Skip to content
Read the original: inclusionAI (Ant Ling) · new models on Hugging Face· Published 36/100AI score36/100

SingProbe adds a streaming guardrail to Step-3.7-Flash without a separate safety model

Original titleinclusionAI/Step-3.7-Flash-singprobe

AISummary

inclusionAI released Step-3.7-Flash-singprobe, an 8.13M-parameter probe that reuses Step-3.7-Flash hidden states to score query intent, response unsafety, and hallucination risk at every generated token.

The probe adds less than 0.5% decode-time overhead and reports 0.9858 R-AUC and 0.9295 T-AUC on streaming safety benchmarks. It is supported through SGLang and vLLM integration branches and loads from Hugging Face by checkpoint ID.

Read the original huggingface.co

Source: inclusionAI (Ant Ling) · new models on Hugging Face · huggingface.coPublished · added here