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 8, 11:15 UTC
[Bugfix] fix default is_neox_style is True for deepseek (#34353)
Signed-off-by: dongxinyu03 <dongxinyu03@baidu.com>
Xinyu Dong committed
Feb 12, 2026 at 02:20 UTC
be7f3d5d2016b326d12ff582a8c9f96a68217c7a
1 file changed
+1
-1
vllm/model_executor/models/deepseek_v2.py
+1
-1
@@ -836,7 +836,7 @@ class DeepseekV2MLAAttention(nn.Module):
836
qk_rope_head_dim,
837
max_position=max_position_embeddings,
838
rope_parameters=config.rope_parameters,
839
- is_neox_style=not getattr(config, "indexer_rope_interleave", True),
839
+ is_neox_style=not getattr(config, "indexer_rope_interleave", False),
840
)
841
self.indexer = Indexer(
842
vllm_config,