From 85bda9e7d05371af6bb9d0052b1eb2f85d3cde29 Mon Sep 17 00:00:00 2001
From: Yuxuan Zhang <2448370773@qq.com>
Date: Thu, 24 Jul 2025 16:52:43 +0800
Subject: [PATCH] remove GLM-4.5 quantization wrong Code (#21435)

---
 vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py | 2 +-
 vllm/model_executor/models/glm4_moe.py                       | 1 -
 vllm/reasoning/glm4_moe_reasoning_parser.py                  | 2 +-
 3 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py b/vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py
index c3f9d7923..40cdf7275 100644
--- a/vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py
+++ b/vllm/entrypoints/openai/tool_parsers/glm4_moe_tool_parser.py
@@ -20,7 +20,7 @@ from vllm.transformers_utils.tokenizer import AnyTokenizer
 logger = init_logger(__name__)
 
 
-@ToolParserManager.register_module("glm4_moe")
+@ToolParserManager.register_module("glm45")
 class Glm4MoeModelToolParser(ToolParser):
 
     def __init__(self, tokenizer: AnyTokenizer):
diff --git a/vllm/model_executor/models/glm4_moe.py b/vllm/model_executor/models/glm4_moe.py
index bdca293d2..095bfbc40 100644
--- a/vllm/model_executor/models/glm4_moe.py
+++ b/vllm/model_executor/models/glm4_moe.py
@@ -390,7 +390,6 @@ class Glm4MoeModel(nn.Module):
             self.embed_tokens = VocabParallelEmbedding(
                 config.vocab_size,
                 config.hidden_size,
-                quant_config=quant_config,
                 prefix=f"{prefix}.embed_tokens")
         else:
             self.embed_tokens = PPMissingLayer()
diff --git a/vllm/reasoning/glm4_moe_reasoning_parser.py b/vllm/reasoning/glm4_moe_reasoning_parser.py
index 6511fb49d..460e38d2d 100644
--- a/vllm/reasoning/glm4_moe_reasoning_parser.py
+++ b/vllm/reasoning/glm4_moe_reasoning_parser.py
@@ -14,7 +14,7 @@ from vllm.reasoning import ReasoningParser, ReasoningParserManager
 logger = init_logger(__name__)
 
 
-@ReasoningParserManager.register_module("glm4_moe")
+@ReasoningParserManager.register_module("glm45")
 class Glm4MoeModelReasoningParser(ReasoningParser):
     """
     Reasoning parser for the Glm4MoeModel model.