Spaces:
Runtime error
Runtime error
elo calculations and update arena metadta
Browse files- README.md +8 -2
- calculate_elo.py +276 -0
README.md
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
emoji: π
|
| 4 |
colorFrom: yellow
|
| 5 |
colorTo: blue
|
|
@@ -10,4 +10,10 @@ pinned: false
|
|
| 10 |
license: apache-2.0
|
| 11 |
---
|
| 12 |
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
title: Community ChatBot Arena
|
| 3 |
emoji: π
|
| 4 |
colorFrom: yellow
|
| 5 |
colorTo: blue
|
|
|
|
| 10 |
license: apache-2.0
|
| 11 |
---
|
| 12 |
|
| 13 |
+
# OpenAccess AI Collective Community ChatBot Arena
|
| 14 |
+
|
| 15 |
+
- Arena: https://huggingface.co/spaces/openaccess-ai-collective/rlhf-arena
|
| 16 |
+
- GitHub: https://github.com/OpenAccess-AI-Collective/rlhf-arena
|
| 17 |
+
- Built using Runpod Serverless. See our writeup here: https://medium.com/@winglian/inference-any-llm-with-serverless-in-15-minutes-69eeb548a41d
|
| 18 |
+
- Want to have your language model added to the Arena? [Create an Issue](https://github.com/OpenAccess-AI-Collective/rlhf-arena/issues) or reach out on [Discord](https://discord.gg/PugNNHAF5r)
|
| 19 |
+
- [π΅ Consider Donating on our Patreon](http://patreon.com/OpenAccessAICollective)
|
calculate_elo.py
ADDED
|
@@ -0,0 +1,276 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import logging
|
| 2 |
+
import os
|
| 3 |
+
from datetime import datetime
|
| 4 |
+
from decimal import Decimal
|
| 5 |
+
|
| 6 |
+
import boto3
|
| 7 |
+
from boto3.dynamodb.conditions import Attr
|
| 8 |
+
|
| 9 |
+
logging.basicConfig(level=os.getenv("LOG_LEVEL", "INFO"))
|
| 10 |
+
|
| 11 |
+
# Create a DynamoDB client
|
| 12 |
+
dynamodb = boto3.resource('dynamodb', region_name='us-east-1')
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def _create_arena_table():
|
| 16 |
+
table = dynamodb.create_table(
|
| 17 |
+
TableName='oaaic_chatbot_arena',
|
| 18 |
+
KeySchema=[
|
| 19 |
+
{
|
| 20 |
+
'AttributeName': 'arena_battle_id',
|
| 21 |
+
'KeyType': 'HASH'
|
| 22 |
+
},
|
| 23 |
+
],
|
| 24 |
+
AttributeDefinitions=[
|
| 25 |
+
{
|
| 26 |
+
'AttributeName': 'arena_battle_id',
|
| 27 |
+
'AttributeType': 'S'
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
'AttributeName': 'timestamp',
|
| 31 |
+
'AttributeType': 'S'
|
| 32 |
+
},
|
| 33 |
+
],
|
| 34 |
+
ProvisionedThroughput={
|
| 35 |
+
'ReadCapacityUnits': 5,
|
| 36 |
+
'WriteCapacityUnits': 5
|
| 37 |
+
},
|
| 38 |
+
GlobalSecondaryIndexes=[
|
| 39 |
+
{
|
| 40 |
+
'IndexName': 'TimestampIndex',
|
| 41 |
+
'KeySchema': [
|
| 42 |
+
{
|
| 43 |
+
'AttributeName': 'arena_battle_id',
|
| 44 |
+
'KeyType': 'HASH'
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
