Add q8_0 kernels to kompute shaders and bump to latest llama/gguf.

1 year ago · f605a5b686
parent 1534df3e9f
commit f605a5b686
2 changed files with 3 additions and 1 deletions
--- a/gpt4all-backend/llama.cpp-mainline
+++ b/gpt4all-backend/llama.cpp-mainline
@ -1 +1 @@
-Subproject commit 7f8f62880cf3242cc0d3b54465bed4e2d59e4774
+Subproject commit 2b7b3f37f9d0193943c3d31e3e617bc6351838c8
--- a/gpt4all-backend/llama.cpp.cmake
+++ b/gpt4all-backend/llama.cpp.cmake
@ -239,6 +239,7 @@ if (LLAMA_KOMPUTE)
          kompute/op_rmsnorm.comp
          kompute/op_diagmask.comp
          kompute/op_mul_mat_f16.comp
+          kompute/op_mul_mat_q8_0.comp
          kompute/op_mul_mat_q4_0.comp
          kompute/op_mul_mat_q4_1.comp
          kompute/op_mul_mat_q6_k.comp
@ -268,6 +269,7 @@ if (LLAMA_KOMPUTE)
          shaderop_rmsnorm.h
          shaderop_diagmask.h
          shaderop_mul_mat_f16.h
+          shaderop_mul_mat_q8_0.h
          shaderop_mul_mat_q4_0.h
          shaderop_mul_mat_q4_1.h
          shaderop_mul_mat_q6_k.h