Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,10 +4,8 @@ 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 |
-
from smolagents.model.base import ModelOutput # import ModelOutput if available
|
| 11 |
|
| 12 |
# System prompt used by the agent
|
| 13 |
SYSTEM_PROMPT = """You are a general AI assistant. I will ask you a question.
|
|
@@ -18,64 +16,36 @@ If you're asked for a string, don’t use articles or abbreviations (e.g. for ci
|
|
| 18 |
|
| 19 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 20 |
|
| 21 |
-
# Gemini model wrapper
|
| 22 |
class GeminiFlashModel:
|
| 23 |
-
def __init__(self, model_id="gemini-1.5-flash"
|
| 24 |
-
|
| 25 |
-
|
|
|
|
|
|
|
|
|
|
| 26 |
self.system_prompt = SYSTEM_PROMPT
|
| 27 |
|
| 28 |
-
def generate(self, messages
|
| 29 |
-
|
| 30 |
-
raise TypeError("Expected 'messages' to be a list of dicts")
|
| 31 |
-
|
| 32 |
if not any(m.get("role") == "system" for m in messages):
|
| 33 |
messages = [{"role": "system", "content": self.system_prompt}] + messages
|
| 34 |
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
|
|
|
| 40 |
|
| 41 |
-
try:
|
| 42 |
-
response = self.model.generate_content(prompt)
|
| 43 |
-
return ModelOutput(
|
| 44 |
-
content=response.text.strip(),
|
| 45 |
-
input_tokens=0,
|
| 46 |
-
output_tokens=0,
|
| 47 |
-
token_usage={}
|
| 48 |
-
)
|
| 49 |
-
except Exception as e:
|
| 50 |
-
return ModelOutput(
|
| 51 |
-
content=f"GENERATION ERROR: {e}",
|
| 52 |
-
input_tokens=0,
|
| 53 |
-
output_tokens=0,
|
| 54 |
-
token_usage={}
|
| 55 |
-
)
|
| 56 |
-
|
| 57 |
-
# Agent wrapper
|
| 58 |
class MyAgent:
|
| 59 |
def __init__(self):
|
| 60 |
-
self.model = GeminiFlashModel(
|
| 61 |
self.agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=self.model)
|
| 62 |
|
| 63 |
def __call__(self, question: str) -> str:
|
| 64 |
-
|
| 65 |
-
print(f"[DEBUG] Agent run result type: {type(result)}; value: {result}")
|
| 66 |
-
|
| 67 |
-
# Return string content only
|
| 68 |
-
if hasattr(result, "content"):
|
| 69 |
-
return result.content
|
| 70 |
-
elif isinstance(result, dict):
|
| 71 |
-
return result.get("content", str(result))
|
| 72 |
-
else:
|
| 73 |
-
return str(result)
|
| 74 |
-
|
| 75 |
-
# Main evaluation function
|
| 76 |
-
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 77 |
-
print("Starting run_and_submit_all...")
