From 28e915b474eba6d132a65b28c8325b1bbc3f572a Mon Sep 17 00:00:00 2001 From: Jimmy Shong <69131491+Jiminator@users.noreply.github.com> Date: Wed, 15 Apr 2026 13:25:36 -0700 Subject: [PATCH] [Bugfix] Preserve auto-detected quant_config for GLM NextN draft model (#22823) --- python/sglang/srt/models/glm4_moe_nextn.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/python/sglang/srt/models/glm4_moe_nextn.py b/python/sglang/srt/models/glm4_moe_nextn.py index 6d2ef8972..dfbd4583d 100644 --- a/python/sglang/srt/models/glm4_moe_nextn.py +++ b/python/sglang/srt/models/glm4_moe_nextn.py @@ -129,7 +129,8 @@ class Glm4MoeForCausalLMNextN(Glm4MoeForCausalLM): self.config = config self.tp_size = get_tensor_model_parallel_world_size() self.needs_quant_draft = ( - get_global_server_args().speculative_draft_model_quantization + get_global_server_args().speculative_draft_model_quantization is not None + or quant_config is not None ) quant_config = quant_config if self.needs_quant_draft else None self.model = Glm4MoeModelNextN(