Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,7 +3,7 @@ from functools import lru_cache
|
|
| 3 |
import random
|
| 4 |
import requests
|
| 5 |
import logging
|
| 6 |
-
import
|
| 7 |
import plotly.graph_objects as go
|
| 8 |
from typing import Dict
|
| 9 |
from leaderboard import (
|
|
@@ -15,7 +15,7 @@ from leaderboard import (
|
|
| 15 |
ensure_elo_ratings_initialized
|
| 16 |
)
|
| 17 |
import sys
|
| 18 |
-
|
| 19 |
import threading
|
| 20 |
import time
|
| 21 |
from collections import Counter
|
|
@@ -31,7 +31,7 @@ start_backup_thread()
|
|
| 31 |
|
| 32 |
# Function to get available models (using predefined list)
|
| 33 |
def get_available_models():
|
| 34 |
-
return [model[0] for model in
|
| 35 |
|
| 36 |
# Function to get recent opponents for a model
|
| 37 |
recent_opponents = {}
|
|
@@ -46,30 +46,28 @@ def update_recent_opponents(model_a, model_b):
|
|
| 46 |
# Function to call Ollama API with caching
|
| 47 |
@lru_cache(maxsize=100)
|
| 48 |
def call_ollama_api(model, prompt):
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
"content": "You are a helpful assistant. At no point should you reveal your name, identity or team affiliation to the user, especially if asked directly!"
|
| 55 |
-
},
|
| 56 |
-
{
|
| 57 |
-
"role": "user",
|
| 58 |
-
"content": prompt
|
| 59 |
-
}
|
| 60 |
-
]
|
| 61 |
-
}
|
| 62 |
try:
|
| 63 |
-
response =
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
timeout=100
|
| 68 |
)
|
| 69 |
-
response.
|
| 70 |
-
|
| 71 |
-
return data["choices"][0]["message"]["content"]
|
| 72 |
-
except requests.exceptions.RequestException as e:
|
| 73 |
logger.error(f"Error calling Ollama API for model {model}: {e}")
|
| 74 |
return f"Error: Unable to get response from the model."
|
| 75 |
|
|
@@ -132,8 +130,8 @@ def battle_arena(prompt):
|
|
| 132 |
gr.update(value="Error: Unable to get response from the model", visible=True)
|
| 133 |
)
|
| 134 |
|
| 135 |
-
nickname_a = random.choice(
|
| 136 |
-
nickname_b = random.choice(
|
| 137 |
|
| 138 |
# Format responses for gr.Chatbot, including the user's prompt
|
| 139 |
response_a_formatted = [
|
|
@@ -283,8 +281,8 @@ def get_leaderboard_chart():
|
|
| 283 |
return fig
|
| 284 |
|
| 285 |
def new_battle():
|
| 286 |
-
nickname_a = random.choice(
|
| 287 |
-
nickname_b = random.choice(
|
| 288 |
return (
|
| 289 |
"", # Reset prompt_input
|
| 290 |
gr.update(value=[], label=nickname_a), # Reset left Chatbot
|
|
@@ -303,18 +301,18 @@ def new_battle():
|
|
| 303 |
|
| 304 |
# Add this new function
|
| 305 |
def get_human_readable_name(model_name: str) -> str:
|
| 306 |
-
model_dict = dict(
|
| 307 |
return model_dict.get(model_name, model_name)
|
| 308 |
|
| 309 |
# Add this new function to randomly select a prompt
|
| 310 |
def random_prompt():
|
| 311 |
-
return random.choice(
|
| 312 |
|
| 313 |
# Modify the continue_conversation function
|
| 314 |
def continue_conversation(prompt, left_chat, right_chat, left_model, right_model, previous_prompt, tie_count):
|
| 315 |
# Check if the prompt is empty or the same as the previous one
|
| 316 |
if not prompt or prompt == previous_prompt:
|
| 317 |
-
prompt = random.choice(
|
| 318 |
|
| 319 |
left_response = call_ollama_api(left_model, prompt)
|
| 320 |
right_response = call_ollama_api(right_model, prompt)
|
|
@@ -337,84 +335,6 @@ def continue_conversation(prompt, left_chat, right_chat, left_model, right_model
|
|
| 337 |
tie_count
|
| 338 |
)
|
| 339 |
|
| 340 |
-
def get_fun_stats_html():
|
| 341 |
-
stats = get_fun_stats()
|
| 342 |
-
|
| 343 |
-
html = f"""
|
| 344 |
-
<style>
|
| 345 |
-
.fun-stats {{
|
