mstar.engine.resources.attn.ragged#
Modules
The cacheless attention resource: the spec-time factory, and the plan machinery its backends share. |
|
Block-causal cacheless attention through FlashInfer's paged prefill. |
|
What a model declares about cacheless (ragged) attention. |
|
Cacheless varlen attention through FlashInfer's ragged prefill wrapper. |
|
Cacheless (ragged) prefill wrappers: queries and keys packed by segment into |