|
| 78 |
|
|
|
|
| 79 |
space_id = os.getenv("SPACE_ID")
|
| 80 |
|
| 81 |
if profile:
|
|
@@ -108,20 +78,15 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 108 |
question_text = item.get("question")
|
| 109 |
if not task_id or question_text is None:
|
| 110 |
continue
|
| 111 |
-
|
| 112 |
-
print(f"Running agent on question: {question_text}")
|
| 113 |
try:
|
| 114 |
submitted_answer = agent(question_text)
|
| 115 |
-
print(f"Agent answer: {submitted_answer} (type: {type(submitted_answer)})")
|
| 116 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 117 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 118 |
except Exception as e:
|
| 119 |
-
error_msg = f"AGENT ERROR: {e}"
|
| 120 |
-
print(error_msg)
|
| 121 |
results_log.append({
|
| 122 |
"Task ID": task_id,
|
| 123 |
"Question": question_text,
|
| 124 |
-
"Submitted Answer":
|
| 125 |
})
|
| 126 |
|
| 127 |
if not answers_payload:
|
|
@@ -148,7 +113,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 148 |
except Exception as e:
|
| 149 |
return f"Submission failed: {e}", pd.DataFrame(results_log)
|
| 150 |
|
| 151 |
-
# Gradio UI setup
|
| 152 |
with gr.Blocks() as demo:
|
| 153 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 154 |
gr.Markdown("""
|
|
@@ -173,5 +137,3 @@ if __name__ == "__main__":
|
|
| 173 |
|
| 174 |
|
| 175 |
|
| 176 |
-
|
| 177 |
-
|
|
|
|
| 4 |
import gradio as gr
|
| 5 |
import requests
|
| 6 |
import pandas as pd
|
| 7 |
+
from openai import OpenAI
|
|
|
|
| 8 |
from smolagents import CodeAgent, DuckDuckGoSearchTool
|
|
|
|
| 9 |
|
| 10 |
# System prompt used by the agent
|
| 11 |
SYSTEM_PROMPT = """You are a general AI assistant. I will ask you a question.
|
|
|
|
| 16 |
|
| 17 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 18 |
|
|
|
|
| 19 |
class GeminiFlashModel:
|
| 20 |
+
def __init__(self, model_id="gemini-1.5-flash"):
|
| 21 |
+
self.client = OpenAI(
|
| 22 |
+
api_key=os.getenv("GEMINI_API_KEY"),
|
| 23 |
+
base_url="https://generativelanguage.googleapis.com/v1beta/openai/"
|
| 24 |
+
)
|
| 25 |
+
self.model_id = model_id
|
| 26 |
self.system_prompt = SYSTEM_PROMPT
|
| 27 |
|
| 28 |
+
def generate(self, messages):
|
| 29 |
+
# Ensure system prompt is present
|
|
|
|
|
|
|
| 30 |
if not any(m.get("role") == "system" for m in messages):
|
| 31 |
messages = [{"role": "system", "content": self.system_prompt}] + messages
|
| 32 |
|
| 33 |
+
response = self.client.chat.completions.create(
|
| 34 |
+
model=self.model_id,
|
| 35 |
+
messages=messages
|
| 36 |
+
)
|
| 37 |
+
# Return the generated content string directly
|
| 38 |
+
return response.choices[0].message.content
|
| 39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
class MyAgent:
|
| 41 |
def __init__(self):
|
| 42 |
+
self.model = GeminiFlashModel()
|
| 43 |
self.agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=self.model)
|
| 44 |
|
| 45 |
def __call__(self, question: str) -> str:
|
| 46 |
+
return self.agent.run(question)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
|
| 48 |
+
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 49 |
space_id = os.getenv("SPACE_ID")
|
| 50 |
|
| 51 |
if profile:
|
|
|
|
| 78 |
question_text = item.get("question")
|
| 79 |
if not task_id or question_text is None:
|
| 80 |
continue
|
|
|
|
|
|
|
| 81 |
try:
|
| 82 |
submitted_answer = agent(question_text)
|
|
|
|
| 83 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 84 |
results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
|
| 85 |
except Exception as e:
|
|
|
|
|
|
|
| 86 |
results_log.append({
|
| 87 |
"Task ID": task_id,
|
| 88 |
"Question": question_text,
|
| 89 |
+
"Submitted Answer": f"AGENT ERROR: {e}"
|
| 90 |
})
|
| 91 |
|
| 92 |
if not answers_payload:
|
|
|
|
| 113 |
except Exception as e:
|
| 114 |
return f"Submission failed: {e}", pd.DataFrame(results_log)
|
| 115 |
|
|
|
|
| 116 |
with gr.Blocks() as demo:
|
| 117 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 118 |
gr.Markdown("""
|
|
|
|
| 137 |
|
| 138 |
|
| 139 |
|
|
|
|
|
|