Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -2,7 +2,8 @@ import os
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
| 5 |
-
|
|
|
|
| 6 |
|
| 7 |
# --- Constants ---
|
| 8 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
|
@@ -14,31 +15,25 @@ class AgentArchitect:
|
|
| 14 |
hf_token = os.getenv("HF_TOKEN")
|
| 15 |
|
| 16 |
if not hf_token:
|
| 17 |
-
print("Warning: HF_TOKEN is missing
|
| 18 |
|
| 19 |
-
#
|
| 20 |
-
#
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
token=hf_token
|
| 25 |
)
|
| 26 |
|
| 27 |
-
# Tools are the agent's 'hands'
|
| 28 |
-
# DuckDuckGo for search, VisitWebpage for reading deep into sites
|
| 29 |
self.tools = [DuckDuckGoSearchTool(), VisitWebpageTool()]
|
| 30 |
|
| 31 |
-
# CodeAgent is the 'Brain' - it can write Python code to solve problems
|
| 32 |
self.agent = CodeAgent(
|
| 33 |
tools=self.tools,
|
| 34 |
model=self.model,
|
| 35 |
-
add_base_tools=True
|
| 36 |
)
|
| 37 |
|
| 38 |
def __call__(self, question: str) -> str:
|
| 39 |
try:
|
| 40 |
-
# We enforce conciseness. The GAIA benchmark grades on 'Exact Match'.
|
| 41 |
-
# If the answer is 'India' and the agent says 'The winner is India', it fails.
|
| 42 |
prompt = (
|
| 43 |
f"{question}\n\n"
|
| 44 |
f"Instructions: Think step-by-step. Use tools if needed. "
|
|
@@ -47,7 +42,6 @@ class AgentArchitect:
|
|
| 47 |
result = self.agent.run(prompt)
|
| 48 |
return str(result)
|
| 49 |
except Exception as e:
|
| 50 |
-
print(f"Agent Runtime Error: {e}")
|
| 51 |
return f"Error: {e}"
|
| 52 |
|
| 53 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
@@ -62,11 +56,9 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 62 |
submit_url = f"{api_url}/submit"
|
| 63 |
|
| 64 |
try:
|
| 65 |
-
# Initialize the Agent
|
| 66 |
agent_instance = AgentArchitect()
|
| 67 |
|
| 68 |
-
# 1. Fetch Questions
|
| 69 |
-
print("Fetching questions...")
|
| 70 |
response = requests.get(questions_url, timeout=15)
|
| 71 |
response.raise_for_status()
|
| 72 |
questions_data = response.json()
|
|
@@ -74,19 +66,17 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 74 |
results_log = []
|
| 75 |
answers_payload = []
|
| 76 |
|
| 77 |
-
# 2.
|
| 78 |
-
# This will take 10-15 minutes to finish all 20.
|
| 79 |
for item in questions_data:
|
| 80 |
task_id = item.get("task_id")
|
| 81 |
question_text = item.get("question")
|
| 82 |
|
| 83 |
-
print(f"Processing Task: {task_id}")
|
| 84 |
submitted_answer = agent_instance(question_text)
|
| 85 |
|
| 86 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 87 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 88 |
|
| 89 |
-
# 3. Submit
|
| 90 |
agent_code_link = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 91 |
submission_data = {
|
| 92 |
"username": username.strip(),
|
|
@@ -94,7 +84,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 94 |
"answers": answers_payload
|
| 95 |
}
|
| 96 |
|
| 97 |
-
print("Submitting answers...")
|
| 98 |
submit_response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 99 |
submit_response.raise_for_status()
|
| 100 |
result_data = submit_response.json()
|
|
@@ -113,18 +102,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 113 |
# --- Gradio UI ---
|
| 114 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
| 115 |
gr.Markdown("# 🚀 Professional Agent Evaluator")
|
| 116 |
-
gr.Markdown("Click 'Login' first, then click the 'Run' button. This may take up to 15 minutes.")
