Data Filtering on OpenWebText Quality
88.2Balanced AccuracyGPT-4o
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4o#Queries to GPT-4o=13.5M, Lightweight Model Cost=$0, Total Cost=$67,0002024.10 | 88.2 | 50 | |
| SIEVE#Queries to GPT-4o=60K, Lightweight Model Cost=$170, Total Cost=$4702024.10 | 86.3 | 53 | |
| GPT-3.5-Turbo#Queries to GPT-4o=13.5M, Lightweight Model Cost=$0, Total Cost=$13,0002024.10 | 74.5 | — |