From 4120a05ff1d03797dbcd506110f0a997520a6395 Mon Sep 17 00:00:00 2001 From: Jim Smith Date: Thu, 19 Mar 2026 19:21:14 -0400 Subject: [PATCH] Fix AttributeError in Qwen3.5 GDN layers with quantized models (#37448) Signed-off-by: Jim Smith Signed-off-by: mgoin Signed-off-by: Michael Goin Co-authored-by: mgoin Co-authored-by: Xin Yang <105740670+xyang16@users.noreply.github.com> --- vllm/model_executor/models/qwen3_5.py | 4 ++-- vllm/model_executor/models/qwen3_next.py | 4 ++-- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/vllm/model_executor/models/qwen3_5.py b/vllm/model_executor/models/qwen3_5.py index e5967c1229..78dda9ff4f 100644 --- a/vllm/model_executor/models/qwen3_5.py +++ b/vllm/model_executor/models/qwen3_5.py @@ -182,8 +182,8 @@ class Qwen3_5GatedDeltaNet(Qwen3NextGatedDeltaNet): # ============================================================ mixed_qkvz, ba = torch.ops.vllm.gdn_in_proj( hidden_states, - self.in_proj_qkvz.weight.shape[0], - self.in_proj_ba.weight.shape[0], + sum(self.in_proj_qkvz.output_sizes) // self.tp_size, + sum(self.in_proj_ba.output_sizes) // self.tp_size, self.prefix, ) qkv_size = (self.key_dim * 2 + self.value_dim) // self.tp_size diff --git a/vllm/model_executor/models/qwen3_next.py b/vllm/model_executor/models/qwen3_next.py index 7aaded7ae4..bf59c0c117 100644 --- a/vllm/model_executor/models/qwen3_next.py +++ b/vllm/model_executor/models/qwen3_next.py @@ -660,8 +660,8 @@ class Qwen3NextGatedDeltaNet(nn.Module, MambaBase): # ============================================================ projected_states_qkvz, projected_states_ba = torch.ops.vllm.gdn_in_proj( hidden_states, - self.in_proj_qkvz.weight.shape[0], - self.in_proj_ba.weight.shape[0], + sum(self.in_proj_qkvz.output_sizes) // self.tp_size, + sum(self.in_proj_ba.output_sizes) // self.tp_size, self.prefix, ) query, key, value, z, b, a = self.fix_query_key_value_ordering(