Upload 12 files
Browse files- .gitattributes +2 -0
- deberta_best/config.json +38 -0
- deberta_best/model.safetensors +3 -0
- deberta_best/tokenizer.json +0 -0
- deberta_best/tokenizer_config.json +24 -0
- ensemble_config.pkl +3 -0
- ensemble_evaluation.png +3 -0
- model_card.json +67 -0
- scaler.pkl +3 -0
- xgb_model.json +0 -0
- xgb_threshold.pkl +3 -0
- zeroday.faiss +3 -0
- zeroday_meta.pkl +3 -0
.gitattributes
CHANGED
|
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
| 33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
| 34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
| 35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
| 36 |
+
ensemble_evaluation.png filter=lfs diff=lfs merge=lfs -text
|
| 37 |
+
zeroday.faiss filter=lfs diff=lfs merge=lfs -text
|
deberta_best/config.json
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"DebertaV2ForSequenceClassification"
|
| 4 |
+
],
|
| 5 |
+
"attention_probs_dropout_prob": 0.1,
|
| 6 |
+
"bos_token_id": null,
|
| 7 |
+
"dtype": "float32",
|
| 8 |
+
"eos_token_id": null,
|
| 9 |
+
"hidden_act": "gelu",
|
| 10 |
+
"hidden_dropout_prob": 0.1,
|
| 11 |
+
"hidden_size": 768,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"intermediate_size": 3072,
|
| 14 |
+
"layer_norm_eps": 1e-07,
|
| 15 |
+
"legacy": true,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"max_relative_positions": -1,
|
| 18 |
+
"model_type": "deberta-v2",
|
| 19 |
+
"norm_rel_ebd": "layer_norm",
|
| 20 |
+
"num_attention_heads": 12,
|
| 21 |
+
"num_hidden_layers": 12,
|
| 22 |
+
"pad_token_id": 0,
|
| 23 |
+
"pooler_dropout": 0.0,
|
| 24 |
+
"pooler_hidden_act": "gelu",
|
| 25 |
+
"pooler_hidden_size": 768,
|
| 26 |
+
"pos_att_type": [
|
| 27 |
+
"p2c",
|
| 28 |
+
"c2p"
|
| 29 |
+
],
|
| 30 |
+
"position_biased_input": false,
|
| 31 |
+
"position_buckets": 256,
|
| 32 |
+
"relative_attention": true,
|
| 33 |
+
"share_att_key": true,
|
| 34 |
+
"tie_word_embeddings": true,
|
| 35 |
+
"transformers_version": "5.10.2",
|
| 36 |
+
"type_vocab_size": 0,
|
| 37 |
+
"vocab_size": 128100
|
| 38 |
+
}
|
deberta_best/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:06acf0f692b96bd60d38237e69438610ee5729c3f4f9b4bae96c18e76b470bf5
|
| 3 |
+
size 737719272
|
deberta_best/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
deberta_best/tokenizer_config.json
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": true,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "[CLS]",
|
| 5 |
+
"cls_token": "[CLS]",
|
| 6 |
+
"do_lower_case": false,
|
| 7 |
+
"eos_token": "[SEP]",
|
| 8 |
+
"extra_special_tokens": [
|
| 9 |
+
"[PAD]",
|
| 10 |
+
"[CLS]",
|
| 11 |
+
"[SEP]"
|
| 12 |
+
],
|
| 13 |
+
"is_local": false,
|
| 14 |
+
"local_files_only": false,
|
| 15 |
+
"mask_token": "[MASK]",
|
| 16 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 17 |
+
"pad_token": "[PAD]",
|
| 18 |
+
"sep_token": "[SEP]",
|
| 19 |
+
"split_by_punct": false,
|
| 20 |
+
"tokenizer_class": "DebertaV2Tokenizer",
|
| 21 |
+
"unk_id": 3,
|
| 22 |
+
"unk_token": "[UNK]",
|
| 23 |
+
"vocab_type": "spm"
|
| 24 |
+
}
|
