Skip to content

Enable FusedSDPA slicing if chunked-prefill is enabled or max_model_len > 32k #2272

Merged
czhu15 merged 1 commit into
HabanaAI:aice/v1.22.0from
yangulei:slice_long_seq
Jun 5, 2026
Merged

Enable FusedSDPA slicing if chunked-prefill is enabled or max_model_len > 32k #2272
czhu15 merged 1 commit into
HabanaAI:aice/v1.22.0from
yangulei:slice_long_seq

Commits

Commits on Jun 5, 2026