llama.cpp/ggml/src/ggml-vulkan
Jeff Bolz 303f8615e9
vulkan: Multi-pass softmax for large number of cols (#17892)
When the number of cols is large, split each row across multiple workgroups.
There are three phases that communicate partial results through temp buffers:
(1) compute max partials
(2) take max of partials, compute sum(exp(x-max)) partials
(3) sum partials, compute scaled result
2025-12-13 10:04:29 +01:00
..
cmake cmake: fix ggml-shaders-gen compiler paths containing spaces (#12747) 2025-04-04 10:12:40 -03:00
vulkan-shaders vulkan: Multi-pass softmax for large number of cols (#17892) 2025-12-13 10:04:29 +01:00
CMakeLists.txt vulkan: Improve build time for MSVC (#16545) 2025-10-14 14:51:36 +02:00
ggml-vulkan.cpp vulkan: Multi-pass softmax for large number of cols (#17892) 2025-12-13 10:04:29 +01:00