yaml | |
{ | |
"zero_optimization": { | |
"stage3_gather_16bit_weights_on_model_save": true | |
} | |
} | |
The full precision weights shouldn't be saved during training because it can require a lot of memory. |
yaml | |
{ | |
"zero_optimization": { | |
"stage3_gather_16bit_weights_on_model_save": true | |
} | |
} | |
The full precision weights shouldn't be saved during training because it can require a lot of memory. |