Datasets:
Tasks:
Feature Extraction
Formats:
parquet
Languages:
English
Size:
1M - 10M
Tags:
mechanistic-interpretability
interpretability
activation-atlas
feature-extraction
atlas
dataset
License:
Download layers/14/component_comparison.json from juiceb0xc0de/Qwen3.5-9B-Base: direct link, hf CLI and curl.
- Browser
- Download file 5.03 kB
-
https://huggingface.co/datasets/juiceb0xc0de/Qwen3.5-9B-Base/resolve/main/layers/14/component_comparison.json
- Command line
-
hf download hf://datasets/juiceb0xc0de/Qwen3.5-9B-Base/layers/14/component_comparison.json
-
curl -L -o component_comparison.json https://huggingface.co/datasets/juiceb0xc0de/Qwen3.5-9B-Base/resolve/main/layers/14/component_comparison.json
5.03 kB
| [ | |
| { | |
| "name": "mlp", | |
| "pooling": "mean", | |
| "shape": [ | |
| 12288, | |
| 8965 | |
| ], | |
| "taxonomy": { | |
| "broadly_shared": 5592, | |
| "partial_shared": 5774, | |
| "all_shared": 606, | |
| "non_activated": 316 | |
| }, | |
| "top_sep_score": 228.7849884033203, | |
| "mean_sep_score": 25.91461181640625, | |
| "top_bucket_quality": 0.1871125642655257, | |
| "top_bucket_feature": 11675, | |
| "top_bucket": "humor", | |
| "bucket_entropy_at_top": 0.21833633653158524, | |
| "eta_squared_at_top": 0.23937733453703836, | |
| "n_contrast_pairs": 0, | |
| "top_contrast_feature": null, | |
| "top_contrast_effect": null, | |
| "n_active": 12279, | |
| "mean": -0.00015965047350618988, | |
| "std": 0.0380469374358654, | |
| "non_zero_var": 12288, | |
| "n_coact_pairs": 0, | |
| "code_entangled": 2, | |
| "code_selective": 28, | |
| "code_bucket": "core_technical" | |
| }, | |
| { | |
| "name": "gate", | |
| "pooling": "mean", | |
| "shape": [ | |
| 12288, | |
| 8965 | |
| ], | |
| "taxonomy": { | |
| "partial_shared": 4697, | |
| "broadly_shared": 1393, | |
| "non_activated": 5443, | |
| "all_shared": 753, | |
| "specific": 2 | |
| }, | |
| "top_sep_score": 359.40057373046875, | |
| "mean_sep_score": 43.128997802734375, | |
| "top_bucket_quality": 0.3246120472556045, | |
| "top_bucket_feature": 2228, | |
| "top_bucket": "data", | |
| "bucket_entropy_at_top": -0.0, | |
| "eta_squared_at_top": 0.3246120472556045, | |
| "n_contrast_pairs": 0, | |
| "top_contrast_feature": null, | |
| "top_contrast_effect": null, | |
| "n_active": 11343, | |
| "mean": -0.06530282646417618, | |
| "std": 0.12949307262897491, | |
| "non_zero_var": 12288, | |
| "n_coact_pairs": 29, | |
| "code_entangled": 3, | |
| "code_selective": 27, | |
| "code_bucket": "core_technical" | |
| }, | |
| { | |
| "name": "up", | |
| "pooling": "mean", | |
| "shape": [ | |
| 12288, | |
| 8965 | |
| ], | |
| "taxonomy": { | |
| "partial_shared": 4965, | |
| "broadly_shared": 4405, | |
| "non_activated": 1179, | |
| "all_shared": 1739 | |
| }, | |
| "top_sep_score": 386.8780517578125, | |
| "mean_sep_score": 42.17911148071289, | |
| "top_bucket_quality": 0.25382282413606977, | |
| "top_bucket_feature": 10874, | |
| "top_bucket": "core_technical", | |
| "bucket_entropy_at_top": -0.0, | |
| "eta_squared_at_top": 0.25382282413606977, | |
