]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
vulkan: add FA dequant for q4_1, q5_0, q5_1, iq4_nl (llama/21029)
authormkoker <redacted>
Tue, 7 Apr 2026 11:41:29 +0000 (07:41 -0400)
committerGeorgi Gerganov <redacted>
Thu, 30 Apr 2026 08:29:02 +0000 (11:29 +0300)
commit18c98ffaf7355917935915cfedd95414fccdc1a2
tree9e7f4cb91ade2eb55446c468410f0a88725f6610
parenta1f76fb4cfd05ed08c96d2f569379551f6e6989f
vulkan: add FA dequant for q4_1, q5_0, q5_1, iq4_nl (llama/21029)

Add dequantize4() implementations for Q4_1, Q5_0, Q5_1, and IQ4_NL
in the flash attention base shader. Register them in the shader
generator, pipeline creation, and enable in the scalar/coopmat1 FA
support check.
ggml/src/ggml-vulkan/ggml-vulkan.cpp
ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp