train model
Browse files- scripts/model.yaml +3 -2
scripts/model.yaml
CHANGED
|
@@ -44,7 +44,7 @@ resume: "auto"
|
|
| 44 |
# Data-related arguments. If not provided, the default is ``litgpt.data.TinyLlama``.
|
| 45 |
data:
|
| 46 |
class_path: LitData
|
| 47 |
-
|
| 48 |
init_args:
|
| 49 |
data_path: "../data/"
|
| 50 |
num_workers: 16
|
|
@@ -110,8 +110,9 @@ eval:
|
|
| 110 |
# Optimizer-related arguments
|
| 111 |
optimizer:
|
| 112 |
# class_path: torch.optim.AdamW
|
|
|
|
| 113 |
# class_path: bitsandbytes.optim.PagedAdamW
|
| 114 |
-
class_path: bitsandbytes.optim.AdamW8bit
|
| 115 |
# class_path: bitsandbytes.optim.PagedAdamW8bit
|
| 116 |
|
| 117 |
init_args:
|
|
|
|
| 44 |
# Data-related arguments. If not provided, the default is ``litgpt.data.TinyLlama``.
|
| 45 |
data:
|
| 46 |
class_path: LitData
|
| 47 |
+
|
| 48 |
init_args:
|
| 49 |
data_path: "../data/"
|
| 50 |
num_workers: 16
|
|
|
|
| 110 |
# Optimizer-related arguments
|
| 111 |
optimizer:
|
| 112 |
# class_path: torch.optim.AdamW
|
| 113 |
+
class_path: grokadamw.GrokAdamW
|
| 114 |
# class_path: bitsandbytes.optim.PagedAdamW
|
| 115 |
+
# class_path: bitsandbytes.optim.AdamW8bit
|
| 116 |
# class_path: bitsandbytes.optim.PagedAdamW8bit
|
| 117 |
|
| 118 |
init_args:
|