Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
15dd199
1
Parent(s):
347797e
refresh and persistent logging
Browse files- app.py +7 -14
- utils/leaderboard.py +17 -3
- utils/vote_logger.py +19 -10
app.py
CHANGED
|
@@ -406,36 +406,29 @@ The Elo rating system provides a more accurate ranking than simple win rates:
|
|
| 406 |
outputs=[results_table_display]
|
| 407 |
)
|
| 408 |
|
| 409 |
-
#
|
| 410 |
-
# First click event: Update UI immediately
|
| 411 |
for btn in [random_question_btn, try_another_btn]:
|
| 412 |
-
|
| 413 |
-
|
| 414 |
fn=show_loading_state,
|
| 415 |
inputs=[],
|
| 416 |
outputs=[summary_a_display, summary_b_display, vote_button_a,
|
| 417 |
vote_button_b, vote_button_tie, vote_button_neither]
|
| 418 |
).then(
|
|
|
|
| 419 |
fn=handle_new_example_click,
|
| 420 |
inputs=[],
|
| 421 |
outputs=[current_example]
|
| 422 |
).then(
|
|
|
|
| 423 |
fn=update_ui_for_new_context,
|
| 424 |
inputs=[current_example],
|
| 425 |
outputs=[query_display, context_description, context_display,
|
| 426 |
context_toggle_btn, show_full_context]
|
| 427 |
-
)
|
| 428 |
-
|
| 429 |
-
# Second click event for each button runs in parallel with the first
|
| 430 |
-
for btn in [random_question_btn, try_another_btn]:
|
| 431 |
-
# Generate model outputs (potentially slower operation)
|
| 432 |
-
event2 = btn.click(
|
| 433 |
-
fn=handle_new_example_click, # This will be called separately from the first event
|
| 434 |
-
inputs=[],
|
| 435 |
-
outputs=[current_example]
|
| 436 |
).then(
|
|
|
|
| 437 |
fn=process_example,
|
| 438 |
-
inputs=[current_example],
|
| 439 |
outputs=[model_a_name, model_b_name, summary_a_text, summary_b_text,
|
| 440 |
selected_winner, feedback_list, show_results_state, results_agg,
|
| 441 |
summary_a_display, summary_b_display, vote_button_a, vote_button_b,
|
|
|
|
| 406 |
outputs=[results_table_display]
|
| 407 |
)
|
| 408 |
|
| 409 |
+
# Use a single event chain for each button, structured to update UI first, then run inference
|
|
|
|
| 410 |
for btn in [random_question_btn, try_another_btn]:
|
| 411 |
+
btn.click(
|
| 412 |
+
# Step 1: Show loading state immediately
|
| 413 |
fn=show_loading_state,
|
| 414 |
inputs=[],
|
| 415 |
outputs=[summary_a_display, summary_b_display, vote_button_a,
|
| 416 |
vote_button_b, vote_button_tie, vote_button_neither]
|
| 417 |
).then(
|
| 418 |
+
# Step 2: Get new example
|
| 419 |
fn=handle_new_example_click,
|
| 420 |
inputs=[],
|
| 421 |
outputs=[current_example]
|
| 422 |
).then(
|
| 423 |
+
# Step 3: Update context UI immediately
|
| 424 |
fn=update_ui_for_new_context,
|
| 425 |
inputs=[current_example],
|
| 426 |
outputs=[query_display, context_description, context_display,
|
| 427 |
context_toggle_btn, show_full_context]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 428 |
).then(
|
| 429 |
+
# Step 4: Then process example for model outputs
|
| 430 |
fn=process_example,
|
| 431 |
+
inputs=[current_example],
|
| 432 |
outputs=[model_a_name, model_b_name, summary_a_text, summary_b_text,
|
| 433 |
selected_winner, feedback_list, show_results_state, results_agg,
|
| 434 |
summary_a_display, summary_b_display, vote_button_a, vote_button_b,
|
utils/leaderboard.py
CHANGED
|
@@ -3,6 +3,7 @@ import pandas as pd
|
|
| 3 |
import math
|
| 4 |
from datetime import datetime
|
| 5 |
from .models import models
|
|
|
|
| 6 |
|
| 7 |
# Default K-factor (determines how much a single match affects ratings)
|
| 8 |
DEFAULT_K_FACTOR = 32
|
|
@@ -10,6 +11,18 @@ DEFAULT_K_FACTOR = 32
|
|
| 10 |
# Default starting Elo
|
| 11 |
DEFAULT_ELO = 1500
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
def prepare_url(model_dict: dict):
|
| 14 |
"""
|
| 15 |
Prepare the URL for the model based on its name.
