Update app.py
Browse files
app.py
CHANGED
|
@@ -1,23 +1,121 @@
|
|
| 1 |
import os
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
-
import inspect
|
| 5 |
import pandas as pd
|
| 6 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
# (Keep Constants as is)
|
| 8 |
# --- Constants ---
|
| 9 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 10 |
|
| 11 |
-
# ---
|
| 12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
class BasicAgent:
|
|
|
|
|
|
|
| 14 |
def __init__(self):
|
| 15 |
-
print("
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
def __call__(self, question: str) -> str:
|
| 17 |
-
print(f"Agent received question (first
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
|
| 22 |
def run_and_submit_all( profile: gr.OAuthProfile | None):
|
| 23 |
"""
|
|
@@ -146,11 +244,9 @@ with gr.Blocks() as demo:
|
|
| 146 |
gr.Markdown(
|
| 147 |
"""
|
| 148 |
**Instructions:**
|
| 149 |
-
|
| 150 |
1. Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc ...
|
| 151 |
2. Log in to your Hugging Face account using the button below. This uses your HF username for submission.
|
| 152 |
3. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
|
| 153 |
-
|
| 154 |
---
|
| 155 |
**Disclaimers:**
|
| 156 |
Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
|
|
@@ -193,4 +289,5 @@ if __name__ == "__main__":
|
|
| 193 |
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 194 |
|
| 195 |
print("Launching Gradio Interface for Basic Agent Evaluation...")
|
| 196 |
-
demo.launch(debug=True, share=False)
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
+
import sys
|
| 3 |
import gradio as gr
|
| 4 |
import requests
|
|
|
|
| 5 |
import pandas as pd
|
| 6 |
|
| 7 |
+
# Allow importing smolagents from hf-smolagents venv when running from project root
|
| 8 |
+
_root = os.path.dirname(os.path.abspath(__file__))
|
| 9 |
+
_venv_lib = os.path.join(_root, "hf-smolagents", ".venv", "lib")
|
| 10 |
+
if os.path.isdir(_venv_lib):
|
| 11 |
+
for _name in os.listdir(_venv_lib):
|
| 12 |
+
if _name.startswith("python"):
|
| 13 |
+
_sp = os.path.join(_venv_lib, _name, "site-packages")
|
| 14 |
+
if os.path.isdir(_sp):
|
| 15 |
+
sys.path.insert(0, _sp)
|
| 16 |
+
break
|
| 17 |
+
else:
|
| 18 |
+
_sp = None
|
| 19 |
+
else:
|
| 20 |
+
_sp = None
|
| 21 |
+
|
| 22 |
+
from smolagents import CodeAgent, InferenceClientModel
|
| 23 |
+
from smolagents.default_tools import (
|
| 24 |
+
DuckDuckGoSearchTool,
|
| 25 |
+
FinalAnswerTool,
|
| 26 |
+
PythonInterpreterTool,
|
| 27 |
+
UserInputTool,
|
| 28 |
+
)
|
| 29 |
+
|
| 30 |
# (Keep Constants as is)
|
| 31 |
# --- Constants ---
|
| 32 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 33 |
|
| 34 |
+
# --- Multi-Agent System (smolagents) ---
|
| 35 |
+
def _create_model():
|
| 36 |
+
token = os.environ.get("HF_TOKEN") or os.environ.get("HUGGING_FACE_HUB_TOKEN")
|
| 37 |
+
return InferenceClientModel(
|
| 38 |
+
model_id="Qwen/Qwen2.5-Coder-7B-Instruct",
|
| 39 |
+
token=token,
|
| 40 |
+
)
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
def _create_code_agent(model):
|
| 44 |
+
"""Code agent: Python interpreter + final answer. For math, calculations, data."""
|
| 45 |
+
return CodeAgent(
|
| 46 |
+
tools=[PythonInterpreterTool(), UserInputTool()],
|
| 47 |
+
model=model,
|
| 48 |
+
name="code_agent",
|
| 49 |
+
description="Use for math, calculations, data processing, or when the task can be solved by writing and running Python code. Call with a single clear task.",
|
| 50 |
+
max_steps=15,
|
| 51 |
+
)
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
def _create_web_agent(model):
|
| 55 |
+
"""Web agent: DuckDuckGo search + final answer. For factual/search tasks."""
