Hugging Face's logo

AlignmentResearch
/
diverse-deception-probe-olmo-3-32b-think

deception-detection
linear-probe
mechanistic-interpretability
Model card Files
xet
 Community
taufeeque commited on
Commit
8be9551
·
verified ·
1 Parent(s): 878c252

Add generation (no follow-up) probes — AUC 0.764

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. averaged_over_initial_detector.base_model.on_dataset_linear-probe_generation.txt +1 -0
  2. averaged_violin_initial_detector.base_model.on_dataset_linear-probe_generation.png +0 -0
  3. generation/layer_0/config.json +9 -0
  4. generation/layer_0/model.pt +3 -0
  5. generation/layer_1/config.json +9 -0
  6. generation/layer_1/model.pt +3 -0
  7. generation/layer_10/config.json +9 -0
  8. generation/layer_10/model.pt +3 -0
  9. generation/layer_11/config.json +9 -0
  10. generation/layer_11/model.pt +3 -0
  11. generation/layer_12/config.json +9 -0
  12. generation/layer_12/model.pt +3 -0
  13. generation/layer_13/config.json +9 -0
  14. generation/layer_13/model.pt +3 -0
  15. generation/layer_14/config.json +9 -0
  16. generation/layer_14/model.pt +3 -0
  17. generation/layer_15/config.json +9 -0
  18. generation/layer_15/model.pt +3 -0
  19. generation/layer_16/config.json +9 -0
  20. generation/layer_16/model.pt +3 -0
  21. generation/layer_17/config.json +9 -0
  22. generation/layer_17/model.pt +3 -0
  23. generation/layer_18/config.json +9 -0
  24. generation/layer_18/model.pt +3 -0
  25. generation/layer_19/config.json +9 -0
  26. generation/layer_19/model.pt +3 -0
  27. generation/layer_2/config.json +9 -0
  28. generation/layer_2/model.pt +3 -0
  29. generation/layer_20/config.json +9 -0
  30. generation/layer_20/model.pt +3 -0
  31. generation/layer_21/config.json +9 -0
  32. generation/layer_21/model.pt +3 -0
  33. generation/layer_22/config.json +9 -0
  34. generation/layer_22/model.pt +3 -0
  35. generation/layer_23/config.json +9 -0
  36. generation/layer_23/model.pt +3 -0
  37. generation/layer_24/config.json +9 -0
  38. generation/layer_24/model.pt +3 -0
  39. generation/layer_25/config.json +9 -0
  40. generation/layer_25/model.pt +3 -0
  41. generation/layer_26/config.json +9 -0
  42. generation/layer_26/model.pt +3 -0
  43. generation/layer_27/config.json +9 -0
  44. generation/layer_27/model.pt +3 -0
  45. generation/layer_28/config.json +9 -0
  46. generation/layer_28/model.pt +3 -0
  47. generation/layer_29/config.json +9 -0
  48. generation/layer_29/model.pt +3 -0
  49. generation/layer_3/config.json +9 -0
  50. generation/layer_3/model.pt +3 -0
averaged_over_initial_detector.base_model.on_dataset_linear-probe_generation.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ [17, 23, 24, 25, 26, 27, 28, 29]
averaged_violin_initial_detector.base_model.on_dataset_linear-probe_generation.png ADDED
generation/layer_0/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_0/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:672b65ca6bec27fdf36e3760811eab54a81b0ac84067eee5fa536bc3892d5c8f
3
+ size 23293
generation/layer_1/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_1/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:63de8ef6a539209dd08296b01f8d0addd017b9edac63e245aca5fd6371b1cf32
3
+ size 23293
generation/layer_10/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_10/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:796806c2411c8862a0570af034fd64805e536344dda83194995d91aa2e02660c
3
+ size 23293
generation/layer_11/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_11/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b07e1de24ba8dfff1036d3e55ec7c71417090e3d593a3cc2830f8efcd5819d1
3
+ size 23293
generation/layer_12/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_12/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19b070807bb06ebb6c8cf3cf607ce9e52f4b6dde89f3343c624c53533d963332
3
+ size 23293
generation/layer_13/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_13/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:31eda48c68b9e3864e810566ce5f3e2dac43188d2c61c0dd0b7cb3a038ff141d
3
+ size 23293
generation/layer_14/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_14/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30a2314d5e1bb87dedf1ef331cde4b6890fd946d3a5c6ba3a8da9b18168d87ae
3
+ size 23293
generation/layer_15/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_15/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06adf6a03628a57e0b053fd70dc09977542eb9ce2241a7d31912014e5ff53904
3
+ size 23293
generation/layer_16/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_16/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d8bfdb37a01d4a418eeb55d3b669e9f82a484baa15ced55d8d6f706380559b9a
3
+ size 23293
generation/layer_17/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_17/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6ee021d4c0ba4a4ca7b0d0cafe804cfe1475fcedaf3df9f740475a967b47a5b0
3
+ size 23293
generation/layer_18/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_18/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d21fa7390eb8dbbec6aa9599d34b3c138a67c0520a5eec0b352ad117a3a3f063
3
+ size 23293
generation/layer_19/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_19/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee1047267133616276cbd961b841ff898717f5d9800443fdc75c8ddf31c817e6
3
+ size 23293
generation/layer_2/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_2/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af075332d54b46dbe8c1ead6333b3e81764b96e56c195adaaf9a310820857b3c
3
+ size 23293
generation/layer_20/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_20/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a21640bbd99cfcfed14e1d71605c4307ef2eca58332a1fc13505f6afe136133
3
+ size 23293
generation/layer_21/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_21/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b052233950111d7d7a22a9fcd60f9672aebf83aef43af0c9ab05057e9fcbde8
3
+ size 23293
generation/layer_22/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_22/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6bea57b059aeb185c92ecaadc407f2a1dbbeffa7b65fc91d96f5485c6aaecdf
3
+ size 23293
generation/layer_23/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_23/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8d0b7c2bb8e8387ff2d53afec8f27a70ba6d8cd77f3bbd12762f0b4fae7b0ed
3
+ size 23293
generation/layer_24/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_24/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c4065c5f3b55542d6aeb6bf0d27c35dd4526d1d3b2a994cf137891062ee006c
3
+ size 23293
generation/layer_25/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_25/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:da2b2fffb58fc9ee529b29b1b6a95da3652df7ca9f9c92cf11188c3ffe8349b3
3
+ size 23293
generation/layer_26/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_26/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fba500a2ced31e94a71bf8ffe3eaf92b7a4556cb573131b409fc016792e6f011
3
+ size 23293
generation/layer_27/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_27/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3df63443a4cb37421ad60dde6747b1d1cc2234be898af6ea325d49d2b872520f
3
+ size 23293
generation/layer_28/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_28/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f27c590a5839f9cfac9e2498e995513d9307cd3b3e377e7841af8323d17aa7a9
3
+ size 23293
generation/layer_29/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_29/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5661151d51115eee1cac77bf51c4bc27dc000f54f5bdf1e85b3375309c88b122
3
+ size 23293
generation/layer_3/config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "class_name": "LinearProbe",
3
+ "module": "deception.oa_backdoor.detectors.probe_archs",
4
+ "init_args": {
5
+ "d_model": 5120,
6
+ "nhead": 1,
7
+ "normalize_input": "unit_norm"
8
+ }
9
+ }
generation/layer_3/model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b278a55d61ddac08cf31f135ccb4af295dbd8f326cc13074a29ee47bd910e40c
3
+ size 23293