TP/quantization/weight loading refactor part 1 - Simplify parallel linear logic (#1181)
This commit is contained in:
@@ -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):
|
||||
|
||||
Reference in New Issue
Block a user