Update app.py
Browse files
app.py
CHANGED
|
@@ -1,33 +1,40 @@
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
-
import
|
| 5 |
-
|
|
|
|
| 6 |
from smolagents import CodeAgent, HfApiModel
|
|
|
|
|
|
|
| 7 |
|
| 8 |
# === Constants ===
|
| 9 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 10 |
|
| 11 |
-
# ===
|
| 12 |
class BasicAgent:
|
| 13 |
def __init__(self):
|
| 14 |
-
model = HfApiModel(
|
|
|
|
|
|
|
| 15 |
|
| 16 |
self.agent = CodeAgent(
|
| 17 |
model=model,
|
| 18 |
-
tools=[
|
| 19 |
-
|
|
|
|
|
|
|
| 20 |
max_steps=6,
|
| 21 |
-
name="
|
| 22 |
-
description="Agent using
|
| 23 |
)
|
| 24 |
-
print("✅ BasicAgent initialized with
|
| 25 |
|
| 26 |
def __call__(self, question: str) -> str:
|
| 27 |
return self.agent.run(question)
|
| 28 |
|
| 29 |
|
| 30 |
-
# === Evaluation Logic (
|
| 31 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 32 |
space_id = os.getenv("SPACE_ID")
|
| 33 |
|
|
@@ -38,6 +45,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 38 |
print("User not logged in.")
|
| 39 |
return "Please Login to Hugging Face with the button.", None
|
| 40 |
|
|
|
|
| 41 |
api_url = DEFAULT_API_URL
|
| 42 |
questions_url = f"{api_url}/questions"
|
| 43 |
submit_url = f"{api_url}/submit"
|
|
@@ -45,50 +53,48 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 45 |
try:
|
| 46 |
agent = BasicAgent()
|
| 47 |
except Exception as e:
|
| 48 |
-
print(f"Error instantiating agent: {e}")
|
| 49 |
return f"Error initializing agent: {e}", None
|
| 50 |
|
| 51 |
-
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 52 |
-
print(agent_code)
|
| 53 |
-
|
| 54 |
-
print(f"Fetching questions from: {questions_url}")
|
| 55 |
try:
|
| 56 |
response = requests.get(questions_url, timeout=15)
|
| 57 |
response.raise_for_status()
|
| 58 |
questions_data = response.json()
|
| 59 |
if not questions_data:
|
| 60 |
-
print("Fetched questions list is empty.")
|
| 61 |
return "Fetched questions list is empty or invalid format.", None
|
| 62 |
-
print(f"Fetched {len(questions_data)} questions.")
|
| 63 |
-
except requests.exceptions.RequestException as e:
|
| 64 |
-
print(f"Error fetching questions: {e}")
|
| 65 |
-
return f"Error fetching questions: {e}", None
|
| 66 |
-
except requests.exceptions.JSONDecodeError as e:
|
| 67 |
-
print(f"Error decoding JSON response: {e}")
|
| 68 |
-
return f"Error decoding server response: {e}", None
|
| 69 |
except Exception as e:
|
| 70 |
-
|
| 71 |
-
return f"Unexpected error fetching questions: {e}", None
|
| 72 |
|
| 73 |
results_log = []
|
| 74 |
answers_payload = []
|
|
|
|
| 75 |
for item in questions_data:
|
| 76 |
task_id = item.get("task_id")
|
| 77 |
question_text = item.get("question")
|
| 78 |
-
if not task_id or
|
| 79 |
continue
|
| 80 |
try:
|
| 81 |
submitted_answer = agent(question_text)
|
| 82 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 83 |
-
results_log.append({
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
except Exception as e:
|
| 85 |
-
results_log.append({
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
|
| 87 |
if not answers_payload:
|
| 88 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 89 |
|
| 90 |
-
submission_data = {
|
| 91 |
-
|
|
|
|
|
|
|
|
|
|
| 92 |
|
| 93 |
try:
|
| 94 |
response = requests.post(submit_url, json=submission_data, timeout=60)
|
|
@@ -106,7 +112,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 106 |
return f"Submission Failed: {e}", pd.DataFrame(results_log)
|
| 107 |
|
| 108 |
|
| 109 |
-
# === Gradio UI (
|
| 110 |
with gr.Blocks() as demo:
|
| 111 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 112 |
gr.Markdown("""
|
|
@@ -125,14 +131,13 @@ with gr.Blocks() as demo:
|
|
| 125 |
|
| 126 |
if __name__ == "__main__":
|
| 127 |
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
if
|
| 131 |
-
print(f"✅ SPACE_HOST found: {
|
| 132 |
-
print(f" Runtime URL: https://{
|
| 133 |
-
if
|
| 134 |
-
print(f"✅ SPACE_ID found: {
|
| 135 |
-
print(f" Repo URL: https://huggingface.co/spaces/{
|
| 136 |
-
print(f" Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
| 137 |
print("Launching Gradio Interface...")
