio", "GradioUI"]