'AttributeName': 'timestamp',
|
| 48 |
+
'KeyType': 'RANGE'
|
| 49 |
+
},
|
| 50 |
+
],
|
| 51 |
+
'Projection': {
|
| 52 |
+
'ProjectionType': 'ALL',
|
| 53 |
+
},
|
| 54 |
+
'ProvisionedThroughput': {
|
| 55 |
+
'ReadCapacityUnits': 5,
|
| 56 |
+
'WriteCapacityUnits': 5,
|
| 57 |
+
}
|
| 58 |
+
},
|
| 59 |
+
]
|
| 60 |
+
)
|
| 61 |
+
|
| 62 |
+
def _create_elo_scores_table():
|
| 63 |
+
dynamodb.create_table(
|
| 64 |
+
TableName='elo_scores',
|
| 65 |
+
KeySchema=[
|
| 66 |
+
{
|
| 67 |
+
'AttributeName': 'chatbot_name',
|
| 68 |
+
'KeyType': 'HASH' # Partition key
|
| 69 |
+
},
|
| 70 |
+
],
|
| 71 |
+
AttributeDefinitions=[
|
| 72 |
+
{
|
| 73 |
+
'AttributeName': 'chatbot_name',
|
| 74 |
+
'AttributeType': 'S'
|
| 75 |
+
},
|
| 76 |
+
],
|
| 77 |
+
ProvisionedThroughput={
|
| 78 |
+
'ReadCapacityUnits': 5,
|
| 79 |
+
'WriteCapacityUnits': 5
|
| 80 |
+
}
|
| 81 |
+
)
|
| 82 |
+
|
| 83 |
+
|
| 84 |
+
def _create_elo_logs_table():
|
| 85 |
+
dynamodb.create_table(
|
| 86 |
+
TableName='elo_logs',
|
| 87 |
+
KeySchema=[
|
| 88 |
+
{
|
| 89 |
+
'AttributeName': 'arena_battle_id',
|
| 90 |
+
'KeyType': 'HASH' # Partition key
|
| 91 |
+
},
|
| 92 |
+
],
|
| 93 |
+
AttributeDefinitions=[
|
| 94 |
+
{
|
| 95 |
+
'AttributeName': 'arena_battle_id',
|
| 96 |
+
'AttributeType': 'S'
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
'AttributeName': 'battle_timestamp',
|
| 100 |
+
'AttributeType': 'S'
|
| 101 |
+
},
|
| 102 |
+
],
|
| 103 |
+
ProvisionedThroughput={
|
| 104 |
+
'ReadCapacityUnits': 10,
|
| 105 |
+
'WriteCapacityUnits': 10
|
| 106 |
+
},
|
| 107 |
+
GlobalSecondaryIndexes=[
|
| 108 |
+
{
|
| 109 |
+
'IndexName': 'BattleTimestampIndex',
|
| 110 |
+
'KeySchema': [
|
| 111 |
+
{
|
| 112 |
+
'AttributeName': 'battle_timestamp',
|
| 113 |
+
'KeyType': 'HASH' # Partition key for the GSI
|
| 114 |
+
},
|
| 115 |
+
],
|
| 116 |
+
'Projection': {
|
| 117 |
+
'ProjectionType': 'ALL'
|
| 118 |
+
},
|
| 119 |
+
'ProvisionedThroughput': {
|
| 120 |
+
'ReadCapacityUnits': 10,
|
| 121 |
+
'WriteCapacityUnits': 10
|
| 122 |
+
}
|
| 123 |
+
},
|
| 124 |
+
]
|
| 125 |
+
)
|
| 126 |
+
|
| 127 |
+
|
| 128 |
+
def get_unprocessed_battles(last_processed_timestamp):
|
| 129 |
+
# Use boto3 to create a DynamoDB resource and reference the table
|
| 130 |
+
table = dynamodb.Table('oaaic_chatbot_arena')
|
| 131 |
+
|
| 132 |
+
# Use a query to retrieve unprocessed battles in temporal order
|
| 133 |
+
response = table.scan(
|
| 134 |
+
FilterExpression=Attr('timestamp').gt(last_processed_timestamp),
|
| 135 |
+
# ScanIndexForward=True
|
| 136 |
+
)
|
| 137 |
+
|
| 138 |
+
return response['Items']
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
