[Quantization] Support kimi-k3 routed expert quant (#57430)
Signed-off-by: Kyle Sayers <kylesayrs@gmail.com> Co-authored-by: Misha Goin <mgoin64@gmail.com>
Kyle Sayers committed
Sep 18, 2026 at 10:03 UTC
76d517fd1279805fa2318c28343fcd57b030ead3
1 file changed
+2
-2
vllm/models/kimi_k3/nvidia/model.py
+2
-2
@@ -677,7 +677,7 @@ class KimiMoE(nn.Module):
677
hidden_size,
678
self.moe_hidden_size,
679
bias=False,
680
- quant_config=None,
680
+ quant_config=quant_config,
681
prefix=f"{prefix}.routed_expert_down_proj",
682
)
683
self.routed_expert_norm = (
@@ -694,7 +694,7 @@ class KimiMoE(nn.Module):
694
self.moe_hidden_size,
695
hidden_size,
696
bias=False,
697
- quant_config=None,
697
+ quant_config=quant_config,
698
prefix=f"{prefix}.routed_expert_up_proj",
699
)
700