File size: 4,822 Bytes
0102e16
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
#!/usr/bin/env python3
# -*- encoding: utf-8 -*-
# Copyright FunASR (https://github.com/alibaba-damo-academy/FunASR). All Rights Reserved.
#  MIT License  (https://opensource.org/licenses/MIT)

import torch
from typing import Dict, Optional, Tuple

from funasr_detach.models.transformer.layer_norm import LayerNorm
from funasr_detach.models.rwkv_bat.rwkv_feed_forward import FeedForward
from funasr_detach.models.rwkv_bat.rwkv_attention import (
    EncoderSelfAttention,
    DecoderSelfAttention,
)


class RWKV(torch.nn.Module):
    """RWKV module.

    Args:
        size: Input/Output size.
        linear_size: Feed-forward hidden size.
        attention_size: SelfAttention hidden size.
        context_size: Context size for WKV computation.
        block_id: Block index.
        num_blocks: Number of blocks in the architecture.
        normalization_class: Normalization layer class.
        normalization_args: Normalization layer arguments.
        att_dropout_rate: Dropout rate for the attention module.
        ffn_dropout_rate: Dropout rate for the feed-forward module.

    """

    def __init__(
        self,
        size: int,
        linear_size: int,
        attention_size: int,
        context_size: int,
        block_id: int,
        num_blocks: int,
        att_dropout_rate: float = 0.0,
        ffn_dropout_rate: float = 0.0,
        dropout_rate: float = 0.0,
    ) -> None:
        """Construct a RWKV object."""
        super().__init__()

        self.layer_norm_att = LayerNorm(size)
        self.layer_norm_ffn = LayerNorm(size)

        self.att = EncoderSelfAttention(
            size, attention_size, context_size, block_id, att_dropout_rate, num_blocks
        )
        self.dropout_att = torch.nn.Dropout(p=dropout_rate)

        self.ffn = FeedForward(
            size, linear_size, block_id, ffn_dropout_rate, num_blocks
        )
        self.dropout_ffn = torch.nn.Dropout(p=dropout_rate)

    def forward(
        self,
        x: torch.Tensor,
        state: Optional[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
        """Compute receptance weighted key value.

        Args:
            x: RWKV input sequences. (B, L, size)
            state: Decoder hidden states. [5 x (B, D_att/size, N)]

        Returns:
            x: RWKV output sequences. (B, L, size)
            x: Decoder hidden states. [5 x (B, D_att/size, N)]

        """
        att, state = self.att(self.layer_norm_att(x), state=state)
        x = x + self.dropout_att(att)
        ffn, state = self.ffn(self.layer_norm_ffn(x), state=state)
        x = x + self.dropout_ffn(ffn)
        return x, state


class RWKVDecoderLayer(torch.nn.Module):
    """RWKV module.

    Args:
        size: Input/Output size.
        linear_size: Feed-forward hidden size.
        attention_size: SelfAttention hidden size.
        context_size: Context size for WKV computation.
        block_id: Block index.
        num_blocks: Number of blocks in the architecture.
        normalization_class: Normalization layer class.
        normalization_args: Normalization layer arguments.
        att_dropout_rate: Dropout rate for the attention module.
        ffn_dropout_rate: Dropout rate for the feed-forward module.

    """

    def __init__(
        self,
        size: int,
        linear_size: int,
        attention_size: int,
        context_size: int,
        block_id: int,
        num_blocks: int,
        att_dropout_rate: float = 0.0,
        ffn_dropout_rate: float = 0.0,
        dropout_rate: float = 0.0,
    ) -> None:
        """Construct a RWKV object."""
        super().__init__()

        self.layer_norm_att = LayerNorm(size)
        self.layer_norm_ffn = LayerNorm(size)

        self.att = DecoderSelfAttention(
            size, attention_size, context_size, block_id, att_dropout_rate, num_blocks
        )
        self.dropout_att = torch.nn.Dropout(p=dropout_rate)

        self.ffn = FeedForward(
            size, linear_size, block_id, ffn_dropout_rate, num_blocks
        )
        self.dropout_ffn = torch.nn.Dropout(p=dropout_rate)

    def forward(
        self,
        x: torch.Tensor,
        state: Optional[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
        """Compute receptance weighted key value.

        Args:
            x: RWKV input sequences. (B, L, size)
            state: Decoder hidden states. [5 x (B, D_att/size, N)]

        Returns:
            x: RWKV output sequences. (B, L, size)
            x: Decoder hidden states. [5 x (B, D_att/size, N)]

        """
        att, state = self.att(self.layer_norm_att(x), state=state)
        x = x + self.dropout_att(att)

        ffn, state = self.ffn(self.layer_norm_ffn(x), state=state)
        x = x + self.dropout_ffn(ffn)

        return x, state