A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
not tie_word_embeddings for glm-4.5 and glm-4.5v (#22460)
Signed-off-by: zRzRzRzRzRzRzR <2448370773@qq.com>
Yuxuan Zhang committed
Aug 8, 2025 at 10:37 UTC
c152e2a8a0f49edfc06d760f04ff617310384757
1 file changed
-2
vllm/model_executor/models/glm4_moe.py
-2
@@ -601,8 +601,6 @@ class Glm4MoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
601
quant_config=quant_config)
602
else:
603
self.lm_head = PPMissingLayer()
604
- if self.config.tie_word_embeddings:
605
- self.lm_head.weight = self.model.embed_tokens.weight
604
self.logits_processor = LogitsProcessor(config.vocab_size)
605
self.make_empty_intermediate_tensors = (
606
self.model.make_empty_intermediate_tensors)