adityaardak/Smolagents_Interactive_Coding_Assistant
0
1#!/usr/bin/env python
2# coding=utf-8
3# 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 at
8#
9# http://www.apache.org/licenses/LICENSE-2.0
10#
11# Unless required by applicable law or agreed to in writing, software
12# 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 and
15# limitations under the License.
16import mimetypes
17import os
18import re
19import shutil
20from typing import Optional
21
22from smolagents.agent_types import AgentAudio, AgentImage, AgentText, handle_agent_output_types
23from smolagents.agents import ActionStep, MultiStepAgent
24from smolagents.memory import MemoryStep
25from smolagents.utils import _is_package_available
26
27
28def pull_messages_from_step(
29 step_log: MemoryStep,
30):
31 """Extract ChatMessage objects from agent steps with proper nesting"""
32 import gradio as gr
33
34 if isinstance(step_log, ActionStep):
35 # Output the step number
36 step_number = f"Step {step_log.step_number}" if step_log.step_number is not None else ""
37 yield gr.ChatMessage(role="assistant", content=f"**{step_number}**")
38
39 # First yield the thought/reasoning from the LLM
40 if hasattr(step_log, "model_output") and step_log.model_output is not None:
41 # Clean up the LLM output
42 model_output = step_log.model_output.strip()
43 # Remove any trailing <end_code> and extra backticks, handling multiple possible formats
44 model_output = re.sub(r"```\s*<end_code>", "```", model_output) # handles ```<end_code>
45 model_output = re.sub(r"<end_code>\s*```", "```", model_output) # handles <end_code>```
46 model_output = re.sub(r"```\s*\n\s*<end_code>", "```", model_output) # handles ```\n<end_code>
47 model_output = model_output.strip()
48 yield gr.ChatMessage(role="assistant", content=model_output)
49
50 # For tool calls, create a parent message
51 if hasattr(step_log, "tool_calls") and step_log.tool_calls is not None:
52 first_tool_call = step_log.tool_calls[0]
53 used_code = first_tool_call.name == "python_interpreter"
54 parent_id = f"call_{len(step_log.tool_calls)}"
55
56 # Tool call becomes the parent message with timing info
57 # First we will handle arguments based on type
58 args = first_tool_call.arguments
59 if isinstance(args, dict):
60 content = str(args.get("answer", str(args)))
61 else:
62 content = str(args).strip()
63
64 if used_code:
65 # Clean up the content by removing any end code tags
66 content = re.sub(r"```.*?\n", "", content) # Remove existing code blocks
67 content = re.sub(r"\s*<end_code>\s*", "", content) # Remove end_code tags
68 content = content.strip()
69 if not content.startswith("```python"):
70 content = f"```python\n{content}\n```"
71
72 parent_message_tool = gr.ChatMessage(
73 role="assistant",
74 content=content,
75 metadata={
76 "title": f"🛠️ Used tool {first_tool_call.name}",
77 "id": parent_id,
78 "status": "pending",
79 },
80 )
81 yield parent_message_tool
82
83 # Nesting execution logs under the tool call if they exist
84 if hasattr(step_log, "observations") and (
85 step_log.observations is not None and step_log.observations.strip()
86 ): # Only yield execution logs if there's actual content
87 log_content = step_log.observations.strip()
88 if log_content:
89 log_content = re.sub(r"^Execution logs:\s*", "", log_content)
90 yield gr.ChatMessage(
91 role="assistant",
92 content=f"{log_content}",
93 metadata={"title": "📝 Execution Logs", "parent_id": parent_id, "status": "done"},
94 )
95
96 # Nesting any errors under the tool call
97 if hasattr(step_log, "error") and step_log.error is not None:
98 yield gr.ChatMessage(
99 role="assistant",
100 content=str(step_log.error),
101 metadata={"title": "💥 Error", "parent_id": parent_id, "status": "done"},
102 )
103
104 # Update parent message metadata to done status without yielding a new message
105 parent_message_tool.metadata["status"] = "done"
106
107 # Handle standalone errors but not from tool calls
108 elif hasattr(step_log, "error") and step_log.error is not None:
109 yield gr.ChatMessage(role="assistant", content=str(step_log.error), metadata={"title": "💥 Error"})
110
111 # Calculate duration and token information
112 step_footnote = f"{step_number}"
113 if hasattr(step_log, "input_token_count") and hasattr(step_log, "output_token_count"):
114 token_str = (
115 f" | Input-tokens:{step_log.input_token_count:,} | Output-tokens:{step_log.output_token_count:,}"
116 )
117 step_footnote += token_str
118 if hasattr(step_log, "duration"):
119 step_duration = f" | Duration: {round(float(step_log.duration), 2)}" if step_log.duration else None
120 step_footnote += step_duration
121 step_footnote = f"""<span style="color: #bbbbc2; font-size: 12px;">{step_footnote}</span> """
122 yield gr.ChatMessage(role="assistant", content=f"{step_footnote}")
123 yield gr.ChatMessage(role="assistant", content="-----")
124
125
126def stream_to_gradio(
127 agent,
128 task: str,
129 reset_agent_memory: bool = False,
130 additional_args: Optional[dict] = None,
131):
132 """Runs an agent with the given task and streams the messages from the agent as gradio ChatMessages."""
