pretrain core 0

scripts/pretrain-core-model-0.yaml

@@ -113,8 +113,8 @@ eval:
 # Optimizer-related arguments
 
 optimizer:
-  class_path: torch.optim.AdamW
-  # class_path: torchao.prototype.low_bit_optim.AdamW8bit
+  # class_path: torch.optim.AdamW
+  class_path: torchao.prototype.low_bit_optim.AdamW8bit
   # class_path: torchao.prototype.low_bit_optim.AdamW4bit
   # class_path: bitsandbytes.optim.AdamW8bit
   # class_path: bitsandbytes.optim.PagedAdamW8bit