PR #16309 ggml : remove KQ mask padding - SemanticDiff

Commits

metal : pad K, V and Mask when needed

ggerganov committed 38 days ago
cont : simplify

ggerganov committed 38 days ago
cuda : add TODO about KV padding requirement

ggerganov committed 38 days ago
metal : add comments

ggerganov committed 38 days ago
metal : remove mask padding requirement

ggerganov committed 38 days ago
ggml : remove KQ mask padding

ggerganov committed 38 days ago

Loading