|
| 56 |
+
return CodeAgent(
|
| 57 |
+
tools=[DuckDuckGoSearchTool(), UserInputTool()],
|
| 58 |
+
model=model,
|
| 59 |
+
name="web_agent",
|
| 60 |
+
description="Use for factual questions, current events, or when you need to search the web. Give one search task per call; you can call me multiple times with different queries.",
|
| 61 |
+
max_steps=15,
|
| 62 |
+
)
|
| 63 |
+
|
| 64 |
+
|
| 65 |
+
def _create_evaluator_agent(model):
|
| 66 |
+
"""Evaluator: picks best answer from multiple candidates, returns via final_answer."""
|
| 67 |
+
return CodeAgent(
|
| 68 |
+
tools=[FinalAnswerTool(), UserInputTool()],
|
| 69 |
+
model=model,
|
| 70 |
+
name="evaluator_agent",
|
| 71 |
+
description="Use to pick the single best answer from multiple candidate answers. Pass a task containing the original question and the list of candidate answers; return only the chosen best answer (concise, factual).",
|
| 72 |
+
max_steps=5,
|
| 73 |
+
)
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
def _create_manager_agent(model):
|
| 77 |
+
"""Manager: decides code_agent vs web_agent; if web, runs multiple web queries then evaluator then final_answer."""
|
| 78 |
+
code_agent = _create_code_agent(model)
|
| 79 |
+
web_agent = _create_web_agent(model)
|
| 80 |
+
evaluator_agent = _create_evaluator_agent(model)
|
| 81 |
+
return CodeAgent(
|
| 82 |
+
tools=[UserInputTool()],
|
| 83 |
+
model=model,
|
| 84 |
+
managed_agents=[code_agent, web_agent, evaluator_agent],
|
| 85 |
+
name="manager",
|
| 86 |
+
description="Orchestrator: decide whether to use code_agent or web_agent. For web tasks, call web_agent multiple times with different search queries, then call evaluator_agent with the question and all answers to pick the best one, then call final_answer with that result.",
|
| 87 |
+
max_steps=25,
|
| 88 |
+
planning_interval=3,
|
| 89 |
+
)
|
| 90 |
+
|
| 91 |
+
|
| 92 |
class BasicAgent:
|
| 93 |
+
"""Wrapper: builds manager-led multi-agent system and returns final answer string per question."""
|
| 94 |
+
|
| 95 |
def __init__(self):
|
| 96 |
+
print("Multi-agent system: initializing model and manager...")
|
| 97 |
+
self._model = _create_model()
|
| 98 |
+
self._manager = _create_manager_agent(self._model)
|
| 99 |
+
print("Multi-agent system initialized (manager + code_agent, web_agent, evaluator_agent).")
|
| 100 |
+
|
| 101 |
def __call__(self, question: str) -> str:
|
| 102 |
+
print(f"Agent received question (first 80 chars): {question[:80]}...")
|
| 103 |
+
try:
|
| 104 |
+
result = self._manager.run(
|
| 105 |
+
question,
|
| 106 |
+
reset=True,
|
| 107 |
+
stream=False,
|
| 108 |
+
return_full_result=False,
|
| 109 |
+
)
|
| 110 |
+
if result is None:
|
| 111 |
+
out = "No answer produced."
|
| 112 |
+
else:
|
| 113 |
+
out = str(result).strip()
|
| 114 |
+
print(f"Agent returning answer (first 80 chars): {out[:80]}...")
|
| 115 |
+
return out
|
| 116 |
+
except Exception as e:
|
| 117 |
+
print(f"Agent error: {e}")
|
| 118 |
+
return f"AGENT ERROR: {e}"
|
| 119 |
|
| 120 |
def run_and_submit_all( profile: gr.OAuthProfile | None):
|
| 121 |
"""
|
|
|
|
| 244 |
gr.Markdown(
|
| 245 |
"""
|
| 246 |
**Instructions:**
|
|
|
|
| 247 |
1. Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc ...
|
| 248 |
2. Log in to your Hugging Face account using the button below. This uses your HF username for submission.
|
| 249 |
3. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
|
|
|
|
| 250 |
---
|
| 251 |
**Disclaimers:**
|
| 252 |
Once clicking on the "submit button, it can take quite some time ( this is the time for the agent to go through all the questions).
|
|
|
|
| 289 |
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 290 |
|
| 291 |
print("Launching Gradio Interface for Basic Agent Evaluation...")
|
| 292 |
+
demo.launch(debug=True, share=False)
|
| 293 |
+
|