diff --git a/mlperf_logging/rcp_checker/training_4.1.0/rcps_stable_diffusion.json b/mlperf_logging/rcp_checker/training_4.1.0/rcps_stable_diffusion.json index 1173c11..2a35896 100644 --- a/mlperf_logging/rcp_checker/training_4.1.0/rcps_stable_diffusion.json +++ b/mlperf_logging/rcp_checker/training_4.1.0/rcps_stable_diffusion.json @@ -1,5 +1,29 @@ { + "sd_ref_256": + { + "Benchmark": "stable_diffusion", + "Creator": "NVIDIA", + "When": "Reference RCPs before v4.1", + "Platform": "4xDGX-H100", + "BS": 256, + "Hyperparams": { + "opt_adamw_beta_1": 0.9, + "opt_adamw_beta_2": 0.999, + "opt_adamw_epsilon": 1e-08, + "opt_adamw_weight_decay": 0.01, + "opt_base_learning_rate": 1.8e-7, + "opt_learning_rate_warmup_steps": 1000 + }, + "Epochs to converge": [ + 2048000, 2048000, 2048000, 2048000, + 2048000, 2048000, 2048000, 1536000, + 2048000, 2048000, 2048000, 2048000, + 2048000, 2048000, 2048000, 1536000, + 2048000, 2048000, 2560000, 2560000] + }, + + "sd_ref_384": { "Benchmark": "stable_diffusion",