Workaround broken ROCm p2p copy

Enable the build flag for llama.cpp to use CPU copy for multi-GPU scenarios.
2025-12-10 07:46:59 +00:00 · 2024-07-05 12:46:28 -07:00
parent 53da2c6965
commit 0bacb30007
2 changed files with 2 additions and 1 deletions
--- a/llm/generate/gen_windows.ps1
+++ b/llm/generate/gen_windows.ps1
@@ -366,6 +366,7 @@ function build_rocm() {
            "-DCMAKE_C_COMPILER=clang.exe",
            "-DCMAKE_CXX_COMPILER=clang++.exe",
            "-DGGML_HIPBLAS=on",
+            "-DLLAMA_CUDA_NO_PEER_COPY=on",
            "-DHIP_PLATFORM=amd",
            "-DGGML_AVX=on",
            "-DGGML_AVX2=off",