ensemble_config.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f145e010f99390714dd529b1335f5b2b11128fe64d66058d850f7e3246276f7e
|
| 3 |
+
size 298
|
ensemble_evaluation.png
ADDED
|
Git LFS Details
|
model_card.json
ADDED
|
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"version": "2.0",
|
| 3 |
+
"created_at": "2026-06-21T11:07:28.861636",
|
| 4 |
+
"architecture": "XGBoost + DeBERTa-v3-base ensemble",
|
| 5 |
+
"embedding": "sentence-transformers/all-MiniLM-L6-v2",
|
| 6 |
+
"transformer": "microsoft/deberta-v3-base",
|
| 7 |
+
"feature_dim": 396,
|
| 8 |
+
"ensemble": {
|
| 9 |
+
"w_xgb": 0.05,
|
| 10 |
+
"w_transformer": 0.95,
|
| 11 |
+
"w_faiss-delta": 0.1,
|
| 12 |
+
"threshold": 0.915
|
| 13 |
+
},
|
| 14 |
+
"xgb": {
|
| 15 |
+
"n_estimators": 300,
|
| 16 |
+
"max_depth": 5,
|
| 17 |
+
"threshold": 0.5499999999999998
|
| 18 |
+
},
|
| 19 |
+
"training_data": {
|
| 20 |
+
"adversarial": 12000,
|
| 21 |
+
"benign": 12000,
|
| 22 |
+
"families": [
|
| 23 |
+
"JAILBREAK_PROMPTS",
|
| 24 |
+
"ADVERSARIAL_ATTACKS",
|
| 25 |
+
"TECHNICAL_ATTACKS",
|
| 26 |
+
"ENHANCED_JAILBREAKS",
|
| 27 |
+
"CRITICAL_SCENARIOS",
|
| 28 |
+
"ADVANCED_TECHNIQUES",
|
| 29 |
+
"EXTREME_SCENARIOS",
|
| 30 |
+
"JAILBREAK_PROMPTS_Aditya",
|
| 31 |
+
"ARMAAN_KEY_ATTACK",
|
| 32 |
+
"MULTILINGUAL_PROMPTS",
|
| 33 |
+
"ARMAAN_MULTILINGUAL_PROMPTS",
|
| 34 |
+
"multi_lang",
|
| 35 |
+
"LANGUAGE_BASED_PROMPTS_JIYA",
|
| 36 |
+
"SKELETON_KEY_ATTACKS",
|
| 37 |
+
"COMBINATION_ATTACKS",
|
| 38 |
+
"REFUSAL_SUPPRESSION_ATTACKS",
|
| 39 |
+
"AFFIRMATIVE_FORCING_ATTACKS",
|
| 40 |
+
"ROLEPLAY_PERSONA_ATTACKS",
|
| 41 |
+
"MANY_SHOT_ATTACKS",
|
| 42 |
+
"BEST_OF_N_SEMANTIC_ATTACKS",
|
| 43 |
+
"JAILBREAK_ATTACKS",
|
| 44 |
+
"PROMPT_INJECTION_ATTACKS",
|
| 45 |
+
"LOGIC_TRAP_ATTACKS",
|
| 46 |
+
"ADVERSARIAL_CREATIVE_ATTACK",
|
| 47 |
+
"CHAIN_OF_THOUGHT_ATTACKS",
|
| 48 |
+
"SOCIAL_ENGINEERING_SIMULATION",
|
| 49 |
+
"AMBIGUOUS_LANGUAGE_SIMULATION",
|
| 50 |
+
"MULTI_TURN_PERSISTENCE_SIMULATION",
|
| 51 |
+
"ADVANCED_HIGH_FIDELITY_SIMULATION",
|
| 52 |
+
"PERSUASIVE_ARGUMENT_SIMULATION",
|
| 53 |
+
"PHILOSOPHICAL_TRAP_ATTACKS",
|
| 54 |
+
"CONSISTENCY_EXPLOIT_ATTACKS",
|
| 55 |
+
"FALSE_PREMISE_ATTACKS",
|
| 56 |
+
"REDUCTIO_ATTACKS",
|
| 57 |
+
"META_REASONING_ATTACKS",
|
| 58 |
+
"OBFUSCATION_ATTACKS",
|
| 59 |
+
"PERSONA_ATTACKS"
|
| 60 |
+
]
|
| 61 |
+
},
|
| 62 |
+
"test_metrics": {
|
| 63 |
+
"f1": 0.9808,
|
| 64 |
+
"fpr": 0.0096,
|
| 65 |
+
"tpr": 0.9716
|
| 66 |
+
}
|
| 67 |
+
}
|
scaler.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a0523e04a2d0e46de7b9f431b50a8fbd49c069e281853f27ea69363d88307339
|
| 3 |
+
size 10119
|
xgb_model.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
xgb_threshold.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:98a362bc6e7601f3b4b7d041f278e9780d43cbd16ddb0fc625e5ec59871e0ef0
|
| 3 |
+
size 117
|
zeroday.faiss
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fb484fa9612faadf425d92aa72fabf5657d98b8ecac6b7375a9111458c25d575
|
| 3 |
+
size 12911661
|
zeroday_meta.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e65fccfc9f92461e998b72c2387f710b4d7dd7e4b0931c50f173b1ace7902cae
|
| 3 |
+
size 5050003
|