From 76d517fd1279805fa2318c28343fcd57b030ead3 Mon Sep 17 00:00:00 2001 From: Kyle Sayers Date: Fri, 18 Sep 2026 10:03:59 -0400 Subject: [PATCH] [Quantization] Support kimi-k3 routed expert quant (#57430) Signed-off-by: Kyle Sayers Co-authored-by: Misha Goin --- vllm/models/kimi_k3/nvidia/model.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/models/kimi_k3/nvidia/model.py b/vllm/models/kimi_k3/nvidia/model.py index a970479da8..737184b9d8 100644 --- a/vllm/models/kimi_k3/nvidia/model.py +++ b/vllm/models/kimi_k3/nvidia/model.py @@ -677,7 +677,7 @@ class KimiMoE(nn.Module): hidden_size, self.moe_hidden_size, bias=False, - quant_config=None, + quant_config=quant_config, prefix=f"{prefix}.routed_expert_down_proj", ) self.routed_expert_norm = ( @@ -694,7 +694,7 @@ class KimiMoE(nn.Module): self.moe_hidden_size, hidden_size, bias=False, - quant_config=None, + quant_config=quant_config, prefix=f"{prefix}.routed_expert_up_proj", )