| 346 |
-
font-family: Arial, sans-serif;
|
| 347 |
-
font-size: 18px;
|
| 348 |
-
line-height: 1.6;
|
| 349 |
-
max-width: 800px;
|
| 350 |
-
margin: 0 auto;
|
| 351 |
-
padding: 20px;
|
| 352 |
-
}}
|
| 353 |
-
.fun-stats h2 {{
|
| 354 |
-
font-size: 36px;
|
| 355 |
-
color: inherit;
|
| 356 |
-
text-align: center;
|
| 357 |
-
margin-bottom: 20px;
|
| 358 |
-
}}
|
| 359 |
-
.fun-stats h3 {{
|
| 360 |
-
font-size: 28px;
|
| 361 |
-
color: inherit;
|
| 362 |
-
margin-top: 30px;
|
| 363 |
-
margin-bottom: 15px;
|
| 364 |
-
border-bottom: 2px solid currentColor;
|
| 365 |
-
padding-bottom: 10px;
|
| 366 |
-
}}
|
| 367 |
-
.fun-stats ul {{
|
| 368 |
-
list-style-type: none;
|
| 369 |
-
padding-left: 0;
|
| 370 |
-
}}
|
| 371 |
-
.fun-stats li {{
|
| 372 |
-
margin-bottom: 15px;
|
| 373 |
-
padding: 15px;
|
| 374 |
-
border-radius: 5px;
|
| 375 |
-
box-shadow: 0 2px 5px rgba(0,0,0,0.1);
|
| 376 |
-
}}
|
| 377 |
-
.fun-stats .timestamp {{
|
| 378 |
-
font-style: italic;
|
| 379 |
-
text-align: center;
|
| 380 |
-
margin-bottom: 20px;
|
| 381 |
-
}}
|
| 382 |
-
.fun-stats .highlight {{
|
| 383 |
-
font-weight: bold;
|
| 384 |
-
color: #e74c3c;
|
| 385 |
-
}}
|
| 386 |
-
</style>
|
| 387 |
-
<div class="fun-stats">
|
| 388 |
-
<h2>🎭 Fun Arena Stats 🎭</h2>
|
| 389 |
-
<p class="timestamp">Last updated: {stats['timestamp']}</p>
|
| 390 |
-
|
| 391 |
-
<h3>🏟️ Arena Overview</h3>
|
| 392 |
-
<p>Total Battles Fought: <span class="highlight">{stats['total_battles']}</span></p>
|
| 393 |
-
<p>Active Gladiators (Models): <span class="highlight">{stats['active_models']}</span></p>
|
| 394 |
-
|
| 395 |
-
<h3>🏆 Hall of Fame</h3>
|
| 396 |
-
<p>🥇 Battle Veteran: <span class="highlight">{stats['most_battles']['model']}</span> ({stats['most_battles']['battles']} battles)</p>
|
| 397 |
-
<p>🏹 Sharpshooter: <span class="highlight">{stats['highest_win_rate']['model']}</span> (Win Rate: {stats['highest_win_rate']['win_rate']})</p>
|
| 398 |
-
<p>🌈 Jack of All Trades: <span class="highlight">{stats['most_diverse_opponent']['model']}</span> (Faced {stats['most_diverse_opponent']['unique_opponents']} unique opponents)</p>
|
| 399 |
-
<p>🐕 Underdog Champion: <span class="highlight">{stats['underdog_champion']['model']}</span> ({stats['underdog_champion']['size']} model with {stats['underdog_champion']['win_rate']} win rate)</p>
|
| 400 |
-
<p>⚖️ Mr. Consistent: <span class="highlight">{stats['most_consistent']['model']}</span> (Closest to 50% win rate, difference of {stats['most_consistent']['win_loss_difference']} wins/losses)</p>
|
| 401 |
-
|
| 402 |
-
<h3>🤼 Epic Battles</h3>
|
| 403 |
-
<p>🤼 Biggest Rivalry: <span class="highlight">{stats['biggest_rivalry']['model1']}</span> vs <span class="highlight">{stats['biggest_rivalry']['model2']}</span> ({stats['biggest_rivalry']['total_battles']} fierce battles!)</p>
|
| 404 |
-
<p>🏋️ David vs Goliath: <span class="highlight">{stats['david_vs_goliath']['david']}</span> (David) vs <span class="highlight">{stats['david_vs_goliath']['goliath']}</span> (Goliath)<br>
|
| 405 |
-
David won {stats['david_vs_goliath']['wins']} times despite being {stats['david_vs_goliath']['size_difference']} smaller!</p>
|
| 406 |
-
<p>🔄 Comeback King: <span class="highlight">{stats['comeback_king']['model']}</span> (Overcame a {stats['comeback_king']['comeback_margin']}-battle deficit)</p>
|
| 407 |
-
<p>🏆 Pyrrhic Victor: <span class="highlight">{stats['pyrrhic_victor']['model']}</span> (Lowest win rate among models with more wins than losses: {stats['pyrrhic_victor']['win_rate']})</p>
|
| 408 |
-
</div>
|
| 409 |
-
"""
|
| 410 |
-
|
| 411 |
