Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -8,7 +8,7 @@ import pandas as pd
|
|
| 8 |
import google.generativeai as genai
|
| 9 |
from smolagents import CodeAgent, DuckDuckGoSearchTool
|
| 10 |
|
| 11 |
-
#
|
| 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.
|
|
@@ -17,16 +17,17 @@ If you're asked for a string, don’t use articles or abbreviations (e.g. for ci
|
|
| 17 |
|
| 18 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 19 |
|
| 20 |
-
#
|
| 21 |
class GenerationResult:
|
| 22 |
-
def __init__(self, content, token_usage=None):
|
| 23 |
self.content = content
|
| 24 |
self.token_usage = token_usage or {}
|
|
|
|
|
|
|
| 25 |
|
| 26 |
# Gemini model wrapper
|
| 27 |
class GeminiFlashModel:
|
| 28 |
def __init__(self, model_id="gemini-1.5-flash", api_key=None):
|
| 29 |
-
import google.generativeai as genai
|
| 30 |
genai.configure(api_key=api_key or os.getenv("GEMINI_API_KEY"))
|
| 31 |
self.model = genai.GenerativeModel(model_id)
|
| 32 |
self.system_prompt = SYSTEM_PROMPT
|
|
@@ -46,11 +47,21 @@ class GeminiFlashModel:
|
|
| 46 |
|
| 47 |
try:
|
| 48 |
response = self.model.generate_content(prompt)
|
| 49 |
-
return GenerationResult(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
except Exception as e:
|
| 51 |
-
return GenerationResult(
|
| 52 |
-
|
| 53 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
class MyAgent:
|
| 55 |
def __init__(self):
|
| 56 |
self.model = GeminiFlashModel(model_id="gemini-1.5-flash")
|
|
@@ -59,6 +70,7 @@ class MyAgent:
|
|
| 59 |
def __call__(self, question: str) -> str:
|
| 60 |
return self.agent.run(question)
|
| 61 |
|
|
|
|
| 62 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 63 |
space_id = os.getenv("SPACE_ID")
|
| 64 |
|
|
@@ -127,7 +139,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 127 |
except Exception as e:
|
| 128 |
return f"Submission failed: {e}", pd.DataFrame(results_log)
|
| 129 |
|
| 130 |
-
# Gradio UI
|
| 131 |
with gr.Blocks() as demo:
|
| 132 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 133 |
gr.Markdown("""
|
|
|
|
| 8 |
import google.generativeai as genai
|
| 9 |
from smolagents import CodeAgent, DuckDuckGoSearchTool
|
| 10 |
|
| 11 |
+
# System prompt used by the agent
|
| 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.
|
|
|
|
| 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
|
|
|
|
| 47 |
|
| 48 |
try:
|
| 49 |
response = self.model.generate_content(prompt)
|
| 50 |
+
return GenerationResult(
|
| 51 |
+
content=response.text.strip(),
|
| 52 |
+
token_usage={},
|
| 53 |
+
input_tokens=0,
|
| 54 |
+
output_tokens=0
|
| 55 |
+
)
|
| 56 |
except Exception as e:
|
| 57 |
+
return GenerationResult(
|
| 58 |
+
content=f"GENERATION ERROR: {e}",
|
| 59 |
+
token_usage={},
|
| 60 |
+
input_tokens=0,
|
| 61 |
+
output_tokens=0
|
| 62 |
+
)
|
| 63 |
+
|
| 64 |
+
# Agent wrapper
|
| 65 |
class MyAgent:
|
| 66 |
def __init__(self):
|
| 67 |
self.model = GeminiFlashModel(model_id="gemini-1.5-flash")
|
|
|
|
| 70 |
def __call__(self, question: str) -> str:
|
| 71 |
return self.agent.run(question)
|
| 72 |
|
| 73 |
+
# Main evaluation function
|
| 74 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 75 |
space_id = os.getenv("SPACE_ID")
|
| 76 |
|
|
|
|
| 139 |
except Exception as e:
|
| 140 |
return f"Submission failed: {e}", pd.DataFrame(results_log)
|
| 141 |
|
| 142 |
+
# Gradio UI setup
|
| 143 |
with gr.Blocks() as demo:
|
| 144 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 145 |
gr.Markdown("""
|