From ba9d5aed98ef304338da1f7c48bd71456095b728 Mon Sep 17 00:00:00 2001 From: Brayden Zhong Date: Mon, 22 Jun 2026 17:27:33 -0700 Subject: [PATCH] Fix nightly CI test for Kimi K2.5 INT4 + H200 (#28746) --- .../schemes/compressed_tensors_wNa16_moe.py | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py index 5b0198712..942e3b8f0 100644 --- a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py +++ b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_wNa16_moe.py @@ -130,7 +130,11 @@ class CompressedTensorsWNA16MoE(CompressedTensorsMoEScheme): # In the case where we have actorder/g_idx, # we do not partition the w2 scales - load_full_w2 = (self.actorder != "static") and self.group_size != -1 + load_full_w2 = ( + self.actorder is not None + and self.actorder != "static" + and self.group_size != -1 + ) if load_full_w2: w2_scales_size = intermediate_size_per_partition * layer.moe_tp_size