Article

TRIAGE: VLA inference acceleration requires both reusing and pruning visual tokens

TRIAGE proposes a training-free visual token acceleration method that decides whether to reuse cached KV or prune based on a token drift threshold, and fixes the recomputation budget to directly control prefill latency.

Part of the event: TRIAGE: VLA Inference Reuse and Pruned Visual Tokens

This page provides a summary and source link only.