Skip to content

Should Exist catalogue

Attention-score output from vLLM

An inference server exposes model attention scores for visualization and sampler research.