nanoBERT, inspired by @karpathy's nanoGPT

nanoBERT.py

import torch.nn as nn
import torch.nn.functional as F
import math
from typing import Optional, Tuple

class BertSelfAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        
        if config.hidden_size % config.num_attention_heads != 0:
            raise ValueError(
                f"The hidden size ({config.hidden_size}) is not a multiple of the number of attention "
                f"heads ({config.num_attention_heads})"
            )

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        self.key = nn.Linear(config.hidden_size, self.all_head_size)
        self.value = nn.Linear(config.hidden_size, self.all_head_size)

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
        self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention')
        if not self.flash:
            print("WARNING: using slow attention. Flash Attention requires PyTorch >= 2.0")

    def transpose_for_scores(self, x: torch.Tensor) -> torch.Tensor:
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(self, hidden_states: torch.Tensor, attention_mask: Optional[torch.FloatTensor] = None) -> Tuple[torch.Tensor]:
        
        # Compute query, key, value tensors
        query_layer = self.transpose_for_scores(self.query(hidden_states))
        key_layer = self.transpose_for_scores(self.key(hidden_states))
        value_layer = self.transpose_for_scores(self.value(hidden_states))

        # Use flash implementation if config.flash is set
        if self.flash:
            context_layer = torch.nn.functional.scaled_dot_product_attention(
                query_layer, 
                key_layer, 
                value_layer, 
                attn_mask=attention_mask, 
                dropout_p=self.dropout.p if self.training else 0,
                is_causal=False
            )
        else:
            # Compute attention scores
            attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
            attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        
            if attention_mask is not None:
                attention_scores += attention_mask
        
            # Convert attention scores to probabilities
            attention_probs = nn.functional.softmax(attention_scores, dim=-1)
            attention_probs = self.dropout(attention_probs)

            # Compute context layer
            context_layer = torch.matmul(attention_probs, value_layer)
        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        context_layer = context_layer.view(context_layer.size()[:-2] + (self.all_head_size,))

        # Return accordingly
        outputs = context_layer
        return outputs

  
class BertSelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states

class BertAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.self = BertSelfAttention(config)
        self.output = BertSelfOutput(config)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
    ) -> Tuple[torch.Tensor]:
        
        attention_outputs = self.self(hidden_states=hidden_states, attention_mask=attention_mask)
        attention_output = self.output(input_tensor=hidden_states, hidden_states=attention_outputs)
        
        outputs = attention_output
        
        return outputs



class BertIntermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        self.intermediate_act_fn = nn.GELU()

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states

class BertOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states: torch.Tensor, input_tensor: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states

class BertLayer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.attention = BertAttention(config)
        self.intermediate = BertIntermediate(config)
        self.output = BertOutput(config)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.FloatTensor] = None,
    ) -> Tuple[torch.Tensor]:
        
        self_attention_outputs = self.attention(
            hidden_states=hidden_states,
            attention_mask=attention_mask
        )
        attention_output = self_attention_outputs
        
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)

        return layer_output


class BertEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.layer = nn.ModuleList([BertLayer(config) for _ in range(config.num_hidden_layers)])

    def forward(self, hidden_states, attention_mask=None):
        for layer_module in self.layer:
            hidden_states = layer_module(hidden_states=hidden_states, attention_mask=attention_mask)

        return hidden_states

class BertEmbeddings(nn.Module):
    """Construct the embeddings from word, position and token_type embeddings."""

    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
        self.position_embeddings = nn.Embedding(config.max_position_embeddings, config.hidden_size)
        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)
        
        self.LayerNorm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)
        

    def forward(self, input_ids=None, token_type_ids=None) -> torch.Tensor:
        b, t = input_ids.size()
        device = input_ids.device
        # Use the position_ids created in the constructor if none are provided
        position_ids = torch.arange(0, t, dtype=torch.long, device=device)

        # Use the buffered token_type_ids if none are provided
        if token_type_ids is None:
            token_type_ids = torch.zeros(self.position_ids.size(), dtype=torch.long).expand(b, t)

        # If no input embeddings are provided, transform input_ids into embeddings

        inputs_embeds = self.word_embeddings(input_ids)

        # Create token type embeddings
        token_type_embeddings = self.token_type_embeddings(token_type_ids)

        # Sum word embeddings, position embeddings, and token type embeddings
        embeddings = inputs_embeds + token_type_embeddings
        embeddings += self.position_embeddings(position_ids)

        # Normalize and apply dropout
        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        
        return embeddings

class BertPooler(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.activation = nn.Tanh()

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.
        first_token_tensor = hidden_states[:, 0]
        pooled_output = self.dense(first_token_tensor)
        pooled_output = self.activation(pooled_output)
        return pooled_output

class BertConfig:
    def __init__(self):
        self.hidden_size = 384
        self.num_hidden_layers = 12
        self.hidden_dropout_prob = 0.1
        self.attention_probs_dropout_prob = 0.1
        self.layer_norm_eps = 1e-12
        self.intermediate_size = 1536
        self.vocab_size = 30522
        self.max_position_embeddings = 512
        self.type_vocab_size = 2
        self.pad_token_id = 0
        self.num_attention_heads = 12

class BertModel(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.embeddings = BertEmbeddings(config)
        self.encoder = BertEncoder(config)
        self.pooler = BertPooler(config)
    
    def get_extended_attention_mask(self, attention_mask, input_shape):
        # Convert attention mask to binary:
        extended_attention_mask = attention_mask[:, None, None, :]
        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0
        return extended_attention_mask

    def forward(self,
                input_ids=None,
                attention_mask=None,
                token_type_ids=None):
        
        if input_ids is not None:
            input_shape = input_ids.size()
        else:
            raise ValueError("You have to specify either input_ids")

        if attention_mask is None:
            attention_mask = torch.ones(input_shape, dtype=torch.long, device=input_ids.device)
        
        if token_type_ids is None:
            token_type_ids = torch.zeros(input_shape, dtype=torch.long, device=input_ids.device)

        extended_attention_mask = self.get_extended_attention_mask(attention_mask, input_shape)

        embedding_output = self.embeddings(input_ids, token_type_ids)
        encoder_outputs = self.encoder(embedding_output, extended_attention_mask)

        sequence_output = encoder_outputs
        pooled_output = self.pooler(sequence_output)
        
        return sequence_output, pooled_output

config = BertConfig()
model = BertModel(config)

script to load weights

from transformers import BertModel, BertTokenizerFast
print("loading weