File size: 1,539 Bytes
d643072
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright 2024 MIT Han Lab
#
# 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.
#
# SPDX-License-Identifier: Apache-2.0

import torch
from flash_attn import flash_attn_func
from torch import nn
from torch.nn import functional as F


class FlashAttention(nn.Module):
    def __init__(self, dim: int, num_heads: int):
        super().__init__()
        self.dim = dim
        assert dim % num_heads == 0
        self.num_heads = num_heads
        self.head_dim = dim // num_heads

        self.qkv = nn.Linear(dim, dim * 3, bias=False)
        self.proj_out = torch.nn.Linear(dim, dim)

    def forward(self, x):
        B, N, C = x.shape
        qkv = self.qkv(x).view(B, N, 3, C)  # B, N, 3, C
        q, k, v = qkv.unbind(2)  # B, N, C
        k = k.reshape(B, N, self.num_heads, self.head_dim)
        v = v.reshape(B, N, self.num_heads, self.head_dim)
        q = q.reshape(B, N, self.num_heads, self.head_dim)
        out = flash_attn_func(q, k, v)  # B, N, H, c
        out = self.proj_out(out.view(B, N, C))  # B, N, C
        return out