Datasets:

MInAlA
/

crosscoder-multilayer-split-activations

Tags:

mechanistic-interpretability

sparse-autoencoder

License:

Dataset card Files Files and versions

crosscoder-multilayer-split-activations / v2-differentlayer /smollm3-dpo /singlelayer_separate /base_L20__aligned_L19 /plots

2.2 MB

1 contributor

History: 2 commits

ishangarg183's picture

Copy assembled_activations to v2-differentlayer (batch 9/12)

6364775 verified 8 days ago

cf_distribution_per_class.png

192 kB
xet

Copy assembled_activations to v2-differentlayer (batch 8/12) 8 days ago
cf_shift_per_class.png

127 kB
xet

Copy assembled_activations to v2-differentlayer (batch 8/12) 8 days ago
class_distribution.png

156 kB
xet

Copy assembled_activations to v2-differentlayer (batch 8/12) 8 days ago
loss_curves.png

327 kB
xet

Copy assembled_activations to v2-differentlayer (batch 9/12) 8 days ago
rho_histogram.png

86 kB
xet

Copy assembled_activations to v2-differentlayer (batch 9/12) 8 days ago
rho_theta_scatter.png

354 kB
xet

Copy assembled_activations to v2-differentlayer (batch 9/12) 8 days ago
shared_geometry_histograms.png

141 kB
xet

Copy assembled_activations to v2-differentlayer (batch 9/12) 8 days ago
shared_geometry_theta_vs_norm_ratio.png

434 kB
xet

Copy assembled_activations to v2-differentlayer (batch 9/12) 8 days ago
superposition_analysis.png

387 kB
xet

Copy assembled_activations to v2-differentlayer (batch 9/12) 8 days ago