File size: 14,885 Bytes
331400f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 |
"""Ankelodon Agent Adapter for the Hugging Face Agents Course evaluator.
This module exposes a simple Gradio-powered wrapper around the
`ankelodon_multiagent_system` project. It follows the same high-level flow
as the official GAIA template provided in the course materials: fetch
evaluation questions from the GAIA API, run your agent to produce
responses, and submit those responses back to the leaderboard.
The key differences between this adapter and the GAIA template are:
* It imports and uses your multi‑agent system defined in the `src`
package (see `src/agent.py`) via the `build_workflow` function. This
function returns a `langgraph` state machine capable of planning,
reasoning and executing tools. The adapter calls into this workflow
with a properly initialised `AgentState` and extracts the final
answer from the resulting state.
* It automatically downloads any file attachments associated with a
task (via the `/files/{task_id}` endpoint exposed by the evaluation
server) and saves them into a temporary directory. The local file
paths are passed into the agent through the `files` field of the
state. Your existing file handling logic (e.g. `preprocess_files`
in `src/tools/tools.py`) will detect the file type and suggest
appropriate tools.
* It strips any leading ``Final answer:`` prefix from the agent's
response. The evaluation server performs an exact string match
against the ground truth answer【842261069842380†L108-L112】, so it is
important that the returned text contains only the answer and
nothing else.
Before running this script yourself, make sure all dependencies in
`requirements.txt` are installed. To use the Gradio interface locally,
run `python ankelodon_adapter.py` from the project root. When deploying
as a Hugging Face Space for leaderboard submission, ensure the
`SPACE_ID` environment variable is set by the platform; it is used to
construct a link back to your code for verification.
"""
from __future__ import annotations
import os
import tempfile
from typing import Optional, List, Dict, Any
import requests
import gradio as gr
import pandas as pd
try:
# Import the multi‑agent system components. When running as a script
# within the project root, Python's module search path should
# already include the `src` directory. If you get import errors,
# ensure that the working directory is the repository root or
# append `src` to `sys.path` manually before these imports.
from agent import build_workflow
from config import config as WORKFLOW_CONFIG
from state import AgentState
except Exception as import_err:
raise RuntimeError(
"Failed to import the Ankelodon multi-agent system. "
"Make sure you are running this script from the repository root "
"and that the project has been installed correctly."
) from import_err
DEFAULT_API_URL: str = "https://agents-course-unit4-scoring.hf.space"
class AnkelodonAgent:
"""Simple callable wrapper around the Ankelodon multi‑agent system.
Instances of this class can be called directly with a natural
language question and an optional task identifier. Under the hood it
builds a `langgraph` workflow using ``build_workflow()``, prepares
an initial state, fetches any file attachments associated with
the task, and invokes the workflow to compute a final answer.
"""
def __init__(self) -> None:
# Initialise the workflow once per agent. Subsequent calls reuse
# the compiled state machine, which is more efficient than
# rebuilding it on every question.
self.workflow = build_workflow()
def _download_attachment(self, task_id: str) -> List[str]:
"""Download a file attachment for the given task ID.
The evaluation API exposes a ``/files/{task_id}`` endpoint【842261069842380†L95-L107】.
This helper downloads the content, infers a file extension
from the HTTP ``Content-Type`` header and writes the bytes to a
temporary file. It returns a list of file paths (zero or one
element) to be included in the agent state.
"""
files: List[str] = []
url = f"{DEFAULT_API_URL}/files/{task_id}"
try:
resp = requests.get(url, timeout=15, allow_redirects=True)
if resp.status_code == 200 and resp.content:
# Map common MIME substrings to file extensions. The
# multi‑agent system's file handling tools use the
# extension to determine how to process the file.
ctype = resp.headers.get("content-type", "").lower()
ext_map = {
"excel": ".xlsx",
"sheet": ".xlsx",
"csv": ".csv",
"python": ".py",
"audio": ".mp3",
"image": ".jpg",
}
extension = ""
for key, val in ext_map.items():
if key in ctype:
extension = val
break
tmp_dir = tempfile.mkdtemp(prefix="ankelodon_task_")
filename = f"attachment{extension}"
path = os.path.join(tmp_dir, filename)
with open(path, "wb") as fh:
fh.write(resp.content)
files.append(path)
except Exception as e:
# Log the error to console but don't fail the entire task.
print(f"[WARNING] Failed to fetch attachment for task {task_id}: {e}")
return files
def __call__(self, question: str, task_id: Optional[str] = None) -> str:
"""Run the multi‑agent system to answer a question.
Parameters
----------
question: str
The natural language query to answer.
task_id: Optional[str]
If provided, the ID used to fetch any associated file
attachment from the evaluation API. Attachments are stored
locally and passed into the agent via the ``files`` field.
Returns
-------
str
The final answer produced by the agent, with any "final
answer" prefix removed. If no answer is produced the empty
string is returned.
