Spaces:
Sleeping
Sleeping
Commit
·
51ce1ed
1
Parent(s):
cbf5963
handle login
Browse files
app.py
CHANGED
|
@@ -11,17 +11,24 @@ class BasicAgent:
|
|
| 11 |
print("BasicAgent initialized.")
|
| 12 |
def __call__(self, question: str) -> str:
|
| 13 |
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
def run_and_submit_all(profile, evt=None):
|
| 19 |
space_id = SPACE_ID
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
print(f"User logged in: {username}")
|
| 23 |
-
else:
|
| 24 |
-
print("User not logged in.")
|
| 25 |
return "Please Login to Hugging Face with the button.", None
|
| 26 |
|
| 27 |
api_url = DEFAULT_API_URL
|
|
@@ -31,121 +38,72 @@ def run_and_submit_all(profile, evt=None):
|
|
| 31 |
try:
|
| 32 |
agent = BasicAgent()
|
| 33 |
except Exception as e:
|
| 34 |
-
print(f"Error instantiating agent: {e}")
|
| 35 |
return f"Error initializing agent: {e}", None
|
| 36 |
|
| 37 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
| 38 |
-
print(agent_code)
|
| 39 |
|
| 40 |
-
print(f"Fetching questions from: {questions_url}")
|
| 41 |
try:
|
| 42 |
response = requests.get(questions_url, timeout=15)
|
| 43 |
response.raise_for_status()
|
| 44 |
questions_data = response.json()
|
| 45 |
if not questions_data:
|
| 46 |
-
print("Fetched questions list is empty.")
|
| 47 |
return "Fetched questions list is empty or invalid format.", None
|
| 48 |
-
print(f"Fetched {len(questions_data)} questions.")
|
| 49 |
-
except requests.exceptions.RequestException as e:
|
| 50 |
-
print(f"Error fetching questions: {e}")
|
| 51 |
-
return f"Error fetching questions: {e}", None
|
| 52 |
-
except requests.exceptions.JSONDecodeError as e:
|
| 53 |
-
print(f"Error decoding JSON response from questions endpoint: {e}")
|
| 54 |
-
print(f"Response text: {response.text[:500]}")
|
| 55 |
-
return f"Error decoding server response for questions: {e}", None
|
| 56 |
except Exception as e:
|
| 57 |
-
|
| 58 |
-
return f"An unexpected error occurred fetching questions: {e}", None
|
| 59 |
|
| 60 |
-
results_log = []
|
| 61 |
-
answers_payload = []
|
| 62 |
-
print(f"Running agent on {len(questions_data)} questions...")
|
| 63 |
for item in questions_data:
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
if not
|
| 67 |
-
print(f"Skipping item with missing task_id or question: {item}")
|
| 68 |
continue
|
| 69 |
try:
|
| 70 |
-
|
| 71 |
-
answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
|
| 72 |
-
results_log.append({
|
| 73 |
-
"Task ID": task_id,
|
| 74 |
-
"Question": question_text,
|
| 75 |
-
"Submitted Answer": submitted_answer
|
| 76 |
-
})
|
| 77 |
except Exception as e:
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
"Question": question_text,
|
| 82 |
-
"Submitted Answer": f"AGENT ERROR: {e}"
|
| 83 |
-
})
|
| 84 |
|
| 85 |
if not answers_payload:
|
| 86 |
-
print("Agent did not produce any answers to submit.")
|
| 87 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 88 |
|
| 89 |
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
| 90 |
-
status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
|
| 91 |
-
print(status_update)
|
| 92 |
-
|
| 93 |
-
print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
|
| 94 |
try:
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
| 100 |
-
f"User: {
|
| 101 |
-
f"Overall Score: {
|
| 102 |
-
f"({
|
| 103 |
-
f"Message: {
|
| 104 |
)
|
| 105 |
-
|
| 106 |
-
results_df = pd.DataFrame(results_log)
|
| 107 |
-
return final_status, results_df
|
| 108 |
except requests.exceptions.HTTPError as e:
|
| 109 |
-
error_detail = f"Server responded with status {e.response.status_code}."
