kokolamba/keen_popqa_gpt2xl_generations

Upload qa accuracy data with avg and layer mlp outputs

Upload qa accuracy data with avg and layer residual stream(aka hidden state)

Add per-layer MinMaxScalers for mlp output (train-only) for GPT-2 XL: layers=range(33, 36)

Add per-layer MinMaxScalers for attention output (train-only) for GPT-2 XL: layers=range(33, 36)

Add top-k indices for gp2-xl model

Upload qa accuracy data with avg and layer logits

Add per-layer MinMaxScalers for logits (train-set-only) for GPT-2 XL: layers=[33, 34, 35]

Delete scalers/gpt2xl_keen_minmax_scalers_200_logits.joblib

Upload qa accuracy data with avg and layer top_200 logits

Add per-layer MinMaxScalers (train-only) for GPT-2 XL: layers=range(33, 36), top-k=200

Upload qa accuracy data with avg and layer residual stream(aka hidden state)

Delete features/residual_stream.parquet

Upload qa accuracy data with avg and layer residual stream(aka hidden state)

Add per-layer MinMaxScalers (train-only) for GPT-2 XL: layers=range(33, 36)

Delete scalers/gpt2xl_keen_minmax_scalers.joblib

Add per-layer MinMaxScalers (train-only) for GPT-2 XL: layers=range(33, 36)

Upload dataset

initial commit