Costa HuangandGitHub 8506aa0dbc add reward training script (#217) ...

* add reward trainer

* push changes

* format

* bug fix

* only do visualize in main process

* quick fix

* black

* update docs even more

* formatting

* fix eval logging

* add ds config

* maybe use ai2 wandb entity

* quick change

* remove unused print

* make the dataset column configurable; remove eos from prompt token;

* add a generate collator

2024-08-09 09:19:49 -07:00

443 KiB

2046x1240px

Raw Permalink History