mars_520m_8 / README.md
KaiyueWen's picture
Upload folder using huggingface_hub
f574dc2 verified

Model Card

Best configuration

Hyperparameter Value
beta1 0.95
beta2 0.98
epsilon 9.999999999999999e-26
gamma 0.025
learning_rate 0.004
max_grad_norm 1
min_lr_ratio 0
train_batch_size 256
warmup 2000
weight_decay 0.1