|
import torch
|
|
from torch import nn
|
|
from typing import Optional
|
|
from dataclasses import dataclass
|
|
from transformers import PreTrainedModel
|
|
from .configuration_mlp import MLPConfig
|
|
from transformers.utils import ModelOutput
|
|
from transformers.activations import ACT2FN
|
|
|
|
|
|
@dataclass
|
|
class MLPOutput(ModelOutput):
|
|
loss: Optional[torch.FloatTensor] = None
|
|
logits: Optional[torch.FloatTensor] = None
|
|
|
|
|
|
class MLPPreTrainedModel(PreTrainedModel):
|
|
config_class = MLPConfig
|
|
|
|
def _init_weights(self, module):
|
|
"""Initialize the weights"""
|
|
if isinstance(module, nn.Linear):
|
|
module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
|
|
if module.bias is not None:
|
|
module.bias.data.zero_()
|
|
|
|
|
|
class MLPModel(MLPPreTrainedModel):
|
|
def __init__(self, config):
|
|
super().__init__(config)
|
|
self.act_fn = ACT2FN[config.hidden_act]
|
|
iho = [config.input_size, *config.hidden_size, config.output_size]
|
|
self.linears = nn.ModuleList([
|
|
nn.Linear(iho[i], iho[i+1])
|
|
for i in range(config.num_hidden_layers + 1)
|
|
])
|
|
self.loss_fn = nn.CrossEntropyLoss(reduce="mean")
|
|
|
|
self.post_init()
|
|
|
|
def forward(self, inputs, labels=None):
|
|
for i in range(len(self.linears) - 1):
|
|
inputs = self.act_fn(self.linears[i](inputs))
|
|
logits = self.linears[-1](inputs)
|
|
|
|
loss = None
|
|
if labels is None:
|
|
return ModelOutput(loss=loss, logits=logits)
|
|
else:
|
|
loss = self.loss_fn(logits, labels)
|
|
return ModelOutput(loss=loss, logits=logits) |
