A single-layer action head with cross-layer KV fusion and RoPE realignment on a pretrained video DiT matches deeper World Action Models while cutting inference latency to 66.5 ms.
Title resolution pending
1 Pith paper cite this work, alongside 2 external citations. Polarity classification is still indexing.
1
Pith paper citing it
2
external citations · OpenAlex
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Faster-WAM: Do World Action Models Need Deep Action Modules?
A single-layer action head with cross-layer KV fusion and RoPE realignment on a pretrained video DiT matches deeper World Action Models while cutting inference latency to 66.5 ms.