Update app.py
Browse files
app.py
CHANGED
|
@@ -7,10 +7,17 @@ import requests
|
|
| 7 |
import pandas as pd
|
| 8 |
import gradio as gr
|
| 9 |
|
| 10 |
-
#
|
| 11 |
-
API_URL
|
| 12 |
-
MODEL_ID
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
|
| 15 |
WELCOME = """
|
| 16 |
## GAIA Benchmark Runner π
|
|
@@ -19,14 +26,26 @@ Build your agent, score **β₯30%** to earn your Certificate,
|
|
| 19 |
and see where you land on the Student Leaderboard!
|
| 20 |
"""
|
| 21 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
class GAIAAgent:
|
| 23 |
-
def __init__(self, model_id: str
|
| 24 |
-
print(f"[DEBUG] Initializing
|
| 25 |
self.model_id = model_id
|
| 26 |
-
self.headers =
|
| 27 |
|
| 28 |
def answer(self, prompt: str) -> str:
|
| 29 |
-
print(f"[DEBUG] Sending prompt of length {len(prompt)} to HF Inference") # debug
|
| 30 |
payload = {
|
| 31 |
"inputs": prompt,
|
| 32 |
"parameters": {"max_new_tokens": 512, "temperature": 0.2}
|
|
@@ -35,65 +54,50 @@ class GAIAAgent:
|
|
| 35 |
resp = requests.post(url, headers=self.headers, json=payload, timeout=60)
|
| 36 |
resp.raise_for_status()
|
| 37 |
data = resp.json()
|
| 38 |
-
print(f"[DEBUG] Got response from model: {data!r}") # debug
|
| 39 |
if isinstance(data, list) and data and "generated_text" in data[0]:
|
| 40 |
return data[0]["generated_text"].strip()
|
| 41 |
return str(data)
|
| 42 |
|
| 43 |
-
|
|
|
|
| 44 |
try:
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
return
|
| 49 |
-
print(f"[DEBUG] Logged in as: {profile.username}") # debug
|
| 50 |
-
username = profile.username
|
| 51 |
-
|
| 52 |
-
hf_token = HF_TOKEN_ENV or getattr(profile, "access_token", None)
|
| 53 |
-
print(f"[DEBUG] Using HF token from {'env' if HF_TOKEN_ENV else 'profile'}") # debug
|
| 54 |
-
if not hf_token:
|
| 55 |
-
print("[DEBUG] No HF token found") # debug
|
| 56 |
-
return (
|
| 57 |
-
"β No Hugging Face token found.\n"
|
| 58 |
-
"Set HUGGINGFACEHUB_API_TOKEN in Secrets or log in via the button.",
|
| 59 |
-
pd.DataFrame()
|
| 60 |
-
)
|
| 61 |
|
| 62 |
# 1) Fetch GAIA questions
|
| 63 |
-
print(f"[DEBUG] Fetching questions from {API_URL}/questions") # debug
|
| 64 |
q_resp = requests.get(f"{API_URL}/questions", timeout=15)
|
| 65 |
q_resp.raise_for_status()
|
| 66 |
questions = q_resp.json() or []
|
| 67 |
-
print(f"[DEBUG] Received {len(questions)} questions") # debug
|
| 68 |
if not questions:
|
| 69 |
-
return
|
| 70 |
|
| 71 |
# 2) Init agent
|
