Safetensors
English
qwen3_vl

How to use the Rule Isolation Mask (RI-Mask)

#2
by YoungEver - opened

The paper introduced the Rule Isolation Mask (RI-Mask), which can execute all rule checks in parallel within one packed forward pass. It is very useful in production scenario. It is implemented using PyTorch’s FlexAttention API. So could you please provide a demo for RI-Mask inference? Thanks.

Sign up or log in to comment