Token Classification
Transformers
Safetensors
English
distilbert
secret-detection
security
credentials
dlp
code
Instructions to use AndrewAndrewsen/distilbert-secret-masker-v3.3a-rs with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use AndrewAndrewsen/distilbert-secret-masker-v3.3a-rs with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("token-classification", model="AndrewAndrewsen/distilbert-secret-masker-v3.3a-rs")# Load model directly from transformers import AutoTokenizer, AutoModelForTokenClassification tokenizer = AutoTokenizer.from_pretrained("AndrewAndrewsen/distilbert-secret-masker-v3.3a-rs") model = AutoModelForTokenClassification.from_pretrained("AndrewAndrewsen/distilbert-secret-masker-v3.3a-rs", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -62,12 +62,12 @@ gating; chat/log/prose is weaker than source code (see *Limitations*).
|
|
| 62 |
|
| 63 |
## Evaluation
|
| 64 |
|
| 65 |
-
All evaluation is on **frozen
|
| 66 |
metric is **strict exact span + line** (exact character span AND correct line);
|
| 67 |
file-level and line-level are also reported because rule scanners are line/file
|
| 68 |
detectors, not span labelers.
|
| 69 |
|
| 70 |
-
### Strict real-code benchmarks (v3.3a-RS @ τ=0.99)
|
| 71 |
|
| 72 |
| set | files (pos/total) | strict span+line P/R/F1 | line P/R/F1 | file P/R/F1 |
|
| 73 |
|---|---|---|---|---|
|
|
|
|
| 62 |
|
| 63 |
## Evaluation
|
| 64 |
|
| 65 |
+
All evaluation is on **frozen real code held out from training** (repo-disjoint). Benchmarks **B/C/D are observed regression sets** — repo-disjoint from training, but watched repeatedly during development, so they are **not blind**. **RealCode-1** (below) is the purpose-built, never-observed adjudicator. Headline
|
| 66 |
metric is **strict exact span + line** (exact character span AND correct line);
|
| 67 |
file-level and line-level are also reported because rule scanners are line/file
|
| 68 |
detectors, not span labelers.
|
| 69 |
|
| 70 |
+
### Strict real-code regression benchmarks — B/C/D, observed (v3.3a-RS @ τ=0.99)
|
| 71 |
|
| 72 |
| set | files (pos/total) | strict span+line P/R/F1 | line P/R/F1 | file P/R/F1 |
|
| 73 |
|---|---|---|---|---|
|