Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -4,7 +4,8 @@ import os
|
|
| 4 |
import gradio as gr
|
| 5 |
import requests
|
| 6 |
import pandas as pd
|
| 7 |
-
|
|
|
|
| 8 |
from smolagents import CodeAgent, DuckDuckGoSearchTool
|
| 9 |
|
| 10 |
# System prompt used by the agent
|
|
@@ -16,35 +17,70 @@ If you're asked for a string, don’t use articles or abbreviations (e.g. for ci
|
|
| 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 |
-
|
| 22 |
-
|
| 23 |
-
base_url="https://generativelanguage.googleapis.com/v1beta/openai/"
|
| 24 |
-
)
|
| 25 |
-
self.model_id = model_id
|
| 26 |
self.system_prompt = SYSTEM_PROMPT
|
| 27 |
|
| 28 |
-
|
| 29 |
-
|
|
|
|
|
|
|
|
|
|
| 30 |
if not any(m.get("role") == "system" for m in messages):
|
| 31 |
messages = [{"role": "system", "content": self.system_prompt}] + messages
|
| 32 |
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 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 |
-
|
| 47 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 49 |
space_id = os.getenv("SPACE_ID")
|
| 50 |
|
|
@@ -113,6 +149,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 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("""
|
|
@@ -136,4 +173,3 @@ if __name__ == "__main__":
|
|
| 136 |
|
| 137 |
|
| 138 |
|
| 139 |
-
|
|
|
|
| 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 used by the agent
|
|
|
|
| 17 |
|
| 18 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 19 |
|
| 20 |
+
# Generation result wrapper to match smolagents expectations
|
| 21 |
+
class GenerationResult:
|
| 22 |
+
def __init__(self, content, token_usage=None, input_tokens=0, output_tokens=0):
|
| 23 |
+
self.content = content
|
| 24 |
+
self.token_usage = token_usage or {}
|
| 25 |
+
self.input_tokens = input_tokens
|
| 26 |
+
self.output_tokens = output_tokens
|
| 27 |
+
|
| 28 |
+
# Gemini model wrapper
|
| 29 |
class GeminiFlashModel:
|
| 30 |
+
def __init__(self, model_id="gemini-1.5-flash", api_key=None):
|
| 31 |
+
genai.configure(api_key=api_key or os.getenv("GEMINI_API_KEY"))
|
| 32 |
+
self.model = genai.GenerativeModel(model_id)
|
|
|
|
|
|
|
|
|
|
| 33 |
self.system_prompt = SYSTEM_PROMPT
|
| 34 |
|
| 35 |
+
# Accept stop_sequences explicitly to avoid unexpected kwarg errors
|
| 36 |
+
def generate(self, messages, stop_sequences=None, **kwargs):
|
| 37 |
+
if not isinstance(messages, list) or not all(isinstance(m, dict) for m in messages):
|
| 38 |
+
raise TypeError("Expected 'messages' to be a list of dicts")
|
| 39 |
+
|
| 40 |
if not any(m.get("role") == "system" for m in messages):
|
| 41 |
messages = [{"role": "system", "content": self.system_prompt}] + messages
|
| 42 |
|
| 43 |
+
prompt = ""
|
| 44 |
+
for m in messages:
|
| 45 |
+
role = m["role"].capitalize()
|
| 46 |
+
content = m["content"]
|
| 47 |
+
prompt += f"{role}: {content}\n"
|
|
|
|
| 48 |
|
| 49 |
+
try:
|
| 50 |
+
# Note: genai.GenerativeModel.generate_content may not support stop_sequences
|
| 51 |
+
response = self.model.generate_content(prompt)
|
| 52 |
+
return GenerationResult(
|
| 53 |
+
content=response.text.strip(),
|
| 54 |
+
token_usage={}, # you can extend if API provides token info
|
| 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 wrapper
|
| 67 |
class MyAgent:
|
| 68 |
def __init__(self):
|
| 69 |
+
self.model = GeminiFlashModel(model_id="gemini-1.5-flash")
|
| 70 |
self.agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=self.model)
|
| 71 |
|
| 72 |
def __call__(self, question: str) -> str:
|
| 73 |
+
# The agent.run expects a string answer
|
| 74 |
+
result = self.agent.run(question)
|
| 75 |
+
# If result is GenerationResult or dict-like, convert to string
|
| 76 |
+
if hasattr(result, "content"):
|
| 77 |
+
return result.content
|
| 78 |
+
elif isinstance(result, dict):
|
| 79 |
+
return result.get("content", str(result))
|
| 80 |
+
else:
|
| 81 |
+
return str(result)
|
| 82 |
+
|
| 83 |
+
# Main evaluation function
|
| 84 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 85 |
space_id = os.getenv("SPACE_ID")
|
| 86 |
|
|
|
|
| 149 |
except Exception as e:
|
| 150 |
return f"Submission failed: {e}", pd.DataFrame(results_log)
|
| 151 |
|
| 152 |
+
# Gradio UI setup
|
| 153 |
with gr.Blocks() as demo:
|
| 154 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 155 |
gr.Markdown("""
|
|
|
|
| 173 |
|
| 174 |
|
| 175 |
|
|
|