updates
Browse files
README.md
CHANGED
|
@@ -10,13 +10,16 @@ tags:
|
|
| 10 |
- qwen3
|
| 11 |
- vllm
|
| 12 |
pipeline_tag: token-classification
|
| 13 |
-
base_model: perplexity-ai/
|
| 14 |
---
|
| 15 |
|
| 16 |
-
#
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
vLLM packaging of
|
| 19 |
-
[`perplexity-ai/
|
| 20 |
a PII masking model for conversational data. The model uses a ~600M-parameter
|
| 21 |
bidirectional Qwen3 encoder with a
|
| 22 |
[`perplexity-ai/pplx-embed-v1-0.6b`](https://huggingface.co/perplexity-ai/pplx-embed-v1-0.6b)
|
|
@@ -41,8 +44,8 @@ from the same shell:
|
|
| 41 |
|
| 42 |
```bash
|
| 43 |
pip install huggingface_hub numpy tokenizers requests
|
| 44 |
-
hf download perplexity-ai/
|
| 45 |
-
cd
|
| 46 |
docker compose -f serving/docker-compose.yml up -d
|
| 47 |
|
| 48 |
curl -s http://127.0.0.1:8002/health
|
|
@@ -60,7 +63,7 @@ The example client expects nonempty text.
|
|
| 60 |
To run vLLM directly instead of using Docker Compose:
|
| 61 |
|
| 62 |
```bash
|
| 63 |
-
vllm serve perplexity-ai/
|
| 64 |
--served-model-name pii-masker --host 127.0.0.1 --port 8003 \
|
| 65 |
--runner pooling --convert classify \
|
| 66 |
--pooler-config '{"use_activation": false}' \
|
|
@@ -131,3 +134,17 @@ of the selected token labels; this differs from the base model's span score.
|
|
| 131 |
## License
|
| 132 |
|
| 133 |
MIT, matching the base model. See [LICENSE](LICENSE).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
- qwen3
|
| 11 |
- vllm
|
| 12 |
pipeline_tag: token-classification
|
| 13 |
+
base_model: perplexity-ai/PII-Tracer
|
| 14 |
---
|
| 15 |
|
| 16 |
+
# PII-Tracer-vLLM
|
| 17 |
+
|
| 18 |
+
This repository provides PII-Tracer, the detector introduced in
|
| 19 |
+
[PII-TRACE](https://arxiv.org/abs/2609.22200).
|
| 20 |
|
| 21 |
vLLM packaging of
|
| 22 |
+
[`perplexity-ai/PII-Tracer`](https://huggingface.co/perplexity-ai/PII-Tracer),
|
| 23 |
a PII masking model for conversational data. The model uses a ~600M-parameter
|
| 24 |
bidirectional Qwen3 encoder with a
|
| 25 |
[`perplexity-ai/pplx-embed-v1-0.6b`](https://huggingface.co/perplexity-ai/pplx-embed-v1-0.6b)
|
|
|
|
| 44 |
|
| 45 |
```bash
|
| 46 |
pip install huggingface_hub numpy tokenizers requests
|
| 47 |
+
hf download perplexity-ai/PII-Tracer-vLLM --local-dir PII-Tracer-vLLM
|
| 48 |
+
cd PII-Tracer-vLLM
|
| 49 |
docker compose -f serving/docker-compose.yml up -d
|
| 50 |
|
| 51 |
curl -s http://127.0.0.1:8002/health
|
|
|
|
| 63 |
To run vLLM directly instead of using Docker Compose:
|
| 64 |
|
| 65 |
```bash
|
| 66 |
+
vllm serve perplexity-ai/PII-Tracer-vLLM \
|
| 67 |
--served-model-name pii-masker --host 127.0.0.1 --port 8003 \
|
| 68 |
--runner pooling --convert classify \
|
| 69 |
--pooler-config '{"use_activation": false}' \
|
|
|
|
| 134 |
## License
|
| 135 |
|
| 136 |
MIT, matching the base model. See [LICENSE](LICENSE).
|
| 137 |
+
|
| 138 |
+
## Citation
|
| 139 |
+
|
| 140 |
+
If you use or reference this work, please cite:
|
| 141 |
+
|
| 142 |
+
```bibtex
|
| 143 |
+
@article{zhang2026piitrace,
|
| 144 |
+
title = {{PII-TRACE}: A Benchmark for Context-Aware {PII} Detection in Multi-Turn {LLM} Conversations},
|
| 145 |
+
author = {Zhang, Kaiyuan and Wang, Chuan and Zhong, Joey and Fryzel, Paul and Polley, Kyle and Ma, Jerry and Li, Ninghui},
|
| 146 |
+
journal = {arXiv preprint arXiv:2609.22200},
|
| 147 |
+
year = {2026},
|
| 148 |
+
url = {https://arxiv.org/abs/2609.22200}
|
| 149 |
+
}
|
| 150 |
+
```
|