Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,16 +1,16 @@
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
-
from smolagents import
|
| 5 |
-
import pandas as pd
|
| 6 |
-
import tempfile
|
| 7 |
from pathlib import Path
|
|
|
|
|
|
|
| 8 |
import re
|
| 9 |
|
| 10 |
# --- Constants ---
|
| 11 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 12 |
|
| 13 |
-
# --- File
|
| 14 |
def download_file_if_any(base_api_url: str, task_id: str) -> str | None:
|
| 15 |
url = f"{base_api_url}/files/{task_id}"
|
| 16 |
try:
|
|
@@ -20,12 +20,14 @@ def download_file_if_any(base_api_url: str, task_id: str) -> str | None:
|
|
| 20 |
resp.raise_for_status()
|
| 21 |
except requests.exceptions.HTTPError as e:
|
| 22 |
raise e
|
|
|
|
| 23 |
cdisp = resp.headers.get("content-disposition", "")
|
| 24 |
filename = task_id
|
| 25 |
if "filename=" in cdisp:
|
| 26 |
m = re.search(r'filename="([^\"]+)"', cdisp)
|
| 27 |
if m:
|
| 28 |
filename = m.group(1)
|
|
|
|
| 29 |
tmp_dir = Path(tempfile.gettempdir()) / "gaia_files"
|
| 30 |
tmp_dir.mkdir(exist_ok=True)
|
| 31 |
file_path = tmp_dir / filename
|
|
@@ -33,31 +35,26 @@ def download_file_if_any(base_api_url: str, task_id: str) -> str | None:
|
|
| 33 |
f.write(resp.content)
|
| 34 |
return str(file_path)
|
| 35 |
|
| 36 |
-
# --- Basic Agent
|
| 37 |
class BasicAgent:
|
| 38 |
def __init__(self):
|
| 39 |
-
model=OpenAIServerModel(model_id="gpt-4o"),
|
| 40 |
self.agent = CodeAgent(
|
| 41 |
-
model=
|
| 42 |
tools=[DuckDuckGoSearchTool(), WikipediaSearchTool()],
|
| 43 |
add_base_tools=True,
|
| 44 |
-
additional_authorized_imports=[
|
| 45 |
)
|
| 46 |
print("BasicAgent initialized.")
|
| 47 |
|
| 48 |
def __call__(self, question: str) -> str:
|
| 49 |
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
return fixed_answer
|
| 54 |
-
except Exception as e:
|
| 55 |
-
print(f"Error during inference: {e}")
|
| 56 |
-
return f"AGENT ERROR: {e}"
|
| 57 |
|
| 58 |
-
# ---
|
| 59 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 60 |
-
space_id = "
|
| 61 |
if profile:
|
| 62 |
username = f"{profile.username}"
|
| 63 |
else:
|
|
@@ -86,16 +83,20 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 86 |
for item in questions_data:
|
| 87 |
task_id = item.get("task_id")
|
| 88 |
question_text = item.get("question")
|
|
|
|
| 89 |
try:
|
| 90 |
file_path = download_file_if_any(api_url, task_id)
|
| 91 |
except Exception as e:
|
| 92 |
file_path = None
|
|
|
|
| 93 |
|
| 94 |
-
q_for_agent =
|
|
|
|
|
|
|
|
|
|
| 95 |
|
| 96 |
if not task_id or question_text is None:
|
| 97 |
continue
|
| 98 |
-
|
| 99 |
try:
|
| 100 |
submitted_answer = agent(q_for_agent)
|
| 101 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
|
@@ -123,7 +124,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 123 |
except Exception as e:
|
| 124 |
return f"Submission Failed: {e}", pd.DataFrame(results_log)
|
| 125 |
|
| 126 |
-
# --- UI ---
|
| 127 |
with gr.Blocks() as demo:
|
| 128 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 129 |
