File size: 6,474 Bytes
8481452
 
 
 
 
 
bc2395e
8481452
 
 
 
bc2395e
 
 
 
 
 
 
 
 
 
 
 
 
 
8481452
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
import math
from functools import partial
import os
import json
import torch
import torch.nn as nn
import numpy as np

from mamba_ssm.modules.mamba_simple import Mamba, Block
from huggingface_hub import PyTorchModelHubMixin

# convert to one hot
def seq_to_oh(seq):
    oh = np.zeros((len(seq), 4), dtype=int)
    for i, base in enumerate(seq):
        if base == 'A':
            oh[i, 0] = 1
        elif base == 'C':
            oh[i, 1] = 1
        elif base == 'G':
            oh[i, 2] = 1
        elif base == 'T':
            oh[i, 3] = 1
    return oh

def create_block(
    d_model,
    ssm_cfg=None,
    norm_epsilon=1e-5,
    residual_in_fp32=False,
    fused_add_norm=False,
    layer_idx=None,
    device=None,
    dtype=None,
):
    if ssm_cfg is None:
        ssm_cfg = {}
    factory_kwargs = {"device": device, "dtype": dtype}
    mix_cls = partial(Mamba, layer_idx=layer_idx, **ssm_cfg, **factory_kwargs)
    norm_cls = partial(nn.LayerNorm, eps=norm_epsilon, **factory_kwargs)
    block = Block(
        d_model,
        mix_cls,
        norm_cls=norm_cls,
        fused_add_norm=fused_add_norm,
        residual_in_fp32=residual_in_fp32,
    )
    block.layer_idx = layer_idx
    return block


class MixerModel(
    nn.Module,
    PyTorchModelHubMixin,
):

    def __init__(
        self,
        d_model: int,
        n_layer: int,
        input_dim: int,
        ssm_cfg=None,
        norm_epsilon: float = 1e-5,
        rms_norm: bool = False,
        initializer_cfg=None,
        fused_add_norm=False,
        residual_in_fp32=False,
        device=None,
        dtype=None,
    ) -> None:
        factory_kwargs = {"device": device, "dtype": dtype}
        super().__init__()
        self.residual_in_fp32 = residual_in_fp32

        self.embedding = nn.Linear(input_dim, d_model, **factory_kwargs)

        self.layers = nn.ModuleList(
            [
                create_block(
                    d_model,
                    ssm_cfg=ssm_cfg,
                    norm_epsilon=norm_epsilon,
                    residual_in_fp32=residual_in_fp32,
                    fused_add_norm=fused_add_norm,
                    layer_idx=i,
                    **factory_kwargs,
                )
                for i in range(n_layer)
            ]
        )

        self.norm_f = nn.LayerNorm(d_model, eps=norm_epsilon, **factory_kwargs)

        self.apply(
            partial(
                _init_weights,
                n_layer=n_layer,
                **(initializer_cfg if initializer_cfg is not None else {}),
            )
        )

    def forward(self, x, inference_params=None, channel_last=False):
        if not channel_last:
            x = x.transpose(1, 2)

        hidden_states = self.embedding(x)
        residual = None
        for layer in self.layers:
            hidden_states, residual = layer(
                hidden_states, residual, inference_params=inference_params
            )

        residual = (hidden_states + residual) if residual is not None else hidden_states
        hidden_states = self.norm_f(residual.to(dtype=self.norm_f.weight.dtype))

        hidden_states = hidden_states

        return hidden_states

    def representation(
        self,
        x: torch.Tensor,
        lengths: torch.Tensor,
        channel_last: bool = False,
    ) -> torch.Tensor:
        """Get global representation of input data.

        Args:
            x: Data to embed. Has shape (B x C x L) if not channel_last.
            lengths: Unpadded length of each data input.
            channel_last: Expects input of shape (B x L x C).

        Returns:
            Global representation vector of shape (B x H).
        """
        out = self.forward(x, channel_last=channel_last)

        mean_tensor = mean_unpadded(out, lengths)
        return mean_tensor


def mean_unpadded(x: torch.Tensor, lengths: torch.Tensor) -> torch.Tensor:
    """Take mean of tensor across second dimension without padding.

    Args:
        x: Tensor to take unpadded mean. Has shape (B x L x H).
        lengths: Tensor of unpadded lengths. Has shape (B)

    Returns:
        Mean tensor of shape (B x H).
    """
    mask = torch.arange(x.size(1), device=x.device)[None, :] < lengths[:, None]
    masked_tensor = x * mask.unsqueeze(-1)
    sum_tensor = masked_tensor.sum(dim=1)
    mean_tensor = sum_tensor / lengths.unsqueeze(-1).float()

    return mean_tensor


def _init_weights(
    module,
    n_layer,
    initializer_range=0.02,  # Now only used for embedding layer.
    rescale_prenorm_residual=True,
    n_residuals_per_layer=1,  # Change to 2 if we have MLP
):
    if isinstance(module, nn.Linear):
        if module.bias is not None:
            if not getattr(module.bias, "_no_reinit", False):
                nn.init.zeros_(module.bias)
    elif isinstance(module, nn.Embedding):
        nn.init.normal_(module.weight, std=initializer_range)

    if rescale_prenorm_residual:
        for name, p in module.named_parameters():
            if name in ["out_proj.weight", "fc2.weight"]:
                nn.init.kaiming_uniform_(p, a=math.sqrt(5))
                with torch.no_grad():
                    p /= math.sqrt(n_residuals_per_layer * n_layer)

def load_model(run_path: str, checkpoint_name: str) -> nn.Module:
    """Load trained model located at specified path.

    Args:
        run_path: Path where run data is located.
        checkpoint_name: Name of model checkpoint to load.

    Returns:
        Model with loaded weights.
    """
    model_config_path = os.path.join(run_path, "model_config.json")
    data_config_path = os.path.join(run_path, "data_config.json")

    with open(model_config_path, "r") as f:
        model_params = json.load(f)

    # TODO: Temp backwards compatibility
    if "n_tracks" not in model_params:
        with open(data_config_path, "r") as f:
            data_params = json.load(f)
        n_tracks = data_params["n_tracks"]
    else:
        n_tracks = model_params["n_tracks"]

    model_path = os.path.join(run_path, checkpoint_name)

    model = MixerModel(
        d_model=model_params["ssm_model_dim"],
        n_layer=model_params["ssm_n_layers"],
        input_dim=n_tracks
    )
    checkpoint = torch.load(model_path, map_location=torch.device('cpu'))

    state_dict = {}
    for k, v in checkpoint["state_dict"].items():
        if k.startswith("model"):
            state_dict[k.lstrip("model")[1:]] = v

    model.load_state_dict(state_dict)
    return model