Text Classification
Transformers
ONNX
Safetensors
modernbert
decision-steering
prompt-injection
indirect-prompt-injection
ai-agent-security
guardrails
llm-security
ai-safety
decision-model
jev-like
model-agnostic
erase-and-check
claim-detection
security
multilingual
mmbert
cpu
patronus
decision-steering-gate
decision-model-guard
claim-classification
decision-security
ai-security
agent-security
prompt-injection-defense
indirect-prompt-injection-defense
llm-guardrails
tool-output-security
model-agnostic-defense
binary-classification
sequence-classification
local-inference
on-device
cpu-inference
onnxruntime
quantized
int8
int4
pruned
mit
Eval Results (legacy)
text-embeddings-inference
Instructions to use patronus-studio/decision-steering-gate with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use patronus-studio/decision-steering-gate with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="patronus-studio/decision-steering-gate")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("patronus-studio/decision-steering-gate") model = AutoModelForSequenceClassification.from_pretrained("patronus-studio/decision-steering-gate", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Model card: link the public benchmark dataset
Browse files
README.md
CHANGED
|
@@ -19,6 +19,8 @@ language:
|
|
| 19 |
base_model:
|
| 20 |
- jhu-clsp/mmBERT-small
|
| 21 |
base_model_relation: finetune
|
|
|
|
|
|
|
| 22 |
library_name: transformers
|
| 23 |
pipeline_tag: text-classification
|
| 24 |
metrics:
|
|
@@ -292,6 +294,8 @@ The detector is evaluated separately from the full gate. Its score measures whet
|
|
| 292 |
| **IFlippedJev**: 1,000 claims that flipped at least one of three local decision models on a real case | 46.8 % found | **96.6 %** found |
|
| 293 |
| **NotClaim**: 1,343 claims and 1,343 benign texts that use the same claim vocabulary | 77.7 % found, 11.0 % of benign flagged, AUROC 0.928 | **98.1 %** found, **2.2 %** of benign flagged, AUROC **0.998** |
|
| 294 |
|
|
|
|
|
|
|
| 295 |
Threshold `0.3`, PyTorch FP32 checkpoints, inputs up to 128 tokens. The INT8 + INT4 graph gives 96.4 %, 96.3 %, 98.4 % and 2.4 % on the same sets.
|
| 296 |
|
| 297 |
<details>
|
|
@@ -535,10 +539,10 @@ Both detectors split these texts into the same number of tokens (34.7 and 35.1 o
|
|
| 535 |
| :--- | :--- | :--- |
|
| 536 |
| Claim-detector training mix | Training | 215,315 text units in 15 languages: 158k units of the v1 corpus (after near-duplicate removal), 41k red-team units (claims tested against local decision models, translated into 14 languages, plus hard negatives), 11k units from two further claim collections and 5.6k contrast negatives |
|
| 537 |
| Detector validation split | Layer selection, checkpoint comparison, ONNX validation | 13,155 units: 5,702 claims and 7,453 content units; same distribution as training |
|
| 538 |
-
| Red-team holdout, IFlippedJev, NotClaim | Held-out evaluation | see [claim detector](#claim-detector) |
|
| 539 |
| Decision-steering benchmark | Held-out evaluation of the full gate | Five tasks, 100 cases per task, three attacks and three neutral additions per case |
|
| 540 |
|
| 541 |
-
This repository publishes [aggregated evaluation results](https://huggingface.co/patronus-studio/decision-steering-gate/blob/main/eval_results.json)
|
| 542 |
|
| 543 |
### Training recipe
|
| 544 |
|
|
|
|
| 19 |
base_model:
|
| 20 |
- jhu-clsp/mmBERT-small
|
| 21 |
base_model_relation: finetune
|
| 22 |
+
datasets:
|
| 23 |
+
- patronus-studio/decision-steering-benchmarks
|
| 24 |
library_name: transformers
|
| 25 |
pipeline_tag: text-classification
|
| 26 |
metrics:
|
|
|
|
| 294 |
| **IFlippedJev**: 1,000 claims that flipped at least one of three local decision models on a real case | 46.8 % found | **96.6 %** found |
|
| 295 |
| **NotClaim**: 1,343 claims and 1,343 benign texts that use the same claim vocabulary | 77.7 % found, 11.0 % of benign flagged, AUROC 0.928 | **98.1 %** found, **2.2 %** of benign flagged, AUROC **0.998** |
|
| 296 |
|
| 297 |
+
The three sets are public: [patronus-studio/decision-steering-benchmarks](https://huggingface.co/datasets/patronus-studio/decision-steering-benchmarks), with the cases of IFlippedJev to replay against your own decision model.
|
| 298 |
+
|
| 299 |
Threshold `0.3`, PyTorch FP32 checkpoints, inputs up to 128 tokens. The INT8 + INT4 graph gives 96.4 %, 96.3 %, 98.4 % and 2.4 % on the same sets.
|
| 300 |
|
| 301 |
<details>
|
|
|
|
| 539 |
| :--- | :--- | :--- |
|
| 540 |
| Claim-detector training mix | Training | 215,315 text units in 15 languages: 158k units of the v1 corpus (after near-duplicate removal), 41k red-team units (claims tested against local decision models, translated into 14 languages, plus hard negatives), 11k units from two further claim collections and 5.6k contrast negatives |
|
| 541 |
| Detector validation split | Layer selection, checkpoint comparison, ONNX validation | 13,155 units: 5,702 claims and 7,453 content units; same distribution as training |
|
| 542 |
+
| Red-team holdout, IFlippedJev, NotClaim | Held-out evaluation | see [claim detector](#claim-detector); [public dataset](https://huggingface.co/datasets/patronus-studio/decision-steering-benchmarks) |
|
| 543 |
| Decision-steering benchmark | Held-out evaluation of the full gate | Five tasks, 100 cases per task, three attacks and three neutral additions per case |
|
| 544 |
|
| 545 |
+
This repository publishes [aggregated evaluation results](https://huggingface.co/patronus-studio/decision-steering-gate/blob/main/eval_results.json). The held-out benchmarks are a public dataset: [patronus-studio/decision-steering-benchmarks](https://huggingface.co/datasets/patronus-studio/decision-steering-benchmarks). The training data is not published.
|
| 546 |
|
| 547 |
### Training recipe
|
| 548 |
|