From 4b643c463e31e0513c4b722c8c0754685159c6fa Mon Sep 17 00:00:00 2001 From: Woosuk Kwon Date: Sun, 28 Jun 2026 08:37:00 -0700 Subject: [PATCH] [GLM5] Fix minor typo (#46961) Signed-off-by: Woosuk Kwon --- vllm/models/deepseek_v32/nvidia/attention.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/models/deepseek_v32/nvidia/attention.py b/vllm/models/deepseek_v32/nvidia/attention.py index 420e4e9378..771a3d4f95 100644 --- a/vllm/models/deepseek_v32/nvidia/attention.py +++ b/vllm/models/deepseek_v32/nvidia/attention.py @@ -392,7 +392,7 @@ class DeepseekV32Attention(MLAAttention): has_indexer = True indexer_k_norm_w = self.indexer.k_norm.weight indexer_k_norm_bias = self.indexer.k_norm.bias - indexer_k_norm_eps = self.indexer.k_norm.variance_epsilon + indexer_k_norm_eps = self.indexer.k_norm.eps indexer_k_rope_cos_sin_cache = self.indexer_rope_emb.cos_sin_cache indexer_k_cache = self.indexer.k_cache.kv_cache indexer_softmax_scale = self.indexer.softmax_scale