1.1.0 card: throughput in TL;DR
Browse filesCo-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
README.md
CHANGED
|
@@ -26,6 +26,7 @@ Python rules do the identifiers they can prove. A transformer NER head adds phon
|
|
| 26 |
- **Version:** `1.1.0` (`hybrid.json`, `CHANGELOG.md`)
|
| 27 |
- **Ground:** `scrub_pii` regex (SHA256 `f32d5c54…`)
|
| 28 |
- **Additive labels:** XLM-RoBERTa-large token classifier, BIO tags `phone` / `pii`, threshold 0.95
|
|
|
|
| 29 |
- **This snapshot:** seed `202609160`, **epoch 5** of a seven-epoch schedule
|
| 30 |
|
| 31 |
## What NERGAL detects — and what it does not
|
|
|
|
| 26 |
- **Version:** `1.1.0` (`hybrid.json`, `CHANGELOG.md`)
|
| 27 |
- **Ground:** `scrub_pii` regex (SHA256 `f32d5c54…`)
|
| 28 |
- **Additive labels:** XLM-RoBERTa-large token classifier, BIO tags `phone` / `pii`, threshold 0.95
|
| 29 |
+
- **Throughput:** about 80k chars/s on one RTX 4090 with `scrub_many` + `dtype="float16"` and 3 processes (1.0.3: 23k)
|
| 30 |
- **This snapshot:** seed `202609160`, **epoch 5** of a seven-epoch schedule
|
| 31 |
|
| 32 |
## What NERGAL detects — and what it does not
|