"""
# Build the initial agent state. The AgentState type defines
# numerous fields, many of which the workflow populates
# internally. We set only the essentials here. Unrecognised
# keys are ignored by the underlying state machine.
state: Dict[str, Any] = {
"query": question,
"final_answer": "",
"plan": None,
"complexity_assessment": None,
"current_step": 0,
"reasoning_done": False,
"messages": [],
"files": [],
"file_contents": {},
"critique_feedback": None,
"iteration_count": 0,
"max_iterations": 3,
"execution_report": None,
"previous_tool_results": {},
}
# If a task ID is provided, attempt to download its attachment.
if task_id:
attachment_paths = self._download_attachment(task_id)
if attachment_paths:
state["files"] = attachment_paths
# Invoke the workflow. The `config` parameter defines runtime
# options such as recursion limits and thread identifiers. It is
# imported from `src.config`.
try:
result_state = self.workflow.invoke(state, config=WORKFLOW_CONFIG)
except Exception as e:
print(f"[ERROR] Failed to run workflow: {e}")
return ""
# Extract the final answer. Depending on the branch taken,
# either the ``final_answer`` key or a generic ``answer`` key may
# be present. Use whichever exists. Some nodes may prepend
# "final answer:"; remove it for exact match scoring【842261069842380†L108-L112】.
answer = ""
if isinstance(result_state, dict):
answer = result_state.get("final_answer") or result_state.get("answer") or ""
if answer:
answer = answer.replace("Final answer:", "").replace("final answer:", "").strip()
return answer
def run_and_submit_all(profile: Optional[gr.OAuthProfile]) -> tuple[str, pd.DataFrame | None]:
"""Fetch all questions, run the agent, and submit the answers.
This function replicates the behaviour of the GAIA template's
``run_and_submit_all`` function【566837548679297†L247-L306】 but uses the
``AnkelodonAgent`` class defined above. It is bound to a Gradio
button in the UI. On success it returns a status message and a
DataFrame of results; on failure it returns an error message and
``None`` or an empty DataFrame.
"""
# Require the user to be logged in so we can report the username.
if not profile:
return "Please Login to Hugging Face with the button.", None
username = getattr(profile, "username", "").strip()
api_url = DEFAULT_API_URL
questions_url = f"{api_url}/questions"
submit_url = f"{api_url}/submit"
# Instantiate the agent once.
try:
agent = AnkelodonAgent()
print("Ankelodon agent initialised successfully")
except Exception as e:
err_msg = f"Error initialising agent: {e}"
print(err_msg)
return err_msg, None
# Fetch questions from the evaluation API.【566837548679297†L247-L268】
try:
print(f"Fetching questions from: {questions_url}")
resp = requests.get(questions_url, timeout=15)
resp.raise_for_status()
questions_data = resp.json()
if not questions_data:
return "Fetched questions list is empty or invalid format.", None
print(f"Fetched {len(questions_data)} questions.")
except Exception as e:
err_msg = f"Error fetching questions: {e}"
print(err_msg)
return err_msg, None
# Run the agent on each question.
results_log: List[Dict[str, Any]] = []
answers_payload: List[Dict[str, str]] = []
print(f"Running agent on {len(questions_data)} questions…")
for item in questions_data:
task_id = item.get("task_id")
question_text = item.get("question")
if not task_id or question_text is None:
print(f"Skipping item with missing task_id or question: {item}")
continue
try:
answer = agent(question_text, task_id)
answers_payload.append({"task_id": task_id, "submitted_answer": answer})
results_log.append({
"Task ID": task_id,
"Question": question_text,
"Submitted Answer": answer,
})
except Exception as e:
print(f"Error running agent on task {task_id}: {e}")
results_log.append({
"Task ID": task_id,
"Question": question_text,
"Submitted Answer": f"AGENT ERROR: {e}",
})
if not answers_payload:
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
# Prepare submission payload. The leaderboard displays a link to your
# code; this is constructed from the SPACE_ID environment variable.
space_id = os.getenv("SPACE_ID", "")
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else ""
submission_data = {
"username": username,
"agent_code": agent_code,
"answers": answers_payload,
}
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
try:
submission_resp = requests.post(submit_url, json=submission_data, timeout=60)
submission_resp.raise_for_status()
result_data = submission_resp.json()
final_status = (
f"Submission Successful!\n"
f"User: {result_data.get('username')}\n"
f"Overall Score: {result_data.get('score', 'N/A')}% "
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
f"Message: {result_data.get('message', 'No message received.')}"
)
print("Submission successful.")
return final_status, pd.DataFrame(results_log)
except Exception as e:
err_msg = f"Submission Failed: {e}"
print(err_msg)
return err_msg, pd.DataFrame(results_log)
# Build the Gradio interface. This interface resembles the official
# GAIA template【566837548679297†L372-L401】 but runs your Ankelodon agent.
with gr.Blocks() as demo:
gr.Markdown("# Ankelodon Agent Evaluation Runner")
gr.Markdown(
"""
**Instructions**
1. Clone this repository or duplicate the associated Hugging Face Space.
2. Log in to your Hugging Face account using the button below. Your HF
username is used to attribute your submission on the leaderboard.
3. Click **Run Evaluation & Submit All Answers** to fetch the questions,
run the Ankelodon agent on each one, submit your answers, and display
the resulting score and answers.
---
This template is intentionally lightweight. Feel free to customise it –
add caching, parallel execution or additional logging as you see fit.
"""
)
gr.LoginButton()
run_button = gr.Button("Run Evaluation & Submit All Answers")
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
if __name__ == "__main__":
# When running locally, print some information about the environment.
print("\n" + "-" * 30 + " Ankelodon Adapter Starting " + "-" * 30)
space_host_startup = os.getenv("SPACE_HOST")
space_id_startup = os.getenv("SPACE_ID")
if space_host_startup:
print(f"✅ SPACE_HOST found: {space_host_startup}")
print(f" Runtime URL should be: https://{space_host_startup}.hf.space")
else:
print("ℹ️ SPACE_HOST environment variable not found (running locally?).")
if space_id_startup:
print(f"✅ SPACE_ID found: {space_id_startup}")
print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}")
print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
else:
print("ℹ️ SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.")
print("-" * (60 + len(" Ankelodon Adapter Starting ")) + "\n")
# Launch the Gradio app.
demo.launch(debug=True, share=False) |