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 12, 22:15 UTC
[CI] Fix precommit `rope_theta` issue (#29040)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Nov 19, 2025 at 17:22 UTC
0075bfffd4201d1377f0d048848f82911e917639
1 file changed
+1
-2
vllm/model_executor/models/deepseek_v2.py
+1
-2
@@ -991,8 +991,7 @@ class DeepseekV2MLAAttention(nn.Module):
991
qk_rope_head_dim,
992
rotary_dim=qk_rope_head_dim,
993
max_position=max_position_embeddings,
994
- base=rope_theta,
995
- rope_scaling=rope_scaling,
994
+ rope_parameters=config.rope_parameters,
995
is_neox_style=True,
996
)
997
self.indexer = Indexer(