]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
vulkan: fix NaN issue in flash attention shader (llama/12776)
authorJeff Bolz <redacted>
Sun, 6 Apr 2025 09:03:47 +0000 (04:03 -0500)
committerGeorgi Gerganov <redacted>
Thu, 24 Apr 2025 17:39:16 +0000 (20:39 +0300)
commit3c26dd3353bced6fa88c2e7e6c7d921a1e09dfcd
treeed365cf195c5d4df09188573eea8703f17c0aa27
parentd792d2a2dc713ee9b4b7835663e694fef9f11a69
vulkan: fix NaN issue in flash attention shader (llama/12776)

Use -FLT_MAX/2 rather than -inf as the initial value for computing the maximum.
ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp