Correct the document-type head's figures to the published head
Browse files
README.md
CHANGED
|
@@ -212,9 +212,9 @@ is: a contract, a proposal, source code and so on. This revision (`main`) is unc
|
|
| 212 |
- **The PII output is byte-identical** to this revision's, on every row of the PII release check (int8 and fp32,
|
| 213 |
onnxruntime 1.19.2 and 1.24.2).
|
| 214 |
- **What the head decides.** `tech.source_code` is the one type the head decides on its own. On a held-out set in
|
| 215 |
-
seven languages it scores F1 0.
|
| 216 |
flags no benign item in any language, not counting 55 benign items that themselves contain source code; counting
|
| 217 |
-
them,
|
| 218 |
the product.
|
| 219 |
- **Languages.** Trained and evaluated in English, German, French, Italian, Spanish, Russian and Arabic.
|
| 220 |
- **Try it in the browser:** [vadimbelsky/scnsoft-pii-encoder-demo](https://huggingface.co/spaces/vadimbelsky/scnsoft-pii-encoder-demo).
|
|
|
|
| 212 |
- **The PII output is byte-identical** to this revision's, on every row of the PII release check (int8 and fp32,
|
| 213 |
onnxruntime 1.19.2 and 1.24.2).
|
| 214 |
- **What the head decides.** `tech.source_code` is the one type the head decides on its own. On a held-out set in
|
| 215 |
+
seven languages it scores F1 0.848 (seed 1234) and 0.825 (seed 4321), against 0.614 for a rule-based classifier. It
|
| 216 |
flags no benign item in any language, not counting 55 benign items that themselves contain source code; counting
|
| 217 |
+
them, seed 4321 flags 1 of 18,704. Eight more types carry advisory scores, which a rule-based classifier decides in
|
| 218 |
the product.
|
| 219 |
- **Languages.** Trained and evaluated in English, German, French, Italian, Spanish, Russian and Arabic.
|
| 220 |
- **Try it in the browser:** [vadimbelsky/scnsoft-pii-encoder-demo](https://huggingface.co/spaces/vadimbelsky/scnsoft-pii-encoder-demo).
|