Update README.md
Browse files
README.md
CHANGED
|
@@ -4,15 +4,83 @@ tags:
|
|
| 4 |
- ner
|
| 5 |
- pii
|
| 6 |
- security
|
|
|
|
|
|
|
|
|
|
| 7 |
language:
|
| 8 |
- en
|
| 9 |
base_model:
|
| 10 |
- microsoft/deberta-v3-large
|
| 11 |
pipeline_tag: token-classification
|
|
|
|
| 12 |
---
|
| 13 |
|
|
|
|
| 14 |
|
| 15 |
-
|
|
|
|
| 16 |
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
- ner
|
| 5 |
- pii
|
| 6 |
- security
|
| 7 |
+
- pii-detection
|
| 8 |
+
- data-privacy
|
| 9 |
+
- compliance
|
| 10 |
language:
|
| 11 |
- en
|
| 12 |
base_model:
|
| 13 |
- microsoft/deberta-v3-large
|
| 14 |
pipeline_tag: token-classification
|
| 15 |
+
license: apache-2.0
|
| 16 |
---
|
| 17 |
|
| 18 |
+
# Dymium PII Named Entity Recognition
|
| 19 |
|
| 20 |
+
Fine-tuned [DeBERTa-v3-large](https://huggingface.co/microsoft/deberta-v3-large)
|
| 21 |
+
for high-accuracy PII detection in enterprise and AI pipeline contexts.
|
| 22 |
|
| 23 |
+
Developed by [Dymium](https://dymium.io) — AI data security platform enabling
|
| 24 |
+
zero-copy data access with built-in governance and compliance.
|
| 25 |
+
|
| 26 |
+
## Entities
|
| 27 |
+
|
| 28 |
+
13 PII entity types:
|
| 29 |
+
|
| 30 |
+
| Entity | Description |
|
| 31 |
+
|--------|-------------|
|
| 32 |
+
| ADDRESS | Physical addresses |
|
| 33 |
+
| AUTH | Credentials, passwords, API keys |
|
| 34 |
+
| DATE | Dates of birth and other sensitive dates |
|
| 35 |
+
| EMAIL | Email addresses |
|
| 36 |
+
| ID_DOC | Passport, driver's license numbers |
|
| 37 |
+
| ID_FIN | Financial identifiers (account, card numbers) |
|
| 38 |
+
| ID_GOV | Government identifiers (SSN, tax IDs) |
|
| 39 |
+
| ID_REF | Internal reference identifiers |
|
| 40 |
+
| ORG | Organization names |
|
| 41 |
+
| PERSON | Personal names |
|
| 42 |
+
| PHONE | Phone numbers |
|
| 43 |
+
| TIME | Timestamps |
|
| 44 |
+
| URL | URLs |
|
| 45 |
+
|
| 46 |
+
## Intended Use
|
| 47 |
+
|
| 48 |
+
- PII detection and redaction in AI pipelines
|
| 49 |
+
- Data governance and compliance enforcement (GDPR, HIPAA, FedRAMP)
|
| 50 |
+
- Sensitive data discovery before feeding to LLMs
|
| 51 |
+
- Real-time data access control
|
| 52 |
+
|
| 53 |
+
## Usage
|
| 54 |
+
```python
|
| 55 |
+
from transformers import pipeline
|
| 56 |
+
|
| 57 |
+
ner = pipeline("token-classification",
|
| 58 |
+
model="dymium/dymium-pii-ner",
|
| 59 |
+
aggregation_strategy="simple")
|
| 60 |
+
|
| 61 |
+
result = ner("Contact John Smith at john@example.com or call 555-123-4567")
|
| 62 |
+
print(result)
|
| 63 |
+
```
|
| 64 |
+
|
| 65 |
+
## Performance
|
| 66 |
+
|
| 67 |
+
| Metric | Score |
|
| 68 |
+
|--------|-------|
|
| 69 |
+
| F1 | [add] |
|
| 70 |
+
| Precision | [add] |
|
| 71 |
+
| Recall | [add] |
|
| 72 |
+
|
| 73 |
+
## Limitations
|
| 74 |
+
|
| 75 |
+
- English language only
|
| 76 |
+
- Performance may vary on highly domain-specific text
|
| 77 |
+
- AUTH entity detection depends on context availability
|
| 78 |
+
|
| 79 |
+
## About Dymium
|
| 80 |
+
|
| 81 |
+
Dymium eliminates data movement risk by enabling AI agents and analytics
|
| 82 |
+
tools to query sensitive data in place — no copying, full governance,
|
| 83 |
+
FedRAMP-ready.
|
| 84 |
+
|
| 85 |
+
🔗 [dymium.io](https://dymium.io) | [Blog](https://dymium.io/blog) |
|
| 86 |
+
[Resources](https://dymium.io/resources)
|