Text Classification
Transformers
ONNX
Safetensors
llama
safety
content-moderation
multi-label
multilingual
tiny-language-model
from-scratch
distilled
text-embeddings-inference
Instructions to use KucLab/Barist.3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use KucLab/Barist.3 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="KucLab/Barist.3")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("KucLab/Barist.3") model = AutoModelForSequenceClassification.from_pretrained("KucLab/Barist.3", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Add report gates.json
Browse files- reports/gates.json +153 -0
reports/gates.json
ADDED
|
@@ -0,0 +1,153 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"tag": "Barist.3",
|
| 3 |
+
"checkpoint": "checkpoints_b3e/best.pt",
|
| 4 |
+
"censer9": "/home/tvadmin/censer.9/checkpoints/best.pt",
|
| 5 |
+
"set": "heldout2 = data/mustcatch_heldout2.jsonl (full 91 rows: 75 unsafe + 16 benign; published copy in data/, working copy in data/processed/; train purged of >0.8 5-gram overlaps, gate in reports/heldout2_dedup.json); heldout3 = data/mustcatch_heldout3.jsonl (fresh phrasings)",
|
| 6 |
+
"gates": [
|
| 7 |
+
{
|
| 8 |
+
"gate": "heldout2_model_only",
|
| 9 |
+
"censer.9": "52/75",
|
| 10 |
+
"Barist.3": "51/75",
|
| 11 |
+
"minimum": "==c9",
|
| 12 |
+
"target": "0.85",
|
| 13 |
+
"pass": false
|
| 14 |
+
},
|
| 15 |
+
{
|
| 16 |
+
"gate": "heldout2_with_prefilter",
|
| 17 |
+
"censer.9": "64/75",
|
| 18 |
+
"Barist.3": "66/75",
|
| 19 |
+
"minimum": "==c9",
|
| 20 |
+
"target": "0.95",
|
| 21 |
+
"pass": true
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"gate": "heldout2_benign_fp",
|
| 25 |
+
"censer.9": "7/16",
|
| 26 |
+
"Barist.3": "5/16",
|
| 27 |
+
"minimum": "2/15",
|
| 28 |
+
"target": "1/15",
|
| 29 |
+
"pass": false
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"gate": "cs_benign_fpr_raw",
|
| 33 |
+
"censer.9": 0.2783,
|
| 34 |
+
"Barist.3": 0.0783,
|
| 35 |
+
"minimum": 0.05,
|
| 36 |
+
"target": 0.03,
|
| 37 |
+
"pass": false
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"gate": "cs_benign_fpr@cs_thr",
|
| 41 |
+
"censer.9": 0.2783,
|
| 42 |
+
"Barist.3": 0.0783,
|
| 43 |
+
"minimum": 0.05,
|
| 44 |
+
"target": 0.03,
|
| 45 |
+
"pass": false
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"gate": "hardneg_fp",
|
| 49 |
+
"censer.9": "6/44",
|
| 50 |
+
"Barist.3": "5/44",
|
| 51 |
+
"minimum": "3/44",
|
| 52 |
+
"target": "1/44",
|
| 53 |
+
"pass": false
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"gate": "train_must_catch",
|
| 57 |
+
"censer.9": "84/88",
|
| 58 |
+
"Barist.3": "81/88",
|
| 59 |
+
"minimum": "84/88",
|
| 60 |
+
"target": "86/88",
|
| 61 |
+
"pass": false
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"gate": "test_f1",
|
| 65 |
+
"censer.9": 0.8239,
|
| 66 |
+
"Barist.3": 0.8228,
|
| 67 |
+
"minimum": 0.83,
|
| 68 |
+
"target": 0.849,
|
| 69 |
+
"pass": false
|
| 70 |
+
},
|
| 71 |
+
{
|
| 72 |
+
"gate": "test_recall",
|
| 73 |
+
"censer.9": 0.7987,
|
| 74 |
+
"Barist.3": 0.7917,
|
| 75 |
+
"minimum": 0.8,
|
| 76 |
+
"target": 0.83,
|
| 77 |
+
"pass": false
|
| 78 |
+
},
|
| 79 |
+
{
|
| 80 |
+
"gate": "test_precision",
|
| 81 |
+
"censer.9": 0.8508,
|
| 82 |
+
"Barist.3": 0.8565,
|
| 83 |
+
"minimum": 0.85,
|
| 84 |
+
"target": 0.86,
|
| 85 |
+
"pass": true
|
| 86 |
+
},
|
| 87 |
+
{
|
| 88 |
+
"gate": "user_benign_fp",
|
| 89 |
+
"censer.9": "35/43",
|
| 90 |
+
"Barist.3": "13/43",
|
| 91 |
+
"minimum": "3/43",
|
| 92 |
+
"target": "1/43",
|
| 93 |
+
"pass": false
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"gate": "ece_calibrated",
|
| 97 |
+
"censer.9": 0.018413447943885906,
|
| 98 |
+
"Barist.3": 0.0250844468628166,
|
| 99 |
+
"minimum": 0.035,
|
| 100 |
+
"target": 0.03,
|
| 101 |
+
"pass": true
|
| 102 |
+
},
|
| 103 |
+
{
|
| 104 |
+
"gate": "onnx_max_abs_diff",
|
| 105 |
+
"censer.9": null,
|
| 106 |
+
"Barist.3": 0.0,
|
| 107 |
+
"minimum": 0.001,
|
| 108 |
+
"target": 0.0001,
|
| 109 |
+
"pass": true
|
| 110 |
+
},
|
| 111 |
+
{
|
| 112 |
+
"gate": "regex_gate_hard_fp",
|
| 113 |
+
"censer.9": null,
|
| 114 |
+
"Barist.3": 0,
|
| 115 |
+
"minimum": 0,
|
| 116 |
+
"target": 0,
|
| 117 |
+
"pass": true
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"gate": "heldout2_dedup_max_overlap",
|
| 121 |
+
"censer.9": null,
|
| 122 |
+
"Barist.3": 0.75,
|
| 123 |
+
"minimum": 0.8,
|
| 124 |
+
"target": 0.5,
|
| 125 |
+
"pass": true
|
| 126 |
+
},
|
| 127 |
+
{
|
| 128 |
+
"gate": "heldout3_model_only",
|
| 129 |
+
"censer.9": "32/50",
|
| 130 |
+
"Barist.3": "29/50",
|
| 131 |
+
"minimum": "==c9",
|
| 132 |
+
"target": "0.85",
|
| 133 |
+
"pass": false
|
| 134 |
+
},
|
| 135 |
+
{
|
| 136 |
+
"gate": "heldout3_with_prefilter",
|
| 137 |
+
"censer.9": "33/50",
|
| 138 |
+
"Barist.3": "31/50",
|
| 139 |
+
"minimum": "==c9",
|
| 140 |
+
"target": "0.95",
|
| 141 |
+
"pass": false
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"gate": "identity_fp",
|
| 145 |
+
"censer.9": 0.6364,
|
| 146 |
+
"Barist.3": 0.3182,
|
| 147 |
+
"minimum": 0.02,
|
| 148 |
+
"target": 0.0,
|
| 149 |
+
"pass": false
|
| 150 |
+
}
|
| 151 |
+
],
|
| 152 |
+
"all_pass": false
|
| 153 |
+
}
|