[refactor] remove redundant code in linear

2025-12-24 12:02:09 +08:00
parent 75d0bdae2f
commit 383eb5459a
8 changed files with 37 additions and 1570 deletions
--- a/vllm_kunlun/models/qwen2.py
+++ b/vllm_kunlun/models/qwen2.py
@@ -40,7 +40,7 @@ from vllm.config import CacheConfig, VllmConfig
 from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm_kunlun.ops.activation import SiluAndMul
 from vllm.model_executor.layers.layernorm import RMSNorm
-from vllm_kunlun.ops.linear import (MergedColumnParallelLinear,
+from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
                                               QKVParallelLinear,
                                               RowParallelLinear)
 from vllm.model_executor.layers.logits_processor import LogitsProcessor