Spaces:
Running
on
Zero
Running
on
Zero
File size: 2,004 Bytes
42f2c22 |
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 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 |
# // Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
# //
# // 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 typing import Optional
import torch
import torch.nn.functional as F
from torch import nn
def get_mlp(mlp_type: Optional[str] = "normal"):
if mlp_type == "normal":
return MLP
elif mlp_type == "swiglu":
return SwiGLUMLP
class MLP(nn.Module):
def __init__(
self,
dim: int,
expand_ratio: int,
):
super().__init__()
self.proj_in = nn.Linear(dim, dim * expand_ratio)
self.act = nn.GELU("tanh")
self.proj_out = nn.Linear(dim * expand_ratio, dim)
def forward(self, x: torch.FloatTensor) -> torch.FloatTensor:
x = self.proj_in(x)
x = self.act(x)
x = self.proj_out(x)
return x
class SwiGLUMLP(nn.Module):
def __init__(
self,
dim: int,
expand_ratio: int,
multiple_of: int = 256,
):
super().__init__()
hidden_dim = int(2 * dim * expand_ratio / 3)
hidden_dim = multiple_of * ((hidden_dim + multiple_of - 1) // multiple_of)
self.proj_in_gate = nn.Linear(dim, hidden_dim, bias=False)
self.proj_out = nn.Linear(hidden_dim, dim, bias=False)
self.proj_in = nn.Linear(dim, hidden_dim, bias=False)
def forward(self, x: torch.FloatTensor) -> torch.FloatTensor:
x = self.proj_out(F.silu(self.proj_in_gate(x)) * self.proj_in(x))
return x
|