133 if not _is_package_available("gradio"):
134 raise ModuleNotFoundError(
135 "Please install 'gradio' extra to use the GradioUI: `pip install 'smolagents[gradio]'`"
136 )
137 import gradio as gr
138
139 total_input_tokens = 0
140 total_output_tokens = 0
141
142 for step_log in agent.run(task, stream=True, reset=reset_agent_memory, additional_args=additional_args):
143 # Track tokens if model provides them
144 if hasattr(agent.model, "last_input_token_count"):
145 total_input_tokens += agent.model.last_input_token_count or 0
146 total_output_tokens += agent.model.last_output_token_count
147 if isinstance(step_log, ActionStep):
148 step_log.input_token_count = agent.model.last_input_token_count
149 step_log.output_token_count = agent.model.last_output_token_count
150
151 for message in pull_messages_from_step(
152 step_log,
153 ):
154 yield message
155
156 final_answer = step_log # Last log is the run's final_answer
157 final_answer = handle_agent_output_types(final_answer)
158
159 if isinstance(final_answer, AgentText):
160 yield gr.ChatMessage(
161 role="assistant",
162 content=f"**Final answer:**\n{final_answer.to_string()}\n",
163 )
164 elif isinstance(final_answer, AgentImage):
165 yield gr.ChatMessage(
166 role="assistant",
167 content={"path": final_answer.to_string(), "mime_type": "image/png"},
168 )
169 elif isinstance(final_answer, AgentAudio):
170 yield gr.ChatMessage(
171 role="assistant",
172 content={"path": final_answer.to_string(), "mime_type": "audio/wav"},
173 )
174 else:
175 yield gr.ChatMessage(role="assistant", content=f"**Final answer:** {str(final_answer)}")
176
177
178class GradioUI:
179 """A one-line interface to launch your agent in Gradio"""
180
181 def __init__(self, agent: MultiStepAgent, file_upload_folder: str | None = None):
182 if not _is_package_available("gradio"):
183 raise ModuleNotFoundError(
184 "Please install 'gradio' extra to use the GradioUI: `pip install 'smolagents[gradio]'`"
185 )
186 self.agent = agent
187 self.file_upload_folder = file_upload_folder
188 if self.file_upload_folder is not None:
189 if not os.path.exists(file_upload_folder):
190 os.mkdir(file_upload_folder)
191
192 def interact_with_agent(self, prompt, messages):
193 import gradio as gr
194
195 messages.append(gr.ChatMessage(role="user", content=prompt))
196 yield messages
197 for msg in stream_to_gradio(self.agent, task=prompt, reset_agent_memory=False):
198 messages.append(msg)
199 yield messages
200 yield messages
201
202 def upload_file(
203 self,
204 file,
205 file_uploads_log,
206 allowed_file_types=[
207 "application/pdf",
208 "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
209 "text/plain",
210 ],
211 ):
212 """
213 Handle file uploads, default allowed types are .pdf, .docx, and .txt
214 """
215 import gradio as gr
216
217 if file is None:
218 return gr.Textbox("No file uploaded", visible=True), file_uploads_log
219
220 try:
221 mime_type, _ = mimetypes.guess_type(file.name)
222 except Exception as e:
223 return gr.Textbox(f"Error: {e}", visible=True), file_uploads_log
224
225 if mime_type not in allowed_file_types:
226 return gr.Textbox("File type disallowed", visible=True), file_uploads_log
227
228 # Sanitize file name
229 original_name = os.path.basename(file.name)
230 sanitized_name = re.sub(
231 r"[^\w\-.]", "_", original_name
232 ) # Replace any non-alphanumeric, non-dash, or non-dot characters with underscores
233
234 type_to_ext = {}
235 for ext, t in mimetypes.types_map.items():
236 if t not in type_to_ext:
237 type_to_ext[t] = ext
238
239 # Ensure the extension correlates to the mime type
240 sanitized_name = sanitized_name.split(".")[:-1]
241 sanitized_name.append("" + type_to_ext[mime_type])
242 sanitized_name = "".join(sanitized_name)
243
244 # Save the uploaded file to the specified folder
245 file_path = os.path.join(self.file_upload_folder, os.path.basename(sanitized_name))
246 shutil.copy(file.name, file_path)
247
248 return gr.Textbox(f"File uploaded: {file_path}", visible=True), file_uploads_log + [file_path]
249
250 def log_user_message(self, text_input, file_uploads_log):
251 return (
252 text_input
253 + (
254 f"\nYou have been provided with these files, which might be helpful or not: {file_uploads_log}"
255 if len(file_uploads_log) > 0
256 else ""
257 ),
258 "",
259 )
260
261 def launch(self, **kwargs):
262 import gradio as gr
263
264 with gr.Blocks(fill_height=True) as demo:
265 stored_messages = gr.State([])
266 file_uploads_log = gr.State([])
267 chatbot = gr.Chatbot(
268 label="Agent",
269 type="messages",
270 avatar_images=(
271 None,
272 "https://huggingface.co/datasets/agents-course/course-images/resolve/main/en/communication/Alfred.png",
273 ),
274 resizeable=True,
275 scale=1,
276 )
277 # If an upload folder is provided, enable the upload feature
278 if self.file_upload_folder is not None:
279 upload_file = gr.File(label="Upload a file")
280 upload_status = gr.Textbox(label="Upload Status", interactive=False, visible=False)
281 upload_file.change(
282 self.upload_file,
283 [upload_file, file_uploads_log],
284 [upload_status, file_uploads_log],
285 )
286 text_input = gr.Textbox(lines=1, label="Chat Message")
287 text_input.submit(
288 self.log_user_message,
289 [text_input, file_uploads_log],
290 [stored_messages, text_input],
291 ).then(self.interact_with_agent, [stored_messages, chatbot], [chatbot])
292
293 demo.launch(debug=True, share=True, **kwargs)
294
295
296__all__ = ["stream_to_gradio", "GradioUI"]