Closed Willmish closed 8 months ago
Adds both evaluation for batch sizes 2 and 8, of eval_framework_tasks and eval_harmfulness.
Training logs available: https://wandb.ai/ucl-willmish-team/snlp_batch_unlearn/table?nw=nwuserwillmish
Logs from training uploaded for batch unlearning 8, unfortunately overwritten for 2 (can download from wandb tho, the important bits (models, analysis, plots) are all here or huggingface (models)
I think it should all be fine now (feel free to double verify AJ, or just approve will wait forthat)
Adds both evaluation for batch sizes 2 and 8, of eval_framework_tasks and eval_harmfulness.
Training logs available: https://wandb.ai/ucl-willmish-team/snlp_batch_unlearn/table?nw=nwuserwillmish
Logs from training uploaded for batch unlearning 8, unfortunately overwritten for 2 (can download from wandb tho, the important bits (models, analysis, plots) are all here or huggingface (models)