Repository navigation

ggml : support broadcast for ggml_soft_max_ext and ggml_flash_attn_ext - #14435

Merged

ggerganov merged 4 commits into

gg/ggml-batch-soft-max-ops

Jul 2, 2025

Merged

ggml : support broadcast for ggml_soft_max_ext and ggml_flash_attn_ext#14435
ggerganov merged 4 commits into
masterfrom
gg/ggml-batch-soft-max-ops

Commits on Jul 2, 2025
- ggml : support bcast ggml_soft_max_ext, ggml_flash_attn_ext (#14435)
  ggerganov
  committed
  3236670
- vulkan: support softmax/FA batch and broadcast (#14449)
  
  jeffbolznv
  authored and
  ggerganov
  committed
  b726564
- CUDA: broadcasting for FlashAttention mask (#14500)
  
  JohannesGaessler
  authored and
  ggerganov
  committed
  3045a1e
- CUDA: add softmax broadcast (#14475)
  
  am17an
  authored and
  ggerganov
  committed
  3b38afd