| 72 |
-
agent = GAIAAgent(MODEL_ID
|
| 73 |
|
| 74 |
# 3) Answer each
|
| 75 |
results = []
|
| 76 |
payload = []
|
| 77 |
for item in questions:
|
| 78 |
-
print(f"[DEBUG] Processing task_id={item.get('task_id')}") # debug
|
| 79 |
tid = item.get("task_id")
|
| 80 |
qtxt = item.get("question", "")
|
| 81 |
try:
|
| 82 |
ans = agent.answer(qtxt)
|
| 83 |
except Exception as e:
|
| 84 |
ans = f"ERROR: {e}"
|
| 85 |
-
print(f"[DEBUG] Error answering: {e}") # debug
|
| 86 |
results.append({"Task ID": tid, "Question": qtxt, "Answer": ans})
|
| 87 |
payload.append({"task_id": tid, "submitted_answer": ans})
|
| 88 |
time.sleep(0.5)
|
| 89 |
|
| 90 |
-
# 4) Submit
|
| 91 |
-
|
| 92 |
-
|
|
|
|
|
|
|
| 93 |
s_resp = requests.post(f"{API_URL}/submit", json=submission, timeout=60)
|
| 94 |
s_resp.raise_for_status()
|
| 95 |
data = s_resp.json()
|
| 96 |
-
print(f"[DEBUG] Submission response: {data!r}") # debug
|
| 97 |
|
| 98 |
# 5) Build status text
|
| 99 |
status = (
|
|
@@ -108,18 +112,18 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
|
|
| 108 |
except Exception as e:
|
| 109 |
tb = traceback.format_exc()
|
| 110 |
print("[ERROR] Unhandled exception:\n", tb)
|
| 111 |
-
return (f"β Unexpected error:\n{e}", pd.DataFrame()
|
| 112 |
|
|
|
|
| 113 |
with gr.Blocks() as demo:
|
| 114 |
gr.Markdown(WELCOME)
|
| 115 |
-
login = gr.LoginButton()
|
| 116 |
run_btn = gr.Button("βΆοΈ Run GAIA Benchmark")
|
| 117 |
status = gr.Markdown()
|
| 118 |
table_df = gr.Dataframe(headers=["Task ID", "Question", "Answer"], wrap=True)
|
| 119 |
|
| 120 |
run_btn.click(
|
| 121 |
fn=run_and_submit_all,
|
| 122 |
-
inputs=[
|
| 123 |
outputs=[status, table_df]
|
| 124 |
)
|
| 125 |
|
|
|
|
| 7 |
import pandas as pd
|
| 8 |
import gradio as gr
|
| 9 |
|
| 10 |
+
# βββ Configuration ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 11 |
+
API_URL = os.getenv("API_URL", "https://agents-course-unit4-scoring.hf.space")
|
| 12 |
+
MODEL_ID = os.getenv("MODEL_ID", "meta-llama/Llama-2-7b-instruct")
|
| 13 |
+
HF_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN")
|
| 14 |
+
|
| 15 |
+
if not HF_TOKEN:
|
| 16 |
+
raise RuntimeError(
|
| 17 |
+
"β Please set HUGGINGFACEHUB_API_TOKEN in your Space Secrets."
|
| 18 |
+
)
|
| 19 |
+
|
| 20 |
+
HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"}
|
| 21 |
|
| 22 |
WELCOME = """
|
| 23 |
## GAIA Benchmark Runner π
|
|
|
|
| 26 |
and see where you land on the Student Leaderboard!
|
| 27 |
"""
|
| 28 |
|
| 29 |
+
# βββ Utility to fetch your HF username from the token ββββββββββββββββββββββββ
|
| 30 |
+
def get_hf_username():
|
| 31 |
+
try:
|
| 32 |
+
resp = requests.get("https://huggingface.co/api/whoami-v2", headers=HEADERS, timeout=10)
|
| 33 |
+
resp.raise_for_status()
|
| 34 |
+
data = resp.json()
|
| 35 |
+
# V2 returns {"user": { "id": ..., "username": ... }, ...}
|
| 36 |
+
return data.get("user", {}).get("username") or data.get("name")
|
| 37 |
+
except Exception as e:
|
| 38 |
+
print("[DEBUG] whoami failed:", e)
|
| 39 |
+
return None
|
| 40 |
+
|
| 41 |
+
# βββ Simple HF-Inference Agent βββββββββββββββββββββββββββββββββββββββββββββ
|
| 42 |
class GAIAAgent:
|
| 43 |
+
def __init__(self, model_id: str):
|
| 44 |
+
print(f"[DEBUG] Initializing with model {model_id}")
|
| 45 |
self.model_id = model_id
|
| 46 |
+
self.headers = HEADERS
|
| 47 |
|
| 48 |
def answer(self, prompt: str) -> str:
|
|
|
|
| 49 |
payload = {
|
| 50 |
"inputs": prompt,
|
| 51 |
"parameters": {"max_new_tokens": 512, "temperature": 0.2}
|
|
|
|
| 54 |
resp = requests.post(url, headers=self.headers, json=payload, timeout=60)
|
| 55 |
resp.raise_for_status()
|
| 56 |
data = resp.json()
|
|
|
|
| 57 |
if isinstance(data, list) and data and "generated_text" in data[0]:
|
| 58 |
return data[0]["generated_text"].strip()
|
| 59 |
return str(data)
|
| 60 |
|
| 61 |
+
# βββ Gradio callback ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 62 |
+
def run_and_submit_all():
|
| 63 |
try:
|
| 64 |
+
# 0) Resolve username
|
| 65 |
+
username = get_hf_username()
|
| 66 |
+
if not username:
|
| 67 |
+
return "β Could not fetch your HF username. Check your token.", pd.DataFrame()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 68 |
|
| 69 |
# 1) Fetch GAIA questions
|
|
|
|
| 70 |
q_resp = requests.get(f"{API_URL}/questions", timeout=15)
|
| 71 |
q_resp.raise_for_status()
|
| 72 |
questions = q_resp.json() or []
|
|
|
|
| 73 |
if not questions:
|
| 74 |
+
return "β No questions returned; check your API_URL.", pd.DataFrame()
|
| 75 |
|
| 76 |
# 2) Init agent
|
| 77 |
+
agent = GAIAAgent(MODEL_ID)
|
| 78 |
|
| 79 |
# 3) Answer each
|
| 80 |
results = []
|
| 81 |
payload = []
|
| 82 |
for item in questions:
|
|
|
|
| 83 |
tid = item.get("task_id")
|
| 84 |
qtxt = item.get("question", "")
|
| 85 |
try:
|
| 86 |
ans = agent.answer(qtxt)
|
| 87 |
except Exception as e:
|
| 88 |
ans = f"ERROR: {e}"
|
|
|
|
| 89 |
results.append({"Task ID": tid, "Question": qtxt, "Answer": ans})
|
| 90 |
payload.append({"task_id": tid, "submitted_answer": ans})
|
| 91 |
time.sleep(0.5)
|
| 92 |
|
| 93 |
+
# 4) Submit all answers
|
| 94 |
+
submission = {
|
| 95 |
+
"username": username,
|
| 96 |
+
"answers": payload
|
| 97 |
+
}
|
| 98 |
s_resp = requests.post(f"{API_URL}/submit", json=submission, timeout=60)
|
| 99 |
s_resp.raise_for_status()
|
| 100 |
data = s_resp.json()
|
|
|
|
| 101 |
|
| 102 |
# 5) Build status text
|
| 103 |
status = (
|
|
|
|
| 112 |
except Exception as e:
|
| 113 |
tb = traceback.format_exc()
|
| 114 |
print("[ERROR] Unhandled exception:\n", tb)
|
| 115 |
+
return (f"β Unexpected error:\n{e}\n\nSee logs for details."), pd.DataFrame()
|
| 116 |
|
| 117 |
+
# βββ Gradio UI ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 118 |
with gr.Blocks() as demo:
|
| 119 |
gr.Markdown(WELCOME)
|
|
|
|
| 120 |
run_btn = gr.Button("βΆοΈ Run GAIA Benchmark")
|
| 121 |
status = gr.Markdown()
|
| 122 |
table_df = gr.Dataframe(headers=["Task ID", "Question", "Answer"], wrap=True)
|
| 123 |
|
| 124 |
run_btn.click(
|
| 125 |
fn=run_and_submit_all,
|
| 126 |
+
inputs=[],
|
| 127 |
outputs=[status, table_df]
|
| 128 |
)
|
| 129 |
|