Add Cuda Graphs in CMakeList (ggml-org#947)

Nexesenex · web-flow · commit a5a32b9179a2 · 2024-06-29T10:10:51.000+08:00
original LCPP PR 3242 by Slaren
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -49,6 +49,7 @@ option(LLAMA_CUDA_F16                        "llama: use 16 bit floats for dmmv
 set(LLAMA_CUDA_KQUANTS_ITER "2" CACHE STRING "llama: iters./thread per block for Q2_K/Q6_K")
 set(LLAMA_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING
                                              "llama: max. batch size for using peer access")
+set(GGML_CUDA_USE_GRAPHS                                                                        ON)
 option(LLAMA_HIPBLAS                         "llama: use hipBLAS"                               OFF)
 
 # Other
