espnet/test/test_positional_encoding.py at master · phorvicheka-python/espnet · GitHub

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
154
155
156
157
158
159
160
161
import pytest
import torch

from espnet.nets.pytorch_backend.transformer.embedding import (
    LearnableFourierPosEnc,
    PositionalEncoding,
    ScaledPositionalEncoding,
)


@pytest.mark.parametrize(
    "dtype, device",
    [(dt, dv) for dt in ("float32", "float64") for dv in ("cpu", "cuda")],
)
def test_pe_extendable(dtype, device):
    if device == "cuda" and not torch.cuda.is_available():
        pytest.skip("no cuda device is available")
    dtype = getattr(torch, dtype)
    dim = 2
    pe = PositionalEncoding(dim, 0.0, 3).to(dtype=dtype, device=device)
    x = torch.rand(2, 3, dim, dtype=dtype, device=device)
    y = pe(x)
    init_cache = pe.pe

    # test not extended from init
    x = torch.rand(2, 3, dim, dtype=dtype, device=device)
    y = pe(x)
    assert pe.pe is init_cache

    x = torch.rand(2, 5, dim, dtype=dtype, device=device)
    y = pe(x)

    sd = pe.state_dict()
    assert len(sd) == 0, "PositionalEncoding should save nothing"
    pe2 = PositionalEncoding(dim, 0.0, 3).to(dtype=dtype, device=device)
    pe2.load_state_dict(sd)
    y2 = pe2(x)
    assert torch.allclose(y, y2)


@pytest.mark.parametrize(
    "dtype, device, apply_scaling, hidden_dim",
    [
        (dt, dv, scal, hd)
        for dt in ("float32", "float64")
        for dv in ("cpu", "cuda")
        for scal in [True, False]
        for hd in [None, 12]
    ],
)
def test_learnedFourierPe_extendable(dtype, device, apply_scaling, hidden_dim):
    if device == "cuda" and not torch.cuda.is_available():
        pytest.skip("no cuda device is available")
    dtype = getattr(torch, dtype)
    dim = 2
    pe = LearnableFourierPosEnc(
        dim, apply_scaling=apply_scaling, hidden_dim=hidden_dim
    ).to(dtype=dtype, device=device)
    x = torch.rand(2, 3, dim, dtype=dtype, device=device)
    pe(x)

    x = torch.rand(2, 5, dim, dtype=dtype, device=device)
    pe(x)


@pytest.mark.parametrize(
    "dtype, device",
    [(dt, dv) for dt in ("float32", "float64") for dv in ("cpu", "cuda")],
)
def test_scaled_pe_extendable(dtype, device):
    if device == "cuda" and not torch.cuda.is_available():
        pytest.skip("no cuda device is available")
    dtype = getattr(torch, dtype)
    dim = 2
    pe = ScaledPositionalEncoding(dim, 0.0, 3).to(dtype=dtype, device=device)
    x = torch.rand(2, 3, dim, dtype=dtype, device=device)
    y = pe(x)
    init_cache = pe.pe

    # test not extended from init
    x = torch.rand(2, 3, dim, dtype=dtype, device=device)
    y = pe(x)
    assert pe.pe is init_cache

    x = torch.rand(2, 5, dim, dtype=dtype, device=device)
    y = pe(x)

    sd = pe.state_dict()
    assert sd == {"alpha": pe.alpha}, "ScaledPositionalEncoding should save only alpha"
    pe2 = ScaledPositionalEncoding(dim, 0.0, 3).to(dtype=dtype, device=device)
    pe2.load_state_dict(sd)
    y2 = pe2(x)
    assert torch.allclose(y, y2)


class LegacyPositionalEncoding(torch.nn.Module):
    """Positional encoding module until v.0.5.2."""

    def __init__(self, d_model, dropout_rate, max_len=5000):
        import math

        super().__init__()
        self.dropout = torch.nn.Dropout(p=dropout_rate)
        # Compute the positional encodings once in log space.
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float32).unsqueeze(1)
        div_term = torch.exp(
            torch.arange(0, d_model, 2, dtype=torch.float32)
            * -(math.log(10000.0) / d_model)
        )
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0)
        self.max_len = max_len
        self.xscale = math.sqrt(d_model)
        self.register_buffer("pe", pe)

    def forward(self, x):
        x = x * self.xscale + self.pe[:, : x.size(1)]
        return self.dropout(x)


class LegacyScaledPositionalEncoding(LegacyPositionalEncoding):
    """Positional encoding module until v.0.5.2."""

    def __init__(self, d_model, dropout_rate, max_len=5000):
        super().__init__(d_model=d_model, dropout_rate=dropout_rate, max_len=max_len)
        self.alpha = torch.nn.Parameter(torch.tensor(1.0))

    def forward(self, x):
        x = x + self.alpha * self.pe[:, : x.size(1)]
        return self.dropout(x)


def test_compatibility():
    """Regression test for #1121"""
    x = torch.rand(2, 3, 4)

    legacy_net = torch.nn.Sequential(
        LegacyPositionalEncoding(4, 0.0), torch.nn.Linear(4, 2)
    )

    latest_net = torch.nn.Sequential(PositionalEncoding(4, 0.0), torch.nn.Linear(4, 2))

    latest_net.load_state_dict(legacy_net.state_dict())
    legacy = legacy_net(x)
    latest = latest_net(x)
    assert torch.allclose(legacy, latest)

    legacy_net = torch.nn.Sequential(
        LegacyScaledPositionalEncoding(4, 0.0), torch.nn.Linear(4, 2)
    )

    latest_net = torch.nn.Sequential(
        ScaledPositionalEncoding(4, 0.0), torch.nn.Linear(4, 2)
    )

    latest_net.load_state_dict(legacy_net.state_dict())
    legacy = legacy_net(x)
    latest = latest_net(x)
    assert torch.allclose(legacy, latest)