gr.Markdown("""
|
|
|
|
| 1 |
import os
|
| 2 |
import gradio as gr
|
| 3 |
import requests
|
| 4 |
+
from smolagents import OpenAIServerModel, DuckDuckGoSearchTool, CodeAgent, WikipediaSearchTool
|
|
|
|
|
|
|
| 5 |
from pathlib import Path
|
| 6 |
+
import tempfile
|
| 7 |
+
import pandas as pd
|
| 8 |
import re
|
| 9 |
|
| 10 |
# --- Constants ---
|
| 11 |
DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
|
| 12 |
|
| 13 |
+
# --- File Download Helper ---
|
| 14 |
def download_file_if_any(base_api_url: str, task_id: str) -> str | None:
|
| 15 |
url = f"{base_api_url}/files/{task_id}"
|
| 16 |
try:
|
|
|
|
| 20 |
resp.raise_for_status()
|
| 21 |
except requests.exceptions.HTTPError as e:
|
| 22 |
raise e
|
| 23 |
+
|
| 24 |
cdisp = resp.headers.get("content-disposition", "")
|
| 25 |
filename = task_id
|
| 26 |
if "filename=" in cdisp:
|
| 27 |
m = re.search(r'filename="([^\"]+)"', cdisp)
|
| 28 |
if m:
|
| 29 |
filename = m.group(1)
|
| 30 |
+
|
| 31 |
tmp_dir = Path(tempfile.gettempdir()) / "gaia_files"
|
| 32 |
tmp_dir.mkdir(exist_ok=True)
|
| 33 |
file_path = tmp_dir / filename
|
|
|
|
| 35 |
f.write(resp.content)
|
| 36 |
return str(file_path)
|
| 37 |
|
| 38 |
+
# --- Basic Agent ---
|
| 39 |
class BasicAgent:
|
| 40 |
def __init__(self):
|
|
|
|
| 41 |
self.agent = CodeAgent(
|
| 42 |
+
model=OpenAIServerModel(model_id="gpt-4o"),
|
| 43 |
tools=[DuckDuckGoSearchTool(), WikipediaSearchTool()],
|
| 44 |
add_base_tools=True,
|
| 45 |
+
additional_authorized_imports=[]
|
| 46 |
)
|
| 47 |
print("BasicAgent initialized.")
|
| 48 |
|
| 49 |
def __call__(self, question: str) -> str:
|
| 50 |
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
| 51 |
+
fixed_answer = self.agent.run(question)
|
| 52 |
+
print(f"Agent returning answer: {fixed_answer}")
|
| 53 |
+
return fixed_answer
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
|
| 55 |
+
# --- Evaluation Logic ---
|
| 56 |
def run_and_submit_all(profile: gr.OAuthProfile | None):
|
| 57 |
+
space_id = "l3xv/Final_Assignment_Template"
|
| 58 |
if profile:
|
| 59 |
username = f"{profile.username}"
|
| 60 |
else:
|
|
|
|
| 83 |
for item in questions_data:
|
| 84 |
task_id = item.get("task_id")
|
| 85 |
question_text = item.get("question")
|
| 86 |
+
|
| 87 |
try:
|
| 88 |
file_path = download_file_if_any(api_url, task_id)
|
| 89 |
except Exception as e:
|
| 90 |
file_path = None
|
| 91 |
+
print(f"[file fetch error] {task_id}: {e}")
|
| 92 |
|
| 93 |
+
q_for_agent = (
|
| 94 |
+
f"{question_text}\n\n---\nA file was downloaded for this task and saved locally at:\n{file_path}\n---\n\n"
|
| 95 |
+
if file_path else question_text
|
| 96 |
+
)
|
| 97 |
|
| 98 |
if not task_id or question_text is None:
|
| 99 |
continue
|
|
|
|
| 100 |
try:
|
| 101 |
submitted_answer = agent(q_for_agent)
|
| 102 |
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
|
|
|
| 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("# Basic Agent Evaluation Runner")
|
| 130 |
gr.Markdown("""
|