deepseek-v4.1POST
V4.1 Flash 1017 uses sparse attention over top-k tokens per block
A user notes that V4.1 Flash 1017 uses sparse attention attending only to top-k tokens per block instead of full context, cutting compute while keeping coherence.
ADL
@ADLXBTArchived 10 days agoOriginally posted 2026.10.17
ADL @ADLXBTOriginally posted 2026.10.17
@scaling01 deepseek v4.1 flash 1017 uses sparse attention that only attends to top k tokens per block instead of full context cutting compute while keeping coherence
x.com/adlxbt/status/2111382997182369923Archived 2026.10.18 16:44