Update app.py
Browse files
app.py
CHANGED
@@ -1,25 +1,21 @@
|
|
1 |
-
import os
|
2 |
-
import gradio as gr
|
3 |
-
import requests
|
4 |
-
import pandas as pd
|
5 |
-
import re
|
6 |
-
|
7 |
-
# --- Constants ---
|
8 |
-
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
9 |
-
|
10 |
-
# --- LangChain Agent Definition ---
|
11 |
from langchain_huggingface import HuggingFaceEndpoint
|
12 |
from langchain.agents import initialize_agent, Tool
|
13 |
from langchain.agents.agent_types import AgentType
|
14 |
from langchain.tools import tool
|
15 |
from duckduckgo_search import DDGS
|
|
|
16 |
|
17 |
@tool
|
18 |
def search_web(query: str) -> str:
|
19 |
"""Search the web and return a relevant text snippet."""
|
20 |
-
|
21 |
-
|
22 |
-
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
@tool
|
25 |
def run_code_snippet(code: str) -> str:
|
@@ -30,15 +26,15 @@ def run_code_snippet(code: str) -> str:
|
|
30 |
result = [v for v in local_env.values() if isinstance(v, (int, float, str))]
|
31 |
return str(result[0]) if result else "0"
|
32 |
except Exception as e:
|
33 |
-
return f"Error: {str(e)}"
|
34 |
|
35 |
class GAIAAgent:
|
36 |
def __init__(self):
|
37 |
print("Initializing LangChain agent with Hugging FaceEndpoint...")
|
38 |
self.llm = HuggingFaceEndpoint(
|
39 |
-
repo_id="google/flan-t5-xl",
|
40 |
temperature=0.3,
|
41 |
-
|
42 |
)
|
43 |
self.tools = [search_web, run_code_snippet]
|
44 |
self.agent = initialize_agent(
|
@@ -50,127 +46,22 @@ class GAIAAgent:
|
|
50 |
|
51 |
def __call__(self, question: str) -> str:
|
52 |
try:
|
|
|
53 |
result = self.agent.invoke(question)
|
|
|
|
|
|
|
|
|
|
|
54 |
return f"FINAL ANSWER: {self.clean_answer(result)}"
|
|
|
|
|
|
|
|
|
55 |
except Exception as e:
|
|
|
56 |
return f"FINAL ANSWER: Error: {e}"
|
57 |
|
58 |
def clean_answer(self, ans: str) -> str:
|
59 |
ans = re.sub(r"[\$,%]", "", ans)
|
60 |
return ans.strip()
|
61 |
-
|
62 |
-
# --- Evaluation & Submission Pipeline ---
|
63 |
-
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
64 |
-
space_id = os.getenv("SPACE_ID")
|
65 |
-
|
66 |
-
if profile:
|
67 |
-
username = f"{profile.username}"
|
68 |
-
print(f"User logged in: {username}")
|
69 |
-
else:
|
70 |
-
return "Please Login to Hugging Face with the button.", None
|
71 |
-
|
72 |
-
api_url = DEFAULT_API_URL
|
73 |
-
questions_url = f"{api_url}/questions"
|
74 |
-
submit_url = f"{api_url}/submit"
|
75 |
-
|
76 |
-
try:
|
77 |
-
agent = GAIAAgent()
|
78 |
-
except Exception as e:
|
79 |
-
return f"Error initializing agent: {e}", None
|
80 |
-
|
81 |
-
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
82 |
-
print(agent_code)
|
83 |
-
|
84 |
-
try:
|
85 |
-
response = requests.get(questions_url, timeout=15)
|
86 |
-
response.raise_for_status()
|
87 |
-
questions_data = response.json()
|
88 |
-
if not questions_data:
|
89 |
-
return "Fetched questions list is empty or invalid format.", None
|
90 |
-
print(f"Fetched {len(questions_data)} questions.")
|
91 |
-
except Exception as e:
|
92 |
-
return f"Error fetching questions: {e}", None
|
93 |
-
|
94 |
-
results_log = []
|
95 |
-
answers_payload = []
|
96 |
-
for item in questions_data:
|
97 |
-
task_id = item.get("task_id")
|
98 |
-
question_text = item.get("question")
|
99 |
-
if not task_id or question_text is None:
|
100 |
-
continue
|
101 |
-
try:
|
102 |
-
submitted_answer = agent(question_text)
|
103 |
-
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
104 |
-
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
105 |
-
except Exception as e:
|
106 |
-
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
|
107 |
-
|
108 |
-
if not answers_payload:
|
109 |
-
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
110 |
-
|
111 |
-
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
112 |
-
try:
|
113 |
-
response = requests.post(submit_url, json=submission_data, timeout=60)
|
114 |
-
response.raise_for_status()
|
115 |
-
result_data = response.json()
|
116 |
-
final_status = (
|
117 |
-
f"Submission Successful!\n"
|
118 |
-
f"User: {result_data.get('username')}\n"
|
119 |
-
f"Overall Score: {result_data.get('score', 'N/A')}% "
|
120 |
-
f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
|
121 |
-
f"Message: {result_data.get('message', 'No message received.')}"
|
122 |
-
)
|
123 |
-
return final_status, pd.DataFrame(results_log)
|
124 |
-
except Exception as e:
|
125 |
-
return f"Submission Failed: {e}", pd.DataFrame(results_log)
|
126 |
-
|
127 |
-
# --- Gradio UI ---
|
128 |
-
with gr.Blocks() as demo:
|
129 |
-
gr.Markdown("# GAIA Agent Evaluation Runner")
|
130 |
-
|
131 |
-
gr.Markdown("""
