zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-08-21 20:46:46 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

pre-commit run

Browse Source
This commit is contained in:
zorea 2023-12-29 17:28:14 +02:00
parent 0abc7f0d10
commit 7bf769a292
2 changed files with 10 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
python/mlx/nn/layers/__init__.py
View File

@ -37,7 +37,7 @@ from mlx.nn.layers.containers import Sequential
from mlx.nn.layers.convolution import Conv1d, Conv2d
from mlx.nn.layers.dropout import Dropout, Dropout2d
from mlx.nn.layers.embedding import Embedding
from mlx.nn.layers.linear import Identity, Linear, Bilinear
from mlx.nn.layers.linear import Bilinear, Identity, Linear
from mlx.nn.layers.normalization import BatchNorm, GroupNorm, LayerNorm, RMSNorm
from mlx.nn.layers.positional_encoding import ALiBi, RoPE, SinusoidalPositionalEncoding
from mlx.nn.layers.quantized import QuantizedLinear

14
python/mlx/nn/layers/linear.py
View File

@ -50,7 +50,7 @@ class Linear(Module):
self.reset_parameters()
def reset_parameters(self):
scale = math.sqrt(1. / self.input_dims)
scale = math.sqrt(1.0 / self.input_dims)
self.weight = mx.random.uniform(
low=-scale,
high=scale,
@ -92,7 +92,9 @@ class Bilinear(Module):
not use a bias. Default ``True``.
"""
def __init__(self, input1_dims: int, input2_dims: int, output_dims: int, bias: bool = True):
def __init__(
self, input1_dims: int, input2_dims: int, output_dims: int, bias: bool = True
):
super().__init__()
self.input1_dims = input1_dims
self.input2_dims = input2_dims
@ -104,7 +106,7 @@ class Bilinear(Module):
self.reset_parameters()
def reset_parameters(self):
scale = math.sqrt(1. / self.input1_dims)
scale = math.sqrt(1.0 / self.input1_dims)
self.weight = mx.random.uniform(
low=-scale,
high=scale,
@ -118,8 +120,10 @@ class Bilinear(Module):
)
def _extra_repr(self):
return (f"input1_dims={self.input1_dims}, input2_dims={self.input2_dims}, output_dims={self.output_dims}, "
f"bias={'bias' in self}")
return (
f"input1_dims={self.input1_dims}, input2_dims={self.input2_dims}, output_dims={self.output_dims}, "
f"bias={'bias' in self}"
)
def __call__(self, input1, input2):
output = (input1 @ self.weight * input2.reshape(1, *input2.shape)).sum(-1).T