biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V0 deprecation] Remove QKVCrossParallelLinear implementation (#26475)

Browse Source 
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
This commit is contained in:
Isotr0py
2025-10-09 18:52:27 +08:00
committed by GitHub
parent ec10fd0abc
commit d1ddf340c8
4 changed files with 2 additions and 255 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/model_loader/utils.py
View File

@@ -17,7 +17,6 @@ from vllm.attention import Attention
from vllm.attention.layer import MLAAttention
from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
from vllm.logger import init_logger
from vllm.model_executor.layers.linear import QKVCrossParallelLinear
from vllm.model_executor.layers.quantization.base_config import (
QuantizationConfig,
QuantizeMethodBase,
@@ -108,11 +107,6 @@ def process_weights_after_loading(
maybe_save_metadata_and_attributes_for_weight_reloading(model, model_config)
for _, module in model.named_modules():
if isinstance(module, QKVCrossParallelLinear):
# NOTE(Isotr0py): special case for cross QKV layer because
# q and kv proj aren't registered as submodules intentionally
module.process_weights_after_loading()
continue
quant_method = getattr(module, "quant_method", None)
if isinstance(quant_method, QuantizeMethodBase):
# When quant methods need to process weights after loading