| "n_contrast_pairs": 0, | |
| "top_contrast_feature": null, | |
| "top_contrast_effect": null, | |
| "n_active": 12120, | |
| "mean": -0.0029763057827949524, | |
| "std": 0.2417440265417099, | |
| "non_zero_var": 12288, | |
| "n_coact_pairs": 30, | |
| "code_entangled": 2, | |
| "code_selective": 28, | |
| "code_bucket": "core_technical" | |
| }, | |
| { | |
| "name": "linattn_qkv", | |
| "pooling": "mean", | |
| "shape": [ | |
| 8192, | |
| 8965 | |
| ], | |
| "taxonomy": { | |
| "partial_shared": 2598, | |
| "broadly_shared": 2285, | |
| "all_shared": 1833, | |
| "non_activated": 1476 | |
| }, | |
| "top_sep_score": 530.696044921875, | |
| "mean_sep_score": 37.63390350341797, | |
| "top_bucket_quality": 0.3068469784910621, | |
| "top_bucket_feature": 6172, | |
| "top_bucket": "ml_ai", | |
| "bucket_entropy_at_top": -0.0, | |
| "eta_squared_at_top": 0.3068469784910621, | |
| "n_contrast_pairs": 0, | |
| "top_contrast_feature": null, | |
| "top_contrast_effect": null, | |
| "n_active": 7663, | |
| "mean": -0.0006362433778122067, | |
| "std": 0.7889477610588074, | |
| "non_zero_var": 8192, | |
| "n_coact_pairs": 563, | |
| "code_entangled": 5, | |
| "code_selective": 25, | |
| "code_bucket": "core_technical" | |
| }, | |
| { | |
| "name": "linattn_z", | |
| "pooling": "mean", | |
| "shape": [ | |
| 4096, | |
| 8965 | |
| ], | |
| "taxonomy": { | |
| "all_shared": 1870, | |
| "non_activated": 1008, | |
| "partial_shared": 668, | |
| "broadly_shared": 550 | |
| }, | |
| "top_sep_score": 288.3899841308594, | |
| "mean_sep_score": 35.13544845581055, | |
| "top_bucket_quality": 0.2718312085088477, | |
| "top_bucket_feature": 588, | |
| "top_bucket": "introspection", | |
| "bucket_entropy_at_top": -0.0, | |
| "eta_squared_at_top": 0.2718312085088477, | |
| "n_contrast_pairs": 0, | |
| "top_contrast_feature": null, | |
| "top_contrast_effect": null, | |
| "n_active": 3554, | |
| "mean": 0.07169578969478607, | |
| "std": 1.2636417150497437, | |
| "non_zero_var": 4096, | |
| "n_coact_pairs": 56, | |
| "code_entangled": 3, | |
| "code_selective": 27, | |
| "code_bucket": "core_technical" | |
| }, | |
| { | |
| "name": "linattn_out", | |
| "pooling": "mean", | |
| "shape": [ | |
| 4096, | |
| 8965 | |
| ], | |
| "taxonomy": { | |
| "broadly_shared": 1458, | |
| "partial_shared": 1696, | |
| "all_shared": 589, | |
| "non_activated": 353 | |
| }, | |
| "top_sep_score": 149.74993896484375, | |
| "mean_sep_score": 25.803455352783203, | |
| "top_bucket_quality": 0.13257640135098067, | |
| "top_bucket_feature": 960, | |
| "top_bucket": "learning", | |
| "bucket_entropy_at_top": -0.0, | |
| "eta_squared_at_top": 0.13257640135098067, | |
| "n_contrast_pairs": 0, | |
| "top_contrast_feature": null, | |
| "top_contrast_effect": null, | |
| "n_active": 4080, | |
| "mean": 0.00008080543193500489, | |
| "std": 0.049896273761987686, | |
| "non_zero_var": 4096, | |
| "n_coact_pairs": 0, | |
| "code_entangled": 10, | |
| "code_selective": 20, | |
| "code_bucket": "core_technical" | |
| } | |
| ] |