|
132 |
-
**Instructions:**
|
133 |
-
1. Set your Hugging Face token in the Space Settings → Secrets as `HUGGINGFACEHUB_API_TOKEN`.
|
134 |
-
2. Log in to Hugging Face below to associate your username.
|
135 |
-
3. Run the agent to fetch, answer, and submit GAIA benchmark questions.
|
136 |
-
""")
|
137 |
-
|
138 |
-
gr.LoginButton()
|
139 |
-
|
140 |
-
gr.Markdown("## 🔍 Try a Question Preview")
|
141 |
-
preview_input = gr.Textbox(label="Your GAIA-style question")
|
142 |
-
preview_output = gr.Textbox(label="Agent's Response", lines=2)
|
143 |
-
preview_button = gr.Button("Preview Answer")
|
144 |
-
preview_button.click(lambda q: GAIAAgent()(q), inputs=preview_input, outputs=preview_output)
|
145 |
-
|
146 |
-
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
147 |
-
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
148 |
-
results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
|
149 |
-
|
150 |
-
run_button.click(
|
151 |
-
fn=run_and_submit_all,
|
152 |
-
outputs=[status_output, results_table]
|
153 |
-
)
|
154 |
-
|
155 |
-
# --- Main Entry ---
|
156 |
-
if __name__ == "__main__":
|
157 |
-
print("\n" + "-"*30 + " App Starting " + "-"*30)
|
158 |
-
space_host_startup = os.getenv("SPACE_HOST")
|
159 |
-
space_id_startup = os.getenv("SPACE_ID")
|
160 |
-
|
161 |
-
if space_host_startup:
|
162 |
-
print(f"✅ SPACE_HOST found: {space_host_startup}")
|
163 |
-
print(f" Runtime URL: https://{space_host_startup}.hf.space")
|
164 |
-
else:
|
165 |
-
print("ℹ️ SPACE_HOST not set.")
|
166 |
-
|
167 |
-
if space_id_startup:
|
168 |
-
print(f"✅ SPACE_ID found: {space_id_startup}")
|
169 |
-
print(f" Repo: https://huggingface.co/spaces/{space_id_startup}")
|
170 |
-
print(f" Code: https://huggingface.co/spaces/{space_id_startup}/tree/main")
|
171 |
-
else:
|
172 |
-
print("ℹ️ SPACE_ID not set.")
|
173 |
-
|
174 |
-
print("-"*(60 + len(" App Starting ")) + "\n")
|
175 |
-
print("Launching Gradio Interface for GAIA Agent Evaluation...")
|
176 |
-
demo.launch(debug=True, share=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
from langchain_huggingface import HuggingFaceEndpoint
|
2 |
from langchain.agents import initialize_agent, Tool
|
3 |
from langchain.agents.agent_types import AgentType
|
4 |
from langchain.tools import tool
|
5 |
from duckduckgo_search import DDGS
|
6 |
+
import re
|
7 |
|
8 |
@tool
|
9 |
def search_web(query: str) -> str:
|
10 |
"""Search the web and return a relevant text snippet."""
|
11 |
+
try:
|
12 |
+
with DDGS() as ddgs:
|
13 |
+
results = list(ddgs.text(query, max_results=3))
|
14 |
+
if not results or not results[0].get("body"):
|
15 |
+
return "No result found"
|
16 |
+
return results[0]["body"]
|
17 |
+
except Exception as e:
|
18 |
+
return f"Web search failed: {e}"
|
19 |
|
20 |
@tool
|
21 |
def run_code_snippet(code: str) -> str:
|
|
|
26 |
result = [v for v in local_env.values() if isinstance(v, (int, float, str))]
|
27 |
return str(result[0]) if result else "0"
|
28 |
except Exception as e:
|
29 |
+
return f"Error in code: {str(e)}"
|
30 |
|
31 |
class GAIAAgent:
|
32 |
def __init__(self):
|
33 |
print("Initializing LangChain agent with Hugging FaceEndpoint...")
|
34 |
self.llm = HuggingFaceEndpoint(
|
35 |
+
repo_id="google/flan-t5-xl",
|
36 |
temperature=0.3,
|
37 |
+
model_kwargs={"max_length": 256} # ✅ fix: max_length passed safely
|
38 |
)
|
39 |
self.tools = [search_web, run_code_snippet]
|
40 |
self.agent = initialize_agent(
|
|
|
46 |
|
47 |
def __call__(self, question: str) -> str:
|
48 |
try:
|
49 |
+
print(f"[Agent] Invoking on question: {question}")
|
50 |
result = self.agent.invoke(question)
|
51 |
+
print(f"[Agent] Raw result: {result}")
|
52 |
+
|
53 |
+
if not result or not isinstance(result, str):
|
54 |
+
raise StopIteration("Model returned empty or non-string result.")
|
55 |
+
|
56 |
return f"FINAL ANSWER: {self.clean_answer(result)}"
|
57 |
+
|
58 |
+
except StopIteration as stop_err:
|
59 |
+
print(f"[Agent] StopIteration: {stop_err}")
|
60 |
+
return "FINAL ANSWER: Error: model or tool returned no result"
|
61 |
except Exception as e:
|
62 |
+
print(f"[Agent] Exception: {e}")
|
63 |
return f"FINAL ANSWER: Error: {e}"
|
64 |
|
65 |
def clean_answer(self, ans: str) -> str:
|
66 |
ans = re.sub(r"[\$,%]", "", ans)
|
67 |
return ans.strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|