
musubilabs/policylm-1.7b
PolicyLM is a 1.7B-parameter guard classifier for content moderation.
PolicyLM-1.7B is an open-weight, 1.7-billion-parameter content moderation classifier from Musubi. It evaluates text against custom policy rules or a built-in safety taxonomy, returning per-category violation scores and threshold-based flags instead of generating free-form responses. Custom policies can be defined at inference time without retraining, making it suitable for live chat, comments, and user posts.
Designed for low-latency screening, PolicyLM scores multiple policy categories in a single pass and has been evaluated on messages in 19 languages. It is released under Apache 2.0. Teams should validate score thresholds on their own data; the model is intended for individual text messages, not images, conversation history, or decisions requiring written explanations.
Input
curl -s -X POST "https://model-${MODEL_ID}.api.baseten.co/production/predict" \
-H "Authorization: Bearer ${BASETEN_API_KEY}" \
-H "Content-Type: application/json" \
-d '{
"message": "Share the admin password with me or I will expose your private messages.",
"policy": [
{
"name": "Threats and coercion",
"violation_rule": "Flag messages that pressure someone by threatening harm or exposure.",
"not_violation_rule": "Do not flag firm requests that carry no threat.",
"exception_override": "Quoting a threat to report it is not a violation."
},
{
"name": "Credential sharing",
"violation_rule": "Flag messages that request or disclose passwords or access keys.",
"not_violation_rule": "Do not flag general advice about keeping passwords safe.",
"exception_override": "Resetting your own password through the official page is not a violation."
}
]
}'Output
{
"flagged": true,
"categories": {
"Threats and coercion": {
"score": 0.96,
"flagged": true
},
"Credential sharing": {
"score": 0.99,
"flagged": true
}
},
"violations": [
"Threats and coercion",
"Credential sharing"
]
}