|
| 117 |
-
|
| 118 |
gr.LoginButton()
|
| 119 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
| 120 |
-
|
| 121 |
status_output = gr.Textbox(label="Status", lines=4)
|
| 122 |
results_table = gr.DataFrame(label="Agent Reasoning Trace", wrap=True)
|
| 123 |
|
| 124 |
-
run_button.click(
|
| 125 |
-
fn=run_and_submit_all,
|
| 126 |
-
outputs=[status_output, results_table]
|
| 127 |
-
)
|
| 128 |
|
| 129 |
if __name__ == "__main__":
|
| 130 |
demo.launch()
|
|
|
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
import pandas as pd
|
| 5 |
+
# Use LiteLLMModel instead of HfApiModel to avoid the import error
|
| 6 |
+
from smolagents import CodeAgent, DuckDuckGoSearchTool, LiteLLMModel, VisitWebpageTool
|
| 7 |
|
| 8 |
# --- Constants ---
|
| 9 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
|
|
|
| 15 |
hf_token = os.getenv("HF_TOKEN")
|
| 16 |
|
| 17 |
if not hf_token:
|
| 18 |
+
print("Warning: HF_TOKEN is missing in Space Secrets.")
|
| 19 |
|
| 20 |
+
# LiteLLMModel is the more robust class in the latest smolagents
|
| 21 |
+
# We add the 'huggingface/' prefix to tell it exactly where to go
|
| 22 |
+
self.model = LiteLLMModel(
|
| 23 |
+
model_id="huggingface/Qwen/Qwen2.5-Coder-32B-Instruct",
|
| 24 |
+
api_key=hf_token
|
|
|
|
| 25 |
)
|
| 26 |
|
|
|
|
|
|
|
| 27 |
self.tools = [DuckDuckGoSearchTool(), VisitWebpageTool()]
|
| 28 |
|
|
|
|
| 29 |
self.agent = CodeAgent(
|
| 30 |
tools=self.tools,
|
| 31 |
model=self.model,
|
| 32 |
+
add_base_tools=True
|
| 33 |
)
|
| 34 |
|
| 35 |
def __call__(self, question: str) -> str:
|
| 36 |
try:
|
|
|
|
|
|
|
| 37 |
prompt = (
|
| 38 |
f"{question}\n\n"
|
| 39 |
f"Instructions: Think step-by-step. Use tools if needed. "
|
|
|
|
| 42 |
result = self.agent.run(prompt)
|
| 43 |
return str(result)
|
| 44 |
except Exception as e:
|
|
|
|
| 45 |
return f"Error: {e}"
|
| 46 |
|
| 47 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
|
|
| 56 |
submit_url = f"{api_url}/submit"
|
| 57 |
|
| 58 |
try:
|
|
|
|
| 59 |
agent_instance = AgentArchitect()
|
| 60 |
|
| 61 |
+
# 1. Fetch Questions
|
|
|
|
| 62 |
response = requests.get(questions_url, timeout=15)
|
| 63 |
response.raise_for_status()
|
| 64 |
questions_data = response.json()
|
|
|
|
| 66 |
results_log = []
|
| 67 |
answers_payload = []
|
| 68 |
|
| 69 |
+
# 2. Run Agent
|
|
|
|
| 70 |
for item in questions_data:
|
| 71 |
task_id = item.get("task_id")
|
| 72 |
question_text = item.get("question")
|
| 73 |
|
|
|
|
| 74 |
submitted_answer = agent_instance(question_text)
|
| 75 |
|
| 76 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 77 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 78 |
|
| 79 |
+
# 3. Submit
|
| 80 |
agent_code_link = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 81 |
submission_data = {
|
| 82 |
"username": username.strip(),
|
|
|
|
| 84 |
"answers": answers_payload
|
| 85 |
}
|
| 86 |
|
|
|
|
| 87 |
submit_response = requests.post(submit_url, json=submission_data, timeout=60)
|
| 88 |
submit_response.raise_for_status()
|
| 89 |
result_data = submit_response.json()
|
|
|
|
| 102 |
# --- Gradio UI ---
|
| 103 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
| 104 |
gr.Markdown("# 🚀 Professional Agent Evaluator")
|
|
|
|
|
|
|
| 105 |
gr.LoginButton()
|
| 106 |
run_button = gr.Button("Run Evaluation & Submit All Answers", variant="primary")
|
|
|
|
| 107 |
status_output = gr.Textbox(label="Status", lines=4)
|
| 108 |
results_table = gr.DataFrame(label="Agent Reasoning Trace", wrap=True)
|
| 109 |
|
| 110 |
+
run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
|
|
|
|
|
|
|
|
|
|
| 111 |
|
| 112 |
if __name__ == "__main__":
|
| 113 |
demo.launch()
|