modeling_gpt_bert.py

import math, torch
import torch.nn as nn
import torch.nn.functional as F
from transformers import PreTrainedModel
from transformers.modeling_outputs import MaskedLMOutput, CausalLMOutputWithCrossAttentions
from .configuration_gpt_bert import GPTBertConfig

class GeGLU(nn.Module):
    def forward(self, x):
        x, gate = x.chunk(2, dim=-1)
        return x * F.gelu(gate, approximate='tanh')

class FeedForward(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.mlp = nn.Sequential(
            nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps, elementwise_affine=False)
            ,nn.Linear(config.hidden_size, 2*config.intermediate_size, bias=False)
            ,GeGLU()
            ,nn.LayerNorm(config.intermediate_size, eps=config.layer_norm_eps, elementwise_affine=False)
            ,nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
            ,nn.Dropout(config.hidden_dropout_prob)
        )
        self._init(config.hidden_size)
    def _init(self, h):
        std = math.sqrt(2.0 / (5.0 * h))
        nn.init.trunc_normal_(self.mlp[1].weight, mean=0.0, std=std, a=-2*std, b=2*std)
        nn.init.trunc_normal_(self.mlp[-2].weight, mean=0.0, std=std, a=-2*std, b=2*std)
    def forward(self, x): return self.mlp(x)

class Attention(nn.Module):
    def __init__(self, config):
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0:
            raise ValueError('hidden not divisible by heads')
        self.nh = config.num_attention_heads
        self.dh = config.hidden_size // config.num_attention_heads
        self.qkv = nn.Linear(config.hidden_size, 3*config.hidden_size, bias=False)
        self.o = nn.Linear(config.hidden_size, config.hidden_size, bias=False)
        self.drop = nn.Dropout(config.hidden_dropout_prob)
    def forward(self, x, attn_mask, rel):
        B,S,H = x.shape
        qkv = self.qkv(x).view(B,S,3,self.nh,self.dh).permute(2,0,3,1,4)
        q,k,v = qkv[0],qkv[1],qkv[2]
        attn = (q @ k.transpose(-1,-2)) / math.sqrt(self.dh)
        if attn_mask is not None: attn = attn.masked_fill(attn_mask[:,None,:, :]==0, float('-inf'))
        attn = torch.softmax(attn, dim=-1)
        attn = self.drop(attn)
        y = attn @ v
        y = y.transpose(1,2).contiguous().view(B,S,H)
        return self.o(y)

class Block(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.attn = Attention(config)
        self.ff = FeedForward(config)
    def forward(self, x, attn_mask, rel):
        x = x + self.attn(x, attn_mask, rel)
        x = x + self.ff(x)
        return x

class Encoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.layers = nn.ModuleList([Block(config) for _ in range(config.num_hidden_layers)])
    def forward(self, x, attn_mask, rel):
        for layer in self.layers:
            x = layer(x, attn_mask, rel)
        return x

class Embedding(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.word_embedding = nn.Embedding(config.vocab_size, config.hidden_size)
        self.pos_embedding = nn.Embedding(config.max_position_embeddings, config.hidden_size)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
    def forward(self, input_ids):
        B,S = input_ids.shape
        pos = torch.arange(0,S, device=input_ids.device).unsqueeze(0).expand(B,S)
        x = self.word_embedding(input_ids) + self.pos_embedding(pos)
        return self.dropout(x), None

class CoreModel(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.embedding = Embedding(config)
        self.transformer = Encoder(config)
        self.layer_norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps, elementwise_affine=False)
        self.head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
        self.head.weight = self.embedding.word_embedding.weight
    def forward(self, input_ids, attention_mask=None):
        x,_ = self.embedding(input_ids)
        if attention_mask is not None: attn = attention_mask.unsqueeze(1)
        else: attn = None
        x = self.transformer(x, attn, None)
        x = self.layer_norm(x)
        return self.head(x)

class GPTBertForMaskedLM(PreTrainedModel):
    config_class = GPTBertConfig
    base_model_prefix = 'gpt_bert'
    def __init__(self, config: GPTBertConfig):
        super().__init__(config)
        self.model = CoreModel(config)
    def forward(self, input_ids, attention_mask=None, labels=None):
        logits = self.model(input_ids, attention_mask)
        loss=None
        if labels is not None:
            loss_fct = nn.CrossEntropyLoss(ignore_index=-100)
            loss = loss_fct(logits.view(-1, logits.size(-1)), labels.view(-1))
        return MaskedLMOutput(loss=loss, logits=logits)

class GPTBertForCausalLM(PreTrainedModel):
    config_class = GPTBertConfig
    base_model_prefix = 'gpt_bert'
    def __init__(self, config: GPTBertConfig):
        super().__init__(config)
        self.model = CoreModel(config)
    def prepare_inputs_for_generation(self, input_ids, **kwargs):
        return {'input_ids': input_ids}
    def forward(self, input_ids, attention_mask=None, labels=None):
        logits = self.model(input_ids, attention_mask)
        loss=None
        if labels is not None:
            shift_logits = logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            loss_fct = nn.CrossEntropyLoss(ignore_index=-100)
            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
        return CausalLMOutputWithCrossAttentions(loss=loss, logits=logits)