flash-attention/training/configs/experiment/pile/gpt3s-hf.yaml
2022-11-29 04:46:43 -08:00

13 lines
232 B
YAML

# @package _global_
defaults:
- /experiment/pile/base.yaml
- override /model: gpt2-hf
- override /model/gpt2model: gpt2-small
datamodule:
batch_size: 8
train:
# Use the standard torch.nn.CrossEntropyLoss
loss_fn: null