-
return html
|
| 412 |
-
|
| 413 |
-
def update_fun_stats_periodically(interval):
|
| 414 |
-
while True:
|
| 415 |
-
time.sleep(interval)
|
| 416 |
-
fun_stats_html.update(value=get_fun_stats_html())
|
| 417 |
-
|
| 418 |
# Initialize Gradio Blocks
|
| 419 |
with gr.Blocks(css="""
|
| 420 |
#dice-button {
|
|
@@ -422,8 +342,8 @@ with gr.Blocks(css="""
|
|
| 422 |
font-size: 35px;
|
| 423 |
}
|
| 424 |
""") as demo:
|
| 425 |
-
gr.Markdown(
|
| 426 |
-
gr.Markdown(
|
| 427 |
|
| 428 |
# Leaderboard Tab (now first)
|
| 429 |
with gr.Tab("Leaderboard"):
|
|
@@ -450,8 +370,8 @@ with gr.Blocks(css="""
|
|
| 450 |
submit_btn = gr.Button("Generate Responses", variant="primary")
|
| 451 |
|
| 452 |
with gr.Row():
|
| 453 |
-
left_output = gr.Chatbot(label=random.choice(
|
| 454 |
-
right_output = gr.Chatbot(label=random.choice(
|
| 455 |
|
| 456 |
with gr.Row():
|
| 457 |
left_vote_btn = gr.Button(f"Vote for {left_output.label}", interactive=False)
|
|
@@ -482,11 +402,6 @@ with gr.Blocks(css="""
|
|
| 482 |
with gr.Tab("ELO Leaderboard"):
|
| 483 |
elo_leaderboard = gr.HTML(label="ELO Leaderboard")
|
| 484 |
|
| 485 |
-
# Add this new tab
|
| 486 |
-
with gr.Tab("Fun Stats"):
|
| 487 |
-
refresh_btn = gr.Button("Refresh Stats")
|
| 488 |
-
fun_stats_html = gr.HTML(label="Fun Arena Stats")
|
| 489 |
-
|
| 490 |
# Add this new tab
|
| 491 |
with gr.Tab("Suggest Models"):
|
| 492 |
with gr.Row():
|
|
@@ -564,16 +479,10 @@ with gr.Blocks(css="""
|
|
| 564 |
demo.load(get_leaderboard, outputs=leaderboard)
|
| 565 |
demo.load(get_elo_leaderboard, outputs=elo_leaderboard)
|
| 566 |
demo.load(get_leaderboard_chart, outputs=leaderboard_chart)
|
| 567 |
-
demo.load(get_fun_stats_html, outputs=fun_stats_html)
|
| 568 |
-
|
| 569 |
-
# Add this event handler for the refresh button
|
| 570 |
-
refresh_btn.click(get_fun_stats_html, outputs=fun_stats_html)
|
| 571 |
-
|
| 572 |
-
# Start the background task to update stats every hour
|
| 573 |
-
update_thread = threading.Thread(target=update_fun_stats_periodically, args=(3600,), daemon=True)
|
| 574 |
-
update_thread.start()
|
| 575 |
|
| 576 |
if __name__ == "__main__":
|
| 577 |
# Initialize ELO ratings before launching the app
|
| 578 |
ensure_elo_ratings_initialized()
|
|
|
|
|
|
|
| 579 |
demo.launch(show_api=False)
|
|
|
|
| 3 |
import random
|
| 4 |
import requests
|
| 5 |
import logging
|
| 6 |
+
import config
|
| 7 |
import plotly.graph_objects as go
|
| 8 |
from typing import Dict
|
| 9 |
from leaderboard import (
|
|
|
|
| 15 |
ensure_elo_ratings_initialized
|
| 16 |
)
|
| 17 |
import sys
|
| 18 |
+
import openai
|
| 19 |
import threading
|
| 20 |
import time
|
| 21 |
from collections import Counter
|
|
|
|
| 31 |
|
| 32 |
# Function to get available models (using predefined list)
|
| 33 |
def get_available_models():
|
| 34 |
+
return [model[0] for model in config.get_approved_models()]
|
| 35 |
|
| 36 |
# Function to get recent opponents for a model
|
| 37 |
recent_opponents = {}
|
|
|
|
| 46 |
# Function to call Ollama API with caching
|
| 47 |
@lru_cache(maxsize=100)
|
| 48 |
def call_ollama_api(model, prompt):
|
| 49 |
+
client = openai.OpenAI(
|
| 50 |
+
api_key=config.API_KEY,
|
| 51 |
+
base_url=config.API_URL
|
| 52 |
+
)
|
| 53 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
try:
|
| 55 |
+
response = client.chat.completions.create(
|
| 56 |
+
model=model,
|
| 57 |
+
messages=[
|
| 58 |
+
{
|
| 59 |
+
"role": "system",
|
| 60 |
+
"content": "You are a helpful assistant. At no point should you reveal your name, identity or team affiliation to the user, especially if asked directly!"
