Source code for tensorrt_llm.layers.normalization

# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ..functional import group_norm, layer_norm, rms_norm
from ..module import Module
from ..parameter import Parameter


[docs] class LayerNorm(Module): def __init__(self, normalized_shape, eps=1e-05, elementwise_affine=True, bias=True, dtype=None, tp_size=1, tp_dim=-1): super().__init__() if isinstance(normalized_shape, int): normalized_shape = (normalized_shape, ) self.normalized_shape = tuple(normalized_shape) self.elementwise_affine = elementwise_affine if self.elementwise_affine: self.weight = Parameter(shape=self.normalized_shape, dtype=dtype) if bias: self.bias = Parameter(shape=self.normalized_shape, dtype=dtype) else: self.register_parameter('bias', None) else: self.register_parameter('weight', None) self.register_parameter('bias', None) self.eps = eps self.dtype = dtype self.tp_size = tp_size self.tp_dim = tp_dim
[docs] def forward(self, x, normalized_shape=None): weight = 1. if self.weight is None else self.weight.value bias = 0. if self.bias is None else self.bias.value if normalized_shape is None: normalized_shape = self.normalized_shape return layer_norm(x, normalized_shape, weight, bias, self.eps)
[docs] class RmsNorm(Module): def __init__(self, normalized_shape, num_groups=1, eps=1e-06, elementwise_affine=True, dtype=None): super().__init__() if isinstance(normalized_shape, int): normalized_shape = (normalized_shape, ) self.normalized_shape = tuple(normalized_shape) self.elementwise_affine = elementwise_affine self.num_groups = num_groups num_channels = normalized_shape[-1] if num_channels % num_groups != 0: raise ValueError('num_channels must be divisible by num_groups') if self.elementwise_affine: self.weight = Parameter(shape=self.normalized_shape, dtype=dtype) else: self.register_parameter('weight', None) self.eps = eps self.dtype = dtype
[docs] def forward(self, x, normalized_shape=None): weight = None if self.weight is None else self.weight.value if normalized_shape is None: normalized_shape = self.normalized_shape return rms_norm(x, normalized_shape, self.num_groups, weight, self.eps)
[docs] class GroupNorm(Module): def __init__(self, num_groups, num_channels, eps=1e-05, affine=True, dtype=None): super().__init__() if num_channels % num_groups != 0: raise ValueError('num_channels must be divisible by num_groups') self.num_groups = num_groups self.num_channels = num_channels self.affine = affine if self.affine: self.weight = Parameter(shape=(self.num_channels, ), dtype=dtype) self.bias = Parameter(shape=(self.num_channels, ), dtype=dtype) else: self.register_parameter('weight', None) self.register_parameter('bias', None) self.eps = eps
[docs] def forward(self, x): weight = None if self.weight is None else self.weight.value bias = None if self.bias is None else self.bias.value return group_norm(x, self.num_groups, weight, bias, self.eps)