|
| 110 |
try:
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
status_message = f"Submission Failed: {error_detail}"
|
| 116 |
-
print(status_message)
|
| 117 |
-
results_df = pd.DataFrame(results_log)
|
| 118 |
-
return status_message, results_df
|
| 119 |
except requests.exceptions.Timeout:
|
| 120 |
-
|
| 121 |
-
print(status_message)
|
| 122 |
-
results_df = pd.DataFrame(results_log)
|
| 123 |
-
return status_message, results_df
|
| 124 |
-
except requests.exceptions.RequestException as e:
|
| 125 |
-
status_message = f"Submission Failed: Network error - {e}"
|
| 126 |
-
print(status_message)
|
| 127 |
-
results_df = pd.DataFrame(results_log)
|
| 128 |
-
return status_message, results_df
|
| 129 |
except Exception as e:
|
| 130 |
-
|
| 131 |
-
print(status_message)
|
| 132 |
-
results_df = pd.DataFrame(results_log)
|
| 133 |
-
return status_message, results_df
|
| 134 |
|
| 135 |
with gr.Blocks() as demo:
|
| 136 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 137 |
-
gr.Markdown(
|
| 138 |
-
"""
|
| 139 |
-
**Instructions:**
|
| 140 |
-
1) Clone this space, then modify the code to define your agent's logic.
|
| 141 |
-
2) Log in to your Hugging Face account using the button below.
|
| 142 |
-
3) Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
|
| 143 |
-
"""
|
| 144 |
-
)
|
| 145 |
|
| 146 |
login = gr.LoginButton()
|
| 147 |
user_state = gr.State()
|
| 148 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 149 |
|
| 150 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 151 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
|
@@ -154,13 +112,4 @@ with gr.Blocks() as demo:
|
|
| 154 |
run_button.click(fn=run_and_submit_all, inputs=[user_state], outputs=[status_output, results_table])
|
| 155 |
|
| 156 |
if __name__ == "__main__":
|
| 157 |
-
|
| 158 |
-
space_host_startup = os.getenv("SPACE_HOST")
|
| 159 |
-
print(f"Repo URL: https://huggingface.co/spaces/{SPACE_ID}")
|
| 160 |
-
print(f"Repo Tree URL: https://huggingface.co/spaces/{SPACE_ID}/tree/main")
|
| 161 |
-
if space_host_startup:
|
| 162 |
-
print(f"Runtime URL should be: https://{space_host_startup}.hf.space")
|
| 163 |
-
else:
|
| 164 |
-
print("SPACE_HOST not found (running locally?).")
|
| 165 |
-
print("-"*(60 + len(" App Starting ")) + "\n")
|
| 166 |
-
demo.launch(debug=True, share=False)
|
|
|
|
| 11 |
print("BasicAgent initialized.")
|
| 12 |
def __call__(self, question: str) -> str:
|
| 13 |
print(f"Agent received question (first 50 chars): {question[:50]}...")
|
| 14 |
+
return "This is a default answer."
