From 65055774596d53362ef8c8340ed155ff12cd0d54 Mon Sep 17 00:00:00 2001 From: Lei Ke <1141466880@qq.com> Date: Mon, 14 Sep 2026 06:20:59 +0800 Subject: [PATCH] [Doc][NPU] Keep AllGather for MiniMax-H3 INT8 DLO (#7444) Signed-off-by: KrystalRay Co-authored-by: KrystalRay --- recipes/MiniMaxAI/MiniMax-H3-NPU.md | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/recipes/MiniMaxAI/MiniMax-H3-NPU.md b/recipes/MiniMaxAI/MiniMax-H3-NPU.md index b4563c1f7..c6ee0904b 100755 --- a/recipes/MiniMaxAI/MiniMax-H3-NPU.md +++ b/recipes/MiniMaxAI/MiniMax-H3-NPU.md @@ -252,8 +252,9 @@ throughput guarantee. supports the H3 native `tile` mode only. - RainFusion block-sparse attention and INT8 quantization are validated for T2VA only; use the BF16 dense configuration for FL2VA and Ref2VA. -- Online quantization cannot be combined with distributed layerwise offload - while AllGather is enabled; pass `--dlo-no-use-allgather` in that case. +- Online INT8 quantization can be combined with distributed layerwise offload + while AllGather is enabled. Keep AllGather enabled for SP>1; use + `--dlo-no-use-allgather` only when collective communication is unavailable. ## Additional resources