From 142e939101d17ca64770e792393bcdde1c5f695b Mon Sep 17 00:00:00 2001 From: AbdullahMPrograms Date: Tue, 28 Jul 2026 10:24:17 -0400 Subject: [PATCH] fix(rpc): remove deprecated -DGGML_HIP_ROCWMMA_FATTN build flag - Remove -DGGML_HIP_ROCWMMA_FATTN=ON from the llama.cpp cmake command and its parameter table row to reflect upstream deprecation --- playbooks/supplemental/clustering-rpc-server/README.md | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/playbooks/supplemental/clustering-rpc-server/README.md b/playbooks/supplemental/clustering-rpc-server/README.md index 756c869c..95656fbc 100644 --- a/playbooks/supplemental/clustering-rpc-server/README.md +++ b/playbooks/supplemental/clustering-rpc-server/README.md @@ -320,7 +320,7 @@ cd llama.cpp Build with ROCm and RPC support: ```bash -cmake -B rocm -DGGML_HIP=ON -DGGML_RPC=ON -DGGML_HIP_ROCWMMA_FATTN=ON -DAMDGPU_TARGETS="gfx1151" +cmake -B rocm -DGGML_HIP=ON -DGGML_RPC=ON -DAMDGPU_TARGETS="gfx1151" cmake --build rocm --config Release -j$(nproc) ``` @@ -328,7 +328,6 @@ cmake --build rocm --config Release -j$(nproc) |-----------|---------| | `-DGGML_HIP=ON` | Enables the ROCm software stack | | `-DGGML_RPC=ON` | Enables RPC for distributed inference | -| `-DGGML_HIP_ROCWMMA_FATTN=ON` | Enables rocWMMA for enhanced Flash Attention on AMD GPUs | | `-DAMDGPU_TARGETS="gfx1151"` | Targets the Ryzen AI Halo GPU (Radeon 8060s) | For more build options, refer to the [llama.cpp build documentation](https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md).