Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -4,20 +4,19 @@ import os
|
|
4 |
import gradio as gr
|
5 |
import requests
|
6 |
import pandas as pd
|
7 |
-
|
8 |
import google.generativeai as genai
|
9 |
from smolagents import CodeAgent, DuckDuckGoSearchTool
|
10 |
|
11 |
-
# System prompt
|
12 |
SYSTEM_PROMPT = """You are a general AI assistant. I will ask you a question.
|
13 |
-
Report your thoughts, and finish your answer with just the answer — no prefixes like "FINAL ANSWER
|
14 |
Your answer should be a number OR as few words as possible OR a comma-separated list of numbers and/or strings.
|
15 |
If you're asked for a number, don’t use commas or units like $ or %, unless specified.
|
16 |
If you're asked for a string, don’t use articles or abbreviations (e.g. for cities), and write digits in plain text unless told otherwise."""
|
17 |
|
18 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
19 |
|
20 |
-
# Generation result wrapper
|
21 |
class GenerationResult:
|
22 |
def __init__(self, content, token_usage=None, input_tokens=0, output_tokens=0):
|
23 |
self.content = content
|
@@ -47,32 +46,31 @@ class GeminiFlashModel:
|
|
47 |
|
48 |
try:
|
49 |
response = self.model.generate_content(prompt)
|
50 |
-
text_output = response.text.strip() if hasattr(response, "text") else str(response)
|
51 |
-
|
52 |
return GenerationResult(
|
53 |
-
content=
|
54 |
-
token_usage={
|
55 |
input_tokens=0,
|
56 |
output_tokens=0
|
57 |
)
|
58 |
except Exception as e:
|
59 |
return GenerationResult(
|
60 |
content=f"GENERATION ERROR: {e}",
|
61 |
-
token_usage={
|
62 |
input_tokens=0,
|
63 |
output_tokens=0
|
64 |
)
|
65 |
|
66 |
-
# Agent
|
67 |
class MyAgent:
|
68 |
def __init__(self):
|
69 |
-
self.model = GeminiFlashModel(
|
70 |
self.agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=self.model)
|
71 |
|
72 |
-
def __call__(self, question: str)
|
73 |
return self.agent.run(question)
|
74 |
|
75 |
-
#
|
|
|
76 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
77 |
space_id = os.getenv("SPACE_ID")
|
78 |
|
@@ -106,10 +104,28 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
106 |
question_text = item.get("question")
|
107 |
if not task_id or question_text is None:
|
108 |
continue
|
|
|
109 |
try:
|
110 |
-
|
111 |
-
|
112 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
except Exception as e:
|
114 |
results_log.append({
|
115 |
"Task ID": task_id,
|
@@ -141,7 +157,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
141 |
except Exception as e:
|
142 |
return f"Submission failed: {e}", pd.DataFrame(results_log)
|
143 |
|
144 |
-
#
|
145 |
with gr.Blocks() as demo:
|
146 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
147 |
gr.Markdown("""
|
@@ -159,7 +175,7 @@ with gr.Blocks() as demo:
|
|
159 |
run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
|
160 |
|
161 |
if __name__ == "__main__":
|
162 |
-
print("
|
163 |
demo.launch(debug=True, share=False)
|
164 |
|
165 |
|
|
|
4 |
import gradio as gr
|
5 |
import requests
|
6 |
import pandas as pd
|
|
|
7 |
import google.generativeai as genai
|
8 |
from smolagents import CodeAgent, DuckDuckGoSearchTool
|
9 |
|
10 |
+
# System prompt for Gemini
|
11 |
SYSTEM_PROMPT = """You are a general AI assistant. I will ask you a question.
|
12 |
+
Report your thoughts, and finish your answer with just the answer — no prefixes like \"FINAL ANSWER:\".
|
13 |
Your answer should be a number OR as few words as possible OR a comma-separated list of numbers and/or strings.
|
14 |
If you're asked for a number, don’t use commas or units like $ or %, unless specified.
|
15 |
If you're asked for a string, don’t use articles or abbreviations (e.g. for cities), and write digits in plain text unless told otherwise."""
|
16 |
|
17 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
18 |
|
19 |
+
# Generation result wrapper
|
20 |
class GenerationResult:
|
21 |
def __init__(self, content, token_usage=None, input_tokens=0, output_tokens=0):
|
22 |
self.content = content
|
|
|
46 |
|
47 |
try:
|
48 |
response = self.model.generate_content(prompt)
|
|
|
|
|
49 |
return GenerationResult(
|
50 |
+
content=response.text.strip(),
|
51 |
+
token_usage={},
|
52 |
input_tokens=0,
|
53 |
output_tokens=0
|
54 |
)
|
55 |
except Exception as e:
|
56 |
return GenerationResult(
|
57 |
content=f"GENERATION ERROR: {e}",
|
58 |
+
token_usage={},
|
59 |
input_tokens=0,
|
60 |
output_tokens=0
|
61 |
)
|
62 |
|
63 |
+
# Agent class
|
64 |
class MyAgent:
|
65 |
def __init__(self):
|
66 |
+
self.model = GeminiFlashModel()
|
67 |
self.agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=self.model)
|
68 |
|
69 |
+
def __call__(self, question: str):
|
70 |
return self.agent.run(question)
|
71 |
|
72 |
+
# Evaluation function
|
73 |
+
|
74 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
75 |
space_id = os.getenv("SPACE_ID")
|
76 |
|
|
|
104 |
question_text = item.get("question")
|
105 |
if not task_id or question_text is None:
|
106 |
continue
|
107 |
+
|
108 |
try:
|
109 |
+
result = agent(question_text)
|
110 |
+
|
111 |
+
if hasattr(result, "content"):
|
112 |
+
submitted_answer = result.content
|
113 |
+
elif isinstance(result, dict) and "content" in result:
|
114 |
+
submitted_answer = result["content"]
|
115 |
+
else:
|
116 |
+
submitted_answer = str(result)
|
117 |
+
|
118 |
+
answers_payload.append({
|
119 |
+
"task_id": task_id,
|
120 |
+
"submitted_answer": submitted_answer
|
121 |
+
})
|
122 |
+
|
123 |
+
results_log.append({
|
124 |
+
"Task ID": task_id,
|
125 |
+
"Question": question_text,
|
126 |
+
"Submitted Answer": submitted_answer
|
127 |
+
})
|
128 |
+
|
129 |
except Exception as e:
|
130 |
results_log.append({
|
131 |
"Task ID": task_id,
|
|
|
157 |
except Exception as e:
|
158 |
return f"Submission failed: {e}", pd.DataFrame(results_log)
|
159 |
|
160 |
+
# UI setup
|
161 |
with gr.Blocks() as demo:
|
162 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
163 |
gr.Markdown("""
|
|
|
175 |
run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
|
176 |
|
177 |
if __name__ == "__main__":
|
178 |
+
print("\ud83d\udd27 App starting...")
|
179 |
demo.launch(debug=True, share=False)
|
180 |
|
181 |
|