pretrain mode
Browse files
scripts/pretrain-model.yaml
CHANGED
|
@@ -66,7 +66,8 @@ train:
|
|
| 66 |
global_batch_size: 512
|
| 67 |
|
| 68 |
# Number of samples per data-parallel rank (type: int, default: 4)
|
| 69 |
-
micro_batch_size: 16
|
|
|
|
| 70 |
|
| 71 |
# Number of iterations with learning rate warmup active (type: int, default: 2000)
|
| 72 |
lr_warmup_steps: 2000
|
|
|
|
| 66 |
global_batch_size: 512
|
| 67 |
|
| 68 |
# Number of samples per data-parallel rank (type: int, default: 4)
|
| 69 |
+
# micro_batch_size: 16
|
| 70 |
+
micro_batch_size: 4
|
| 71 |
|
| 72 |
# Number of iterations with learning rate warmup active (type: int, default: 2000)
|
| 73 |
lr_warmup_steps: 2000
|