forked from modularml/max
-
Notifications
You must be signed in to change notification settings - Fork 0
/
normalization.🔥
28 lines (23 loc) · 1.08 KB
/
normalization.🔥
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
# ===----------------------------------------------------------------------=== #
# Copyright (c) 2024, Modular Inc. All rights reserved.
#
# Licensed under the Apache License v2.0 with LLVM Exceptions:
# https://llvm.org/LICENSE.txt
#
# 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.
# ===----------------------------------------------------------------------=== #
"""Normalization layers."""
from max.graph import ops, Symbol
@value
struct RMSNorm:
var eps: Float64
var weight: Symbol
def __call__(self, input: Symbol) -> Symbol:
scale = ops.rsqrt(ops.mean(input**2.0, axis=-1) + self.eps)
# Since norm weights are float32, cast to input dtype to avoid
# promoting the result to float32 when the input is float16.
return input * scale * ops.cast(self.weight, input.tensor_type().dtype)