From ba0aaed7bb5dc0b790a34e1e5dfa1bd41b67d22d Mon Sep 17 00:00:00 2001 From: wmantly Date: Wed, 2 Sep 2026 16:02:55 +0000 Subject: [PATCH] feat: integrate xrip/llama.cpp-avx1-numa-sm75 fork with NCCL & NUMA optimizations --- scripts/start-server.sh | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/scripts/start-server.sh b/scripts/start-server.sh index 6708e5a..65b5d38 100755 --- a/scripts/start-server.sh +++ b/scripts/start-server.sh @@ -1,7 +1,7 @@ #!/bin/bash # llama-server launcher for Qwen3.8-27B on 3x Turing GPUs (RTX 2060 12GB + 2x CMP 50HX 10GB) # Built with NVIDIA NCCL for hardware-accelerated multi-GPU tensor parallelism. -export LD_LIBRARY_PATH=/opt/llama.cpp/build-nccl/bin:/opt/minicpm-venv/lib/python3.13/site-packages/nvidia/nccl/lib:/lib/x86_64-linux-gnu +export LD_LIBRARY_PATH=/opt/llama.cpp-xrip/build-nccl/bin:/opt/minicpm-venv/lib/python3.13/site-packages/nvidia/nccl/lib:/lib/x86_64-linux-gnu # Low-Latency NCCL & CUDA Driver Pipeline Optimizations export CUDA_DEVICE_MAX_CONNECTIONS=1 @@ -11,10 +11,11 @@ export NCCL_P2P_DISABLE=0 export NCCL_ALGO=RING export NCCL_PROTO=SIMPLE -exec /opt/llama.cpp/build-nccl/bin/llama-server \ +exec /opt/llama.cpp-xrip/build-nccl/bin/llama-server \ -m /opt/models/gguf/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q4_K_P.gguf \ --spec-type draft-mtp \ --spec-draft-n-max 2 \ + --numa split \ -ngl 99 \ -c 262144 \ --parallel 1 \