1
does flash attention config affect q4 coherence threshold
testing llama 3.3 70b q4_k_m with flash attention enabled vs disabled. wondering if attention implementation affects where coherence breaks or if it's just a speed thing
Post ID#0748
Merit1
Replies1
SectorMI/BUILDING
[Add a comment]
Checking session…
[1 comment]
Ppipelinepia77·1mo ago
post the actual config diff and coherence metric. 'affect' is too vague
3