|
| 15 |
+
|
| 16 |
+
def _username_from_profile(p):
|
| 17 |
+
if p is None:
|
| 18 |
+
return None
|
| 19 |
+
# object with attribute
|
| 20 |
+
u = getattr(p, "username", None)
|
| 21 |
+
if u:
|
| 22 |
+
return str(u)
|
| 23 |
+
# dict-style
|
| 24 |
+
if isinstance(p, dict):
|
| 25 |
+
return str(p.get("username") or "")
|
| 26 |
+
return None
|
| 27 |
|
| 28 |
def run_and_submit_all(profile, evt=None):
|
| 29 |
space_id = SPACE_ID
|
| 30 |
+
username = _username_from_profile(profile)
|
| 31 |
+
if not username:
|
|
|
|
|
|
|
|
|
|
| 32 |
return "Please Login to Hugging Face with the button.", None
|
| 33 |
|
| 34 |
api_url = DEFAULT_API_URL
|
|
|
|
| 38 |
try:
|
| 39 |
agent = BasicAgent()
|
| 40 |
except Exception as e:
|
|
|
|
| 41 |
return f"Error initializing agent: {e}", None
|
| 42 |
|
| 43 |
agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
|
|
|
|
| 44 |
|
|
|
|
| 45 |
try:
|
| 46 |
response = requests.get(questions_url, timeout=15)
|
| 47 |
response.raise_for_status()
|
| 48 |
questions_data = response.json()
|
| 49 |
if not questions_data:
|
|
|
|
| 50 |
return "Fetched questions list is empty or invalid format.", None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
except Exception as e:
|
| 52 |
+
return f"Error fetching questions: {e}", None
|
|
|
|
| 53 |
|
| 54 |
+
results_log, answers_payload = [], []
|
|
|
|
|
|
|
| 55 |
for item in questions_data:
|
| 56 |
+
tid = item.get("task_id")
|
| 57 |
+
q = item.get("question")
|
| 58 |
+
if not tid or q is None:
|
|
|
|
| 59 |
continue
|
| 60 |
try:
|
| 61 |
+
ans = agent(q)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
except Exception as e:
|
| 63 |
+
ans = f"AGENT ERROR: {e}"
|
| 64 |
+
answers_payload.append({"task_id": tid, "submitted_answer": ans})
|
| 65 |
+
results_log.append({"Task ID": tid, "Question": q, "Submitted Answer": ans})
|
|
|
|
|
|
|
|
|
|
| 66 |
|
| 67 |
if not answers_payload:
|
|
|
|
| 68 |
return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
|
| 69 |
|
| 70 |
submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
try:
|
| 72 |
+
resp = requests.post(submit_url, json=submission_data, timeout=60)
|
| 73 |
+
resp.raise_for_status()
|
| 74 |
+
data = resp.json()
|
| 75 |
+
status = (
|
| 76 |
+
"Submission Successful!\n"
|
| 77 |
+
f"User: {data.get('username')}\n"
|
| 78 |
+
f"Overall Score: {data.get('score','N/A')}% "
|
| 79 |
+
f"({data.get('correct_count','?')}/{data.get('total_attempted','?')} correct)\n"
|
| 80 |
+
f"Message: {data.get('message','No message received.')}"
|
| 81 |
)
|
| 82 |
+
return status, pd.DataFrame(results_log)
|
|
|
|
|
|
|
| 83 |
except requests.exceptions.HTTPError as e:
|
|
|
|
| 84 |
try:
|
| 85 |
+
detail = e.response.json().get("detail", e.response.text)
|
| 86 |
+
except Exception:
|
| 87 |
+
detail = e.response.text
|
| 88 |
+
return f"Submission Failed: HTTP {e.response.status_code}. Detail: {detail[:500]}", pd.DataFrame(results_log)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 89 |
except requests.exceptions.Timeout:
|
| 90 |
+
return "Submission Failed: The request timed out.", pd.DataFrame(results_log)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
except Exception as e:
|
| 92 |
+
return f"Submission Failed: {e}", pd.DataFrame(results_log)
|
|
|
|
|
|
|
|
|
|
| 93 |
|
| 94 |
with gr.Blocks() as demo:
|
| 95 |
gr.Markdown("# Basic Agent Evaluation Runner")
|
| 96 |
+
gr.Markdown("Click **Log in with Hugging Face** first. After you see your username below, press **Run**.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
|
| 98 |
login = gr.LoginButton()
|
| 99 |
user_state = gr.State()
|
| 100 |
+
whoami = gr.Markdown()
|
| 101 |
+
|
| 102 |
+
def store_and_echo(p):
|
| 103 |
+
u = _username_from_profile(p) or "(not logged in)"
|
| 104 |
+
return p, f"✅ Logged in as **{u}**" if u and u != "(not logged in)" else "❌ Not logged in"
|
| 105 |
+
|
| 106 |
+
login.click(store_and_echo, inputs=login, outputs=[user_state, whoami])
|
| 107 |
|
| 108 |
run_button = gr.Button("Run Evaluation & Submit All Answers")
|
| 109 |
status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
|
|
|
|
| 112 |
run_button.click(fn=run_and_submit_all, inputs=[user_state], outputs=[status_output, results_table])
|
| 113 |
|
| 114 |
if __name__ == "__main__":
|
| 115 |
+
demo.launch(debug=True, share=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|