def calculate_elo(rating1, rating2, result, K=32):
|
| 142 |
+
# Convert ratings to float
|
| 143 |
+
rating1 = float(rating1)
|
| 144 |
+
rating2 = float(rating2)
|
| 145 |
+
|
| 146 |
+
# Calculate the expected outcomes
|
| 147 |
+
expected_outcome1 = 1.0 / (1.0 + 10.0 ** ((rating2 - rating1) / 400.0))
|
| 148 |
+
expected_outcome2 = 1.0 - expected_outcome1
|
| 149 |
+
|
| 150 |
+
# Calculate the new Elo ratings
|
| 151 |
+
new_rating1 = rating1 + K * (result - expected_outcome1)
|
| 152 |
+
new_rating2 = rating2 + K * ((1.0 - result) - expected_outcome2)
|
| 153 |
+
|
| 154 |
+
return Decimal(new_rating1).quantize(Decimal('0.00')), Decimal(new_rating2).quantize(Decimal('0.00'))
|
| 155 |
+
|
| 156 |
+
|
| 157 |
+
def get_last_processed_timestamp():
|
| 158 |
+
table = dynamodb.Table('elo_logs')
|
| 159 |
+
|
| 160 |
+
response = table.update (
|
| 161 |
+
AttributeDefinitions=[
|
| 162 |
+
{
|
| 163 |
+
'AttributeName': 'timestamp',
|
| 164 |
+
'AttributeType': 'S'
|
| 165 |
+
},
|
| 166 |
+
],
|
| 167 |
+
GlobalSecondaryIndexUpdates=[
|
| 168 |
+
{
|
| 169 |
+
'Create': {
|
| 170 |
+
'IndexName': 'TimestampIndex',
|
| 171 |
+
'KeySchema': [
|
| 172 |
+
{
|
| 173 |
+
'AttributeName': 'timestamp',
|
| 174 |
+
'KeyType': 'RANGE'
|
| 175 |
+
},
|
| 176 |
+
],
|
| 177 |
+
'Projection': {
|
| 178 |
+
'ProjectionType': 'ALL',
|
| 179 |
+
}
|
| 180 |
+
},
|
| 181 |
+
},
|
| 182 |
+
]
|
| 183 |
+
)
|
| 184 |
+
|
| 185 |
+
# Scan the table sorted by timestamp in descending order
|
| 186 |
+
response = table.scan(
|
| 187 |
+
Limit=1,
|
| 188 |
+
ScanIndexForward=False
|
| 189 |
+
)
|
| 190 |
+
|
| 191 |
+
# If there are no items in the table, return a default timestamp
|
| 192 |
+
if not response['Items']:
|
| 193 |
+
return '1970-01-01T00:00:00'
|
| 194 |
+
|
| 195 |
+
# Otherwise, return the timestamp of the latest item
|
| 196 |
+
return response['Items'][0]['battle_timestamp']
|
| 197 |
+
|
| 198 |
+
|
| 199 |
+
def log_elo_update(arena_battle_id, battle_timestamp, new_rating1, new_rating2):
|
| 200 |
+
# Reference the elo_logs table
|
| 201 |
+
table = dynamodb.Table('elo_logs')
|
| 202 |
+
|
| 203 |
+
# Update the table
|
| 204 |
+
table.put_item(
|
| 205 |
+
Item={
|
| 206 |
+
'arena_battle_id': arena_battle_id,
|
| 207 |
+
'battle_timestamp': battle_timestamp, # Use the timestamp of the battle
|
| 208 |
+
'log_timestamp': datetime.now().isoformat(), # Also store the timestamp of the log for completeness
|
| 209 |
+
'new_rating1': new_rating1,
|
| 210 |
+
'new_rating2': new_rating2
|
| 211 |
+
}
|
| 212 |
+
)
|
| 213 |
+
|
| 214 |
+
|
| 215 |
+
def get_elo_score(chatbot_name, elo_scores):
|
| 216 |
+
if chatbot_name in elo_scores:
