Foundation Model Training on LAION 500M 5B
74.7Top-1 AccuracySwift Sampler (SS)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Swift Sampler (SS)Model Architecture=GPT-3, Batch Size=2048, Hardware=32 NVIDIA A100 GPUs, Sampling Strategy=Swift Sampler2024.10 | 74.7 | 92.3 | 125 | |
| Baseline (Uniform Sampling)Model Architecture=GPT-3, Batch Size=2048, Hardware=32 NVIDIA A100 GPUs, Sampling Strategy=Uniform Sampling2024.10 | 72.4 | 90.1 | 100 |