|
|
@@ -119,7 +132,7 @@ def load_leaderboard_data():
|
|
| 119 |
|
| 120 |
try:
|
| 121 |
# Define the path to the CSV file for leaderboard
|
| 122 |
-
csv_path =
|
| 123 |
|
| 124 |
# Check if the file exists and load it
|
| 125 |
if os.path.exists(csv_path):
|
|
@@ -247,8 +260,9 @@ def save_leaderboard_data(results):
|
|
| 247 |
df = df.sort_values(by='elo', ascending=False)
|
| 248 |
|
| 249 |
# Save to CSV
|
| 250 |
-
|
| 251 |
-
|
|
|
|
| 252 |
except Exception as e:
|
| 253 |
print(f"Error saving leaderboard data: {e}")
|
| 254 |
|
|
|
|
| 3 |
import math
|
| 4 |
from datetime import datetime
|
| 5 |
from .models import models
|
| 6 |
+
from huggingface_hub import CommitScheduler
|
| 7 |
|
| 8 |
# Default K-factor (determines how much a single match affects ratings)
|
| 9 |
DEFAULT_K_FACTOR = 32
|
|
|
|
| 11 |
# Default starting Elo
|
| 12 |
DEFAULT_ELO = 1500
|
| 13 |
|
| 14 |
+
LEADERBOARD_FN = './arena_elo_leaderboard.csv'
|
| 15 |
+
|
| 16 |
+
csv_path = os.path.join('utils', 'arena_elo_leaderboard.csv')
|
| 17 |
+
|
| 18 |
+
leaderboard_scheduler = CommitScheduler(
|
| 19 |
+
repo_id="aizip-dev/Arena-Metadata",
|
| 20 |
+
folder_path="leaderboard",
|
| 21 |
+
repo_type="dataset",
|
| 22 |
+
every=5
|
| 23 |
+
)
|
| 24 |
+
|
| 25 |
+
|
| 26 |
def prepare_url(model_dict: dict):
|
| 27 |
"""
|
| 28 |
Prepare the URL for the model based on its name.
|
|
|
|
| 132 |
|
| 133 |
try:
|
| 134 |
# Define the path to the CSV file for leaderboard
|
| 135 |
+
csv_path = LEADERBOARD_FN
|
| 136 |
|
| 137 |
# Check if the file exists and load it
|
| 138 |
if os.path.exists(csv_path):
|
|
|
|
| 260 |
df = df.sort_values(by='elo', ascending=False)
|
| 261 |
|
| 262 |
# Save to CSV
|
| 263 |
+
with leaderboard_scheduler.lock:
|
| 264 |
+
df.to_csv(csv_path, index=False)
|
| 265 |
+
print(f"Leaderboard data saved successfully to {csv_path}")
|
| 266 |
except Exception as e:
|
| 267 |
print(f"Error saving leaderboard data: {e}")
|
| 268 |
|
utils/vote_logger.py
CHANGED
|
@@ -3,6 +3,14 @@ import csv
|
|
| 3 |
import json
|
| 4 |
from datetime import datetime
|
| 5 |
import pandas as pd
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
def save_vote_details(example, model_a, model_b, winner, feedback, summary_a, summary_b):
|
| 8 |
"""
|
|
@@ -35,17 +43,18 @@ def save_vote_details(example, model_a, model_b, winner, feedback, summary_a, su
|
|
| 35 |
|
| 36 |
try:
|
| 37 |
# Open the file in append mode
|
| 38 |
-
with
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
|
|
|
| 47 |
|
| 48 |
-
|
| 49 |
except Exception as e:
|
| 50 |
print(f"Error saving vote details: {e}")
|
| 51 |
|
|
|
|
| 3 |
import json
|
| 4 |
from datetime import datetime
|
| 5 |
import pandas as pd
|
| 6 |
+
from huggingface_hub import CommitScheduler
|
| 7 |
+
|
| 8 |
+
vote_details_scheduler = CommitScheduler(
|
| 9 |
+
repo_id="aizip-dev/Arena-Metadata",
|
| 10 |
+
folder_path="votes",
|
| 11 |
+
repo_type="dataset",
|
| 12 |
+
every=5
|
| 13 |
+
)
|
| 14 |
|
| 15 |
def save_vote_details(example, model_a, model_b, winner, feedback, summary_a, summary_b):
|
| 16 |
"""
|
|
|
|
| 43 |
|
| 44 |
try:
|
| 45 |
# Open the file in append mode
|
| 46 |
+
with vote_details_scheduler:
|
| 47 |
+
with open(csv_path, 'a', newline='', encoding='utf-8') as f:
|
| 48 |
+
writer = csv.DictWriter(f, fieldnames=vote_record.keys())
|
| 49 |
+
|
| 50 |
+
# Write header if file doesn't exist
|
| 51 |
+
if not file_exists:
|
| 52 |
+
writer.writeheader()
|
| 53 |
+
|
| 54 |
+
# Write the vote record
|
| 55 |
+
writer.writerow(vote_record)
|
| 56 |
|
| 57 |
+
print(f"Vote details saved to {csv_path}")
|
| 58 |
except Exception as e:
|
| 59 |
print(f"Error saving vote details: {e}")
|
| 60 |
|