From 8dc85b54f4e961513436c7b5a779f2a9134a1811 Mon Sep 17 00:00:00 2001 From: gp513 Date: Fri, 12 Sep 2025 11:18:29 +0800 Subject: [PATCH] update readme of swap-attention for cpu affinity --- docs/features/swap_attention.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/features/swap_attention.md b/docs/features/swap_attention.md index 57aee3e2..434c9f3a 100644 --- a/docs/features/swap_attention.md +++ b/docs/features/swap_attention.md @@ -47,5 +47,5 @@ ## 注意事项： 1. `--recompute-num-layers [int]`中的[int]层数指的是每一个pp stage的层数。[int]的取值应该小于等于num-layers/pipeline-model-parallel-size. -2. 若出现性能波动，可能是跨NUMA内存访问引起，可尝试通过进程绑核缓解 `export CPU_AFFINITY_CONF=1` +2. 若出现性能波动，可能是跨NUMA内存访问引起，可尝试通过进程绑核缓解 `export CPU_AFFINITY_CONF=1,lazy_bind:0` 3. `--swap-attention`暂不兼容LoRA微调。 \ No newline at end of file -- Gitee