deberta-v2-xxlarge / ds_config.json
Pengcheng He
Update deepspeed config
a642358
raw
history blame
294 Bytes
{
"fp16": {
"enabled": true,
"initial_scale_power": 12
},
"zero_optimization": {
"stage": 2,
"reduce_bucket_size": 5e7,
"allgather_bucket_size": 1.25e9,
"overlap_comm": true,
"contiguous_gradients": true
},
"zero_allow_untested_optimizer": true
}