LongHorizonTemperatureScaling
1.0.0
该存储库包含本文的代码:
长度温度缩放
由Andy Shih,Dorsa Sadigh,Stefano Ermon

pip install -r requirements.txt
在此配置文件中查看更多详细的配置。
WANDB_MODE=disabled torchrun --master_port=29601 --nproc_per_node=1 main_gpt2.py gpt_name=gpt2 horizon_loss.T=0.9
WANDB_MODE=disabled torchrun --master_port=29602 --nproc_per_node=1 main_gpt2.py gpt_name=gpt2-medium horizon_loss.T=0.9
WANDB_MODE=disabled torchrun --master_port=29603 --nproc_per_node=1 main_gpt2.py gpt_name=gpt2-large horizon_loss.T=0.9
如果您发现我们的工作有用,请考虑引用:
"Long Horizon Temperature Scaling"
Andy Shih, Dorsa Sadigh, Stefano Ermon
In Proceedings of the 40th International Conference on Machine Learning (ICML), 2023
@inproceedings{shih2023longhorizon,
author = {Andy Shih and Dorsa Sadigh and Stefano Ermon},
title = {Long Horizon Temperature Scaling},
booktitle = {Proceedings of the 40th International Conference on Machine Learning (ICML)},
month = {july},
year = {2023},
}