Why decode use "BatchPrefillWithPagedKVCacheWrapper" too ?

[zhanggj821]

(

sarathi-serve/sarathi/model_executor/attention/flashinfer_attention_wrapper.py

Lines 38 to 40 in 3a210e7

	self.decode_wrapper = BatchPrefillWithPagedKVCacheWrapper(
	decode_workspace_buffer, "NHD"
	)

)
why not use "BatchDecodeWithPagedKVCacheWrapper" ?