|
| 217 |
+
return elo_scores[chatbot_name]
|
| 218 |
+
|
| 219 |
+
table = dynamodb.Table('elo_scores')
|
| 220 |
+
response = table.get_item(Key={'chatbot_name': chatbot_name})
|
| 221 |
+
|
| 222 |
+
# If there is no item in the table, return a default score
|
| 223 |
+
if 'Item' not in response:
|
| 224 |
+
return 1500
|
| 225 |
+
|
| 226 |
+
return response['Item']['elo_score']
|
| 227 |
+
|
| 228 |
+
|
| 229 |
+
def update_elo_score(chatbot_name, new_elo_score):
|
| 230 |
+
table = dynamodb.Table('elo_scores')
|
| 231 |
+
|
| 232 |
+
# This will create a new item if it doesn't exist
|
| 233 |
+
table.put_item(
|
| 234 |
+
Item={
|
| 235 |
+
'chatbot_name': chatbot_name,
|
| 236 |
+
'elo_score': Decimal(str(new_elo_score)),
|
| 237 |
+
}
|
| 238 |
+
)
|
| 239 |
+
|
| 240 |
+
|
| 241 |
+
def main():
|
| 242 |
+
# last_processed_timestamp = get_last_processed_timestamp()
|
| 243 |
+
last_processed_timestamp = '1970-01-01T00:00:00'
|
| 244 |
+
battles = get_unprocessed_battles(last_processed_timestamp)
|
| 245 |
+
|
| 246 |
+
elo_scores = {}
|
| 247 |
+
|
| 248 |
+
for battle in battles:
|
| 249 |
+
if battle['label'] in {0, 1, 2}:
|
| 250 |
+
outcome = battle['label']
|
| 251 |
+
for chatbot_name in [battle['choice1_name'], battle['choice2_name']]:
|
| 252 |
+
if chatbot_name not in elo_scores:
|
| 253 |
+
elo_scores[chatbot_name] = get_elo_score(chatbot_name, elo_scores)
|
| 254 |
+
# 1: This means that the first player (or team) won the match.
|
| 255 |
+
# 0.5: This means that the match ended in a draw.
|
| 256 |
+
# 0: This means that the first player (or team) lost the match.
|
| 257 |
+
if outcome == 0:
|
| 258 |
+
elo_result = 0.5
|
| 259 |
+
elif outcome == 1:
|
| 260 |
+
elo_result = 1
|
| 261 |
+
else:
|
| 262 |
+
elo_result = 0
|
| 263 |
+
|
| 264 |
+
new_rating1, new_rating2 = calculate_elo(elo_scores[battle['choice1_name']], elo_scores[battle['choice2_name']], elo_result)
|
| 265 |
+
elo_scores[battle['choice1_name']] = new_rating1
|
| 266 |
+
elo_scores[battle['choice2_name']] = new_rating2
|
| 267 |
+
log_elo_update(battle['arena_battle_id'], battle['timestamp'], new_rating1, new_rating2)
|
| 268 |
+
logging.info(f"{battle['choice1_name']}: {elo_scores[battle['choice1_name']]} -> {new_rating1} | {battle['choice2_name']}: {elo_scores[battle['choice2_name']]} -> {new_rating2}")
|
| 269 |
+
update_elo_score(battle['choice1_name'], new_rating1)
|
| 270 |
+
update_elo_score(battle['choice2_name'], new_rating2)
|
| 271 |
+
elo_scores[battle['choice1_name']] = new_rating1
|
| 272 |
+
elo_scores[battle['choice2_name']] = new_rating2
|
| 273 |
+
|
| 274 |
+
|
| 275 |
+
if __name__ == "__main__":
|
| 276 |
+
main()
|