From e2ea7aafadaf02f3124051ddd79e64271e43b59e Mon Sep 17 00:00:00 2001 From: Brayden Zhong Date: Tue, 7 Jul 2026 13:20:36 -0700 Subject: [PATCH] [Cherry pick to release/v0.5.15] Fix NVFP4 online quantization (#30397) Co-authored-by: Brayden Zhong --- python/sglang/srt/layers/quantization/nvfp4_online.py | 1 + 1 file changed, 1 insertion(+) diff --git a/python/sglang/srt/layers/quantization/nvfp4_online.py b/python/sglang/srt/layers/quantization/nvfp4_online.py index 89103ab0f..1caa48596 100644 --- a/python/sglang/srt/layers/quantization/nvfp4_online.py +++ b/python/sglang/srt/layers/quantization/nvfp4_online.py @@ -85,6 +85,7 @@ class NvFp4OnlineConfig(ModelOptQuantConfig): self.use_per_token_activation = True self.is_checkpoint_fp8_serialized = is_checkpoint_fp8_serialized self.is_fp4_experts = False + self.dequant_fp4_to_fp8 = False self.activation_scheme = activation_scheme self.weight_block_size = weight_block_size self.use_mxfp8 = use_mxfp8