TP/quantization/weight loading refactor part 1 - Simplify parallel linear logic (#1181)

This commit is contained in:
Zhuohan Li
2023-10-02 15:36:09 -07:00
committed by GitHub
parent 84e4e37d14
commit ba0bfd40e2
42 changed files with 819 additions and 1547 deletions
@@ -1,7 +1,7 @@
from vllm.model_executor.layers.quantized_linear.awq import (
AWQColumnParallelLinear, AWQRowParallelLinear)
from vllm.model_executor.parallel_utils.tensor_parallel import (
ColumnParallelLinear, RowParallelLinear)
from vllm.model_executor.parallel_utils.layers import (ColumnParallelLinear,
RowParallelLinear)
_QUANTIZED_LINEAR_REGISTRY = {
"awq": (AWQColumnParallelLinear, AWQRowParallelLinear),
@@ -4,8 +4,8 @@ import torch
from torch.nn.parameter import Parameter
from vllm import quantization_ops
from vllm.model_executor.parallel_utils.tensor_parallel.layers import (
ColumnParallelLinear, RowParallelLinear)
from vllm.model_executor.parallel_utils.layers import (ColumnParallelLinear,
RowParallelLinear)
class AWQColumnParallelLinear(ColumnParallelLinear):