Skip to content

Conversation

@stas00
Copy link
Collaborator

@stas00 stas00 commented Jul 15, 2025

FA3 is needed for 500K+ seqlen on llama-8b. FA2 crashes with illegal memory access error and won't be fixed according to Tri Dao since FA3 is going to replace FA2.

@stas00 stas00 requested review from tjruwase and tohtana as code owners July 15, 2025 17:41
@stas00 stas00 enabled auto-merge (squash) July 15, 2025 17:41
@stas00 stas00 requested a review from loadams July 15, 2025 19:34
@stas00 stas00 merged commit d33b562 into master Jul 16, 2025
9 checks passed
@stas00 stas00 deleted the stas/ulysses-fa3 branch July 16, 2025 15:51
lpnpcs pushed a commit to lpnpcs/DeepSpeed that referenced this pull request Jul 30, 2025
FA3 is needed for 500K+ seqlen on llama-8b.

Signed-off-by: Stas Bekman <[email protected]>
Co-authored-by: Stas Bekman <[email protected]>
LYMDLUT pushed a commit to LYMDLUT/DeepSpeed that referenced this pull request Aug 20, 2025
FA3 is needed for 500K+ seqlen on llama-8b.

Signed-off-by: Stas Bekman <[email protected]>
Co-authored-by: Stas Bekman <[email protected]>
Signed-off-by: lym <[email protected]>
mauryaavinash95 pushed a commit to DataStates/DeepSpeed that referenced this pull request Oct 4, 2025
FA3 is needed for 500K+ seqlen on llama-8b.

Signed-off-by: Stas Bekman <[email protected]>
Co-authored-by: Stas Bekman <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants