m2l_2022_nlp / v1_config.json
Moreno La Quatra
training config v1 model
9afde18
raw
history blame contribute delete
235 Bytes
{
"batch_size": 64,
"mask_probability": 0.15,
"num_layers": 6,
"num_heads": 8,
"d_model": 128,
"d_ff": 256,
"p_dropout": 0.1,
"max_seq_len": 128,
"vocab_size": 25000,
"learning_rate": 3e-4,
"grad_clip_value": 1,
}