Datasets:

MInAlA
/

crosscoder-multilayer-split-activations

Tags:

mechanistic-interpretability

sparse-autoencoder

License:

Dataset card Files Files and versions

crosscoder-multilayer-split-activations / v2-differentlayer /qwen3-4b-ppo /singlelayer_separate /base_L23__aligned_L20 /plots

2.5 MB

1 contributor

History: 1 commit

ishangarg183's picture

Copy assembled_activations to v2-differentlayer (batch 7/12)

282d256 verified 7 days ago

cf_distribution_per_class.png

200 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
cf_shift_per_class.png

132 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
class_distribution.png

155 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
loss_curves.png

327 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
rho_histogram.png

81.5 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
rho_theta_scatter.png

354 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
shared_geometry_histograms.png

146 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
shared_geometry_theta_vs_norm_ratio.png

418 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago
superposition_analysis.png

685 kB
xet

Copy assembled_activations to v2-differentlayer (batch 7/12) 7 days ago