Hugging Face's logo

alwaysfurther
/
ai-safety-refusal-classifier

Text Classification
Safetensors
English
ai-safety
refusal-detection
bert
multi-task-learning
Model card Files
xet
 Community
lukehinds commited on
Commit
6dca75c
·
verified ·
1 Parent(s): 6d72e8b

Upload folder using huggingface_hub

Browse files
Files changed (15) hide show
  1. bert/model.safetensors +1 -1
  2. checkpoints/checkpoint-30/model.safetensors +3 -0
  3. checkpoints/checkpoint-30/optimizer.pt +3 -0
  4. checkpoints/checkpoint-30/rng_state.pth +0 -0
  5. checkpoints/checkpoint-30/scheduler.pt +3 -0
  6. checkpoints/checkpoint-30/trainer_state.json +81 -0
  7. checkpoints/checkpoint-30/training_args.bin +3 -0
  8. checkpoints/checkpoint-45/model.safetensors +3 -0
  9. checkpoints/checkpoint-45/optimizer.pt +3 -0
  10. checkpoints/checkpoint-45/rng_state.pth +0 -0
  11. checkpoints/checkpoint-45/scheduler.pt +3 -0
  12. checkpoints/checkpoint-45/trainer_state.json +101 -0
  13. checkpoints/checkpoint-45/training_args.bin +3 -0
  14. classifier_config.json +4 -0
  15. classifier_heads.pt +2 -2
bert/model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:873c2a1de6fe216ea3e43e35b8bdbec1c66b30db1d7fc5345f21661147bedb9e
3
  size 265462608
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7c60c6bd1e6ea9ea781888eace2387a8cac7719a0aaa6a1c8fcbedd8c9dda56
3
  size 265462608
checkpoints/checkpoint-30/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a30a19aff498442b477a6929256441b2dfc18f2290167f989844b50966e414fa
3
+ size 265475800
checkpoints/checkpoint-30/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:747530fb5479ac76140ade4d45cfeda844ce8160f9740477a372a710380b29c1
3
+ size 531012363
checkpoints/checkpoint-30/rng_state.pth ADDED
Binary file (14.5 kB). View file
 
checkpoints/checkpoint-30/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15920a3321e711e86e56dc0a004d00271033663edb180dbb76e7855cd403840e
3
+ size 1465
checkpoints/checkpoint-30/trainer_state.json ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 30,
3
+ "best_metric": 0.9583333333333333,
4
+ "best_model_checkpoint": "models/refusal_classifier_bert/checkpoints/checkpoint-30",
5
+ "epoch": 2.0,
6
+ "eval_steps": 500,
7
+ "global_step": 30,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.6666666666666666,
14
+ "grad_norm": 3.4959802627563477,
15
+ "learning_rate": 3.6000000000000003e-06,
16
+ "loss": 1.2976,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 1.0,
21
+ "eval_avg_accuracy": 0.7083333333333334,
22
+ "eval_explanation_accuracy": 0.5833333333333334,
23
+ "eval_explanation_f1": 0.07407407407407407,
24
+ "eval_loss": 1.1716887950897217,
25
+ "eval_refusal_accuracy": 0.8333333333333334,
26
+ "eval_refusal_f1": 0.875,
27
+ "eval_runtime": 0.1483,
28
+ "eval_samples_per_second": 404.703,
29
+ "eval_steps_per_second": 26.98,
30
+ "step": 15
31
+ },
32
+ {
33
+ "epoch": 1.3333333333333333,
34
+ "grad_norm": 5.753523349761963,
35
+ "learning_rate": 7.600000000000001e-06,
36
+ "loss": 1.1349,
37
+ "step": 20
38
+ },
39
+ {
40
+ "epoch": 2.0,
41
+ "grad_norm": 5.5270490646362305,
42
+ "learning_rate": 1.16e-05,
43
+ "loss": 0.7267,
44
+ "step": 30
45
+ },
46
+ {
47
+ "epoch": 2.0,
48
+ "eval_avg_accuracy": 0.9583333333333333,
49
+ "eval_explanation_accuracy": 0.9833333333333333,
50
+ "eval_explanation_f1": 0.9803921568627451,
51
+ "eval_loss": 0.47130146622657776,
52
+ "eval_refusal_accuracy": 0.9333333333333333,
53
+ "eval_refusal_f1": 0.9459459459459459,
54
+ "eval_runtime": 0.1059,
55
+ "eval_samples_per_second": 566.428,
56
+ "eval_steps_per_second": 37.762,
57
+ "step": 30
58
+ }
59
+ ],
60
+ "logging_steps": 10,
61
+ "max_steps": 45,
62
+ "num_input_tokens_seen": 0,
63
+ "num_train_epochs": 3,
64
+ "save_steps": 500,
65
+ "stateful_callbacks": {
66
+ "TrainerControl": {
67
+ "args": {
68
+ "should_epoch_stop": false,
69
+ "should_evaluate": false,
70
+ "should_log": false,
71
+ "should_save": true,
72
+ "should_training_stop": false
73
+ },
74
+ "attributes": {}
75
+ }
76
+ },
77
+ "total_flos": 0.0,
78
+ "train_batch_size": 16,
79
+ "trial_name": null,
80
+ "trial_params": null
81
+ }
checkpoints/checkpoint-30/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:014075be7b60ec5413be99d1f0c5281c647f820e81fd19badc14a7417de3575a
3
+ size 5841
checkpoints/checkpoint-45/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:121b77670648bf437774333341d81f81ad8463b9f44927a239a0a43e5e2792e2
3
+ size 265475800
checkpoints/checkpoint-45/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a7c967d3a68387f696fe7f283307eb62563419c66eeaa0fda4a3dbc6185da5c0
3
+ size 531012363
checkpoints/checkpoint-45/rng_state.pth ADDED
Binary file (14.5 kB). View file
 
checkpoints/checkpoint-45/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc3b0c8f148ee869beb8b03b3564737df8ec6bf57122343ef646ea6f10946369
3
+ size 1465
checkpoints/checkpoint-45/trainer_state.json ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": 45,
3
+ "best_metric": 0.9833333333333333,
4
+ "best_model_checkpoint": "models/refusal_classifier_bert/checkpoints/checkpoint-45",
5
+ "epoch": 3.0,
6
+ "eval_steps": 500,
7
+ "global_step": 45,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.6666666666666666,
14
+ "grad_norm": 3.4959802627563477,
15
+ "learning_rate": 3.6000000000000003e-06,
16
+ "loss": 1.2976,
17
+ "step": 10
18
+ },
19
+ {
20
+ "epoch": 1.0,
21
+ "eval_avg_accuracy": 0.7083333333333334,
22
+ "eval_explanation_accuracy": 0.5833333333333334,
23
+ "eval_explanation_f1": 0.07407407407407407,
24
+ "eval_loss": 1.1716887950897217,
25
+ "eval_refusal_accuracy": 0.8333333333333334,
26
+ "eval_refusal_f1": 0.875,
27
+ "eval_runtime": 0.1483,
28
+ "eval_samples_per_second": 404.703,
29
+ "eval_steps_per_second": 26.98,
30
+ "step": 15
31
+ },
32
+ {
33
+ "epoch": 1.3333333333333333,
34
+ "grad_norm": 5.753523349761963,
35
+ "learning_rate": 7.600000000000001e-06,
36
+ "loss": 1.1349,
37
+ "step": 20
38
+ },
39
+ {
40
+ "epoch": 2.0,
41
+ "grad_norm": 5.5270490646362305,
42
+ "learning_rate": 1.16e-05,
43
+ "loss": 0.7267,
44
+ "step": 30
45
+ },
46
+ {
47
+ "epoch": 2.0,
48
+ "eval_avg_accuracy": 0.9583333333333333,
49
+ "eval_explanation_accuracy": 0.9833333333333333,
50
+ "eval_explanation_f1": 0.9803921568627451,
51
+ "eval_loss": 0.47130146622657776,
52
+ "eval_refusal_accuracy": 0.9333333333333333,
53
+ "eval_refusal_f1": 0.9459459459459459,
54
+ "eval_runtime": 0.1059,
55
+ "eval_samples_per_second": 566.428,
56
+ "eval_steps_per_second": 37.762,
57
+ "step": 30
58
+ },
59
+ {
60
+ "epoch": 2.6666666666666665,
61
+ "grad_norm": 1.8675131797790527,
62
+ "learning_rate": 1.5600000000000003e-05,
63
+ "loss": 0.2737,
64
+ "step": 40
65
+ },
66
+ {
67
+ "epoch": 3.0,
68
+ "eval_avg_accuracy": 0.9833333333333333,
69
+ "eval_explanation_accuracy": 0.9833333333333333,
70
+ "eval_explanation_f1": 0.9803921568627451,
71
+ "eval_loss": 0.10874374210834503,
72
+ "eval_refusal_accuracy": 0.9833333333333333,
73
+ "eval_refusal_f1": 0.9859154929577465,
74
+ "eval_runtime": 0.116,
75
+ "eval_samples_per_second": 517.433,
76
+ "eval_steps_per_second": 34.496,
77
+ "step": 45
78
+ }
79
+ ],
80
+ "logging_steps": 10,
81
+ "max_steps": 45,
82
+ "num_input_tokens_seen": 0,
83
+ "num_train_epochs": 3,
84
+ "save_steps": 500,
85
+ "stateful_callbacks": {
86
+ "TrainerControl": {
87
+ "args": {
88
+ "should_epoch_stop": false,
89
+ "should_evaluate": false,
90
+ "should_log": false,
91
+ "should_save": true,
92
+ "should_training_stop": true
93
+ },
94
+ "attributes": {}
95
+ }
96
+ },
97
+ "total_flos": 0.0,
98
+ "train_batch_size": 16,
99
+ "trial_name": null,
100
+ "trial_params": null
101
+ }
checkpoints/checkpoint-45/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:014075be7b60ec5413be99d1f0c5281c647f820e81fd19badc14a7417de3575a
3
+ size 5841
classifier_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "model_name": "distilbert-base-uncased",
3
+ "max_length": 128
4
+ }
classifier_heads.pt CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:32e8c6b3b594fadc8b878b1975c3f81c2bef69bae5470d6e1030225975780859
3
- size 14967
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94d5e88f9e544f875e6595a4f2f96b1e2397325fb86043cc8eb0007614712398
3
+ size 14839