From 76d17c7ecb03c3a181fb8cb053ba28611a76b210 Mon Sep 17 00:00:00 2001 From: lambert0312 Date: Wed, 23 Apr 2025 00:22:26 +0800 Subject: [PATCH] Fix shared experts fusion error without quantization (#5632) --- python/sglang/srt/models/deepseek_v2.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/sglang/srt/models/deepseek_v2.py b/python/sglang/srt/models/deepseek_v2.py index 241c46cf9..c5afc559d 100644 --- a/python/sglang/srt/models/deepseek_v2.py +++ b/python/sglang/srt/models/deepseek_v2.py @@ -1604,7 +1604,7 @@ class DeepseekV2ForCausalLM(nn.Module): if self.n_share_experts_fusion > 0: weights_list = list(weights) weights_dict = dict(weights_list) - if self.quant_config.get_name() == "w8a8_int8": + if self.quant_config is None or self.quant_config.get_name() == "w8a8_int8": suffix_list = [ "down_proj.weight", "down_proj.weight_scale",