Open weights
The models border runs
The nine published models this library reaches: the PII tagger and eight classifiers, one per model-backed detector. Each carries the export gate that let it ship, and the languages it was actually trained in.
9models, all Apache-2.0
26languages, every model evaluated in each
28 msp95 for one encoder pass, 87 tokens on one CPU thread
9 of 9
| Select | Model | Wired to | Languages | Base | Artifact | Size | Export gate |
|---|---|---|---|---|---|---|---|
| piiguardtoken-classificationOpenNERXLM-RoBERTa base · 533 MB | piioutput_leakage | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 533 MB | 120/120 identical, 0 missed | |
| toxicitymulti-label-classificationXLM-RoBERTa base · 535 MB | toxicity | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 535 MB | 0/300 decisions moved | |
| nsfwmulti-label-classificationXLM-RoBERTa base · 535 MB | nsfw | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 535 MB | 0/300 decisions moved | |
| biasmulti-label-classificationXLM-RoBERTa base · 535 MB | bias | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 535 MB | 0/300 decisions moved | |
| gibberishtext-classificationXLM-RoBERTa base · 535 MB | gibberish | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 535 MB | 0/300 decisions moved | |
| politenesstext-classificationXLM-RoBERTa base · 535 MB | politeness | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 535 MB | 0/300 decisions moved | |
| injectiontext-classificationXLM-RoBERTa base · 535 MB | injection | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 535 MB | 1/300 decisions moved | |
| regulated_advicemulti-label-classificationXLM-RoBERTa base · 535 MB | regulated_advice | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 535 MB | 0/300 decisions moved | |
| topic_scopesentence-similarityXLM-RoBERTa base · 533 MB | topic_scope | 26 of 26 | XLM-RoBERTa base | ONNX INT8ONNX | 533 MB | 2/200 nodes moved, cos 0.9993 |