Token Classification
Transformers
Safetensors
halo_stream_guard
guard
safety
moderation
output-safety
response-moderation
streaming
multilingual
custom_code
Instructions to use astroware/HaloGuard2-MultiStream-0.8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use astroware/HaloGuard2-MultiStream-0.8B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("token-classification", model="astroware/HaloGuard2-MultiStream-0.8B", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForTokenClassification model = AutoModelForTokenClassification.from_pretrained("astroware/HaloGuard2-MultiStream-0.8B", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
Download serving_rule.json from astroware/HaloGuard2-MultiStream-0.8B: direct link, hf CLI and curl.
- Browser
- Download file 750 Bytes
-
https://huggingface.co/astroware/HaloGuard2-MultiStream-0.8B/resolve/main/serving_rule.json
- Command line
-
hf download hf://astroware/HaloGuard2-MultiStream-0.8B/serving_rule.json
-
curl -L -o serving_rule.json https://huggingface.co/astroware/HaloGuard2-MultiStream-0.8B/resolve/main/serving_rule.json
750 Bytes
| { | |
| "rule_id": "poly-v2.0-2026-10-29", | |
| "model": "HaloGuard2.0-Poly-0.8B", | |
| "rule": { | |
| "max_p_threshold": null, | |
| "run_threshold": 0.675, | |
| "run_length": 3, | |
| "latched": true, | |
| "aggregation": "max over RESPONSE positions only", | |
| "note": "Single run arm: fire when 3 consecutive response tokens are all >= 0.675. run_length stays <= 3 by design." | |
| }, | |
| "scores": { | |
| "polyguard_prompts_response_f1": 83.08, | |
| "protocol": "PolyGuardPrompts, response_harm_label, 29,053 scorable rows, 17 languages, macro over languages", | |
| "macro_fpr": 6.1, | |
| "macro_fnr": 7.5, | |
| "by_group_f1": {"English": 87.1, "CJK": 83.1, "Arabic-script": 82.7, | |
| "European/Semitic": 85.5, "Southeast Asian": 73.9, "Indic": 64.2} | |
| } | |
| } | |