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"role": "user",
|
| 64 |
+
"content": prompt
|
| 65 |
+
}
|
| 66 |
+
],
|
| 67 |
timeout=100
|
| 68 |
)
|
| 69 |
+
return response.choices[0].message.content
|
| 70 |
+
except Exception as e:
|
|
|
|
|
|
|
| 71 |
logger.error(f"Error calling Ollama API for model {model}: {e}")
|
| 72 |
return f"Error: Unable to get response from the model."
|
| 73 |
|
|
|
|
| 130 |
gr.update(value="Error: Unable to get response from the model", visible=True)
|
| 131 |
)
|
| 132 |
|
| 133 |
+
nickname_a = random.choice(config.model_nicknames)
|
| 134 |
+
nickname_b = random.choice(config.model_nicknames)
|
| 135 |
|
| 136 |
# Format responses for gr.Chatbot, including the user's prompt
|
| 137 |
response_a_formatted = [
|
|
|
|
| 281 |
return fig
|
| 282 |
|
| 283 |
def new_battle():
|
| 284 |
+
nickname_a = random.choice(config.model_nicknames)
|
| 285 |
+
nickname_b = random.choice(config.model_nicknames)
|
| 286 |
return (
|
| 287 |
"", # Reset prompt_input
|
| 288 |
gr.update(value=[], label=nickname_a), # Reset left Chatbot
|
|
|
|
| 301 |
|
| 302 |
# Add this new function
|
| 303 |
def get_human_readable_name(model_name: str) -> str:
|
| 304 |
+
model_dict = dict(config.get_approved_models())
|
| 305 |
return model_dict.get(model_name, model_name)
|
| 306 |
|
| 307 |
# Add this new function to randomly select a prompt
|
| 308 |
def random_prompt():
|
| 309 |
+
return random.choice(config.example_prompts)
|
| 310 |
|
| 311 |
# Modify the continue_conversation function
|
| 312 |
def continue_conversation(prompt, left_chat, right_chat, left_model, right_model, previous_prompt, tie_count):
|
| 313 |
# Check if the prompt is empty or the same as the previous one
|
| 314 |
if not prompt or prompt == previous_prompt:
|
| 315 |
+
prompt = random.choice(config.example_prompts)
|
| 316 |
|
| 317 |
left_response = call_ollama_api(left_model, prompt)
|
| 318 |
right_response = call_ollama_api(right_model, prompt)
|
|
|
|
| 335 |
tie_count
|
| 336 |
)
|
| 337 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 338 |
# Initialize Gradio Blocks
|
| 339 |
with gr.Blocks(css="""
|
| 340 |
#dice-button {
|
|
|
|
| 342 |
font-size: 35px;
|
| 343 |
}
|
| 344 |
""") as demo:
|
| 345 |
+
gr.Markdown(config.ARENA_NAME)
|
| 346 |
+
gr.Markdown(config.ARENA_DESCRIPTION)
|
| 347 |
|
| 348 |
# Leaderboard Tab (now first)
|
| 349 |
with gr.Tab("Leaderboard"):
|
|
|
|
| 370 |
submit_btn = gr.Button("Generate Responses", variant="primary")
|
| 371 |
|
| 372 |
with gr.Row():
|
| 373 |
+
left_output = gr.Chatbot(label=random.choice(config.model_nicknames), type="messages")
|
| 374 |
+
right_output = gr.Chatbot(label=random.choice(config.model_nicknames), type="messages")
|
| 375 |
|
| 376 |
with gr.Row():
|
| 377 |
left_vote_btn = gr.Button(f"Vote for {left_output.label}", interactive=False)
|
|
|
|
| 402 |
with gr.Tab("ELO Leaderboard"):
|
| 403 |
elo_leaderboard = gr.HTML(label="ELO Leaderboard")
|
| 404 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 405 |
# Add this new tab
|
| 406 |
with gr.Tab("Suggest Models"):
|
| 407 |
with gr.Row():
|
|
|
|
| 479 |
demo.load(get_leaderboard, outputs=leaderboard)
|
| 480 |
demo.load(get_elo_leaderboard, outputs=elo_leaderboard)
|
| 481 |
demo.load(get_leaderboard_chart, outputs=leaderboard_chart)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 482 |
|
| 483 |
if __name__ == "__main__":
|
| 484 |
# Initialize ELO ratings before launching the app
|
| 485 |
ensure_elo_ratings_initialized()
|
| 486 |
+
# Start the model refresh thread
|
| 487 |
+
config.start_model_refresh_thread()
|
| 488 |
demo.launch(show_api=False)
|