xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-22 20:35:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

[ Misc ] Remove separate bias add (#6353)

Browse Source
This commit is contained in:
Robert Shaw 2024-07-12 01:06:09 -04:00 committed by GitHub
parent b6c16cf8ff
commit 6047187cd8
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
vllm/model_executor/layers/linear.py
View File

@ -99,15 +99,7 @@ class LinearMethodBase(QuantizeMethodBase):
class UnquantizedLinearMethod(LinearMethodBase): class UnquantizedLinearMethod(LinearMethodBase):
"""Linear method without quantization. """Linear method without quantization."""
Args:
separate_bias_add: If true, add bias separately after matrix
multiplication.
"""
def __init__(self, separate_bias_add: bool = False):
self.separate_bias_add = separate_bias_add
def create_weights(self, layer: torch.nn.Module, def create_weights(self, layer: torch.nn.Module,
input_size_per_partition: int, input_size_per_partition: int,
@ -126,12 +118,8 @@ class UnquantizedLinearMethod(LinearMethodBase):
layer: torch.nn.Module, layer: torch.nn.Module,
x: torch.Tensor, x: torch.Tensor,
bias: Optional[torch.Tensor] = None) -> torch.Tensor: bias: Optional[torch.Tensor] = None) -> torch.Tensor:
weight = layer.weight
if self.separate_bias_add: return F.linear(x, layer.weight, bias)
if bias is not None:
return F.linear(x, weight) + bias
return F.linear(x, weight)
return F.linear(x, weight, bias)
class LinearBase(torch.nn.Module): class LinearBase(torch.nn.Module):