|
| 138 |
demo.launch(debug=True, share=False)
|
|
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
+
import pandas as pd # Required for result table formatting
|
| 5 |
+
|
| 6 |
+
# === Hugging Face SmolAgents imports ===
|
| 7 |
from smolagents import CodeAgent, HfApiModel
|
| 8 |
+
from smolagents.tools.web_search import DuckDuckGoSearchTool
|
| 9 |
+
from smolagents.tools.vision import ImageCaptionTool
|
| 10 |
|
| 11 |
# === Constants ===
|
| 12 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 13 |
|
| 14 |
+
# === Agent Definition with 2 Tools ===
|
| 15 |
class BasicAgent:
|
| 16 |
def __init__(self):
|
| 17 |
+
model = HfApiModel(
|
| 18 |
+
model_id="Qwen/Qwen2.5-Coder-32B-Instruct"
|
| 19 |
+
)
|
| 20 |
|
| 21 |
self.agent = CodeAgent(
|
| 22 |
model=model,
|
| 23 |
+
tools=[
|
| 24 |
+
DuckDuckGoSearchTool(),
|
| 25 |
+
ImageCaptionTool()
|
| 26 |
+
],
|
| 27 |
max_steps=6,
|
| 28 |
+
name="EvalAgent",
|
| 29 |
+
description="Agent using DuckDuckGoSearchTool and ImageCaptionTool for HF certification"
|
| 30 |
)
|
| 31 |
+
print("✅ BasicAgent initialized with 2 certified tools.")
|
| 32 |
|
| 33 |
def __call__(self, question: str) -> str:
|
| 34 |
return self.agent.run(question)
|
| 35 |
|
| 36 |
|
| 37 |
+
# === Evaluation Logic (Unmodified) ===
|
| 38 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 39 |
space_id = os.getenv("SPACE_ID")
|
| 40 |
|
|
|
|
| 45 |
print("User not logged in.")
|
| 46 |
return "Please Login to Hugging Face with the button.", None
|
| 47 |
|
| 48 |
+
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 49 |
api_url = DEFAULT_API_URL
|
| 50 |
questions_url = f"{api_url}/questions"
|
| 51 |
submit_url = f"{api_url}/submit"
|
|
|
|
| 53 |
try:
|
| 54 |
agent = BasicAgent()
|
| 55 |
except Exception as e:
|
|
|
|
| 56 |
return f"Error initializing agent: {e}", None
|
| 57 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
try:
|
| 59 |
response = requests.get(questions_url, timeout=15)
|
| 60 |
response.raise_for_status()
|
| 61 |
questions_data = response.json()
|
| 62 |
if not questions_data:
|
|
|
|
| 63 |
return "Fetched questions list is empty or invalid format.", None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
except Exception as e:
|
| 65 |
+
return f"Error fetching questions: {e}", None
|
|
|
|
| 66 |
|
| 67 |
results_log = []
|
| 68 |
answers_payload = []
|
| 69 |
+
|
| 70 |
for item in questions_data:
|
| 71 |
task_id = item.get("task_id")
|
| 72 |
question_text = item.get("question")
|
| 73 |
+
if not task_id or not question_text:
|
| 74 |
continue
|
| 75 |
try:
|
| 76 |
submitted_answer = agent(question_text)
|
| 77 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 78 |
+
results_log.append({
|
| 79 |
+
"Task ID": task_id,
|
| 80 |
+
"Question": question_text,
|
| 81 |
+
"Submitted Answer": submitted_answer
|
| 82 |
+
})
|
| 83 |
except Exception as e:
|
| 84 |
+
results_log.append({
|
| 85 |
+
"Task ID": task_id,
|
| 86 |
+
"Question": question_text,
|
| 87 |
+
"Submitted Answer": f"AGENT ERROR: {e}"
|
| 88 |
+
})
|
| 89 |
|
| 90 |
if not answers_payload:
|
| 91 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 92 |
|
| 93 |
+
submission_data = {
|
| 94 |
+
"username": username.strip(),
|
| 95 |
+
"agent_code": agent_code,
|
| 96 |
+
"answers": answers_payload
|
| 97 |
+
}
|
| 98 |
|
| 99 |
try:
|
| 100 |
response = requests.post(submit_url, json=submission_data, timeout=60)
|
|
|
|
| 112 |
return f"Submission Failed: {e}", pd.DataFrame(results_log)
|
| 113 |
|
| 114 |
|
| 115 |
+
# === Gradio UI (Unmodified) ===
|
| 116 |
with gr.Blocks() as demo:
|
| 117 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 118 |
gr.Markdown("""
|
|
|
|
| 131 |
|
| 132 |
if __name__ == "__main__":
|
| 133 |
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
| 134 |
+
space_host = os.getenv("SPACE_HOST")
|
| 135 |
+
space_id = os.getenv("SPACE_ID")
|
| 136 |
+
if space_host:
|
| 137 |
+
print(f"✅ SPACE_HOST found: {space_host}")
|
| 138 |
+
print(f" Runtime URL: https://{space_host}.hf.space")
|
| 139 |
+
if space_id:
|
| 140 |
+
print(f"✅ SPACE_ID found: {space_id}")
|
| 141 |
+
print(f" Repo URL: https://huggingface.co/spaces/{space_id}")
|
|
|
|
| 142 |
print("Launching Gradio Interface...")
|
| 143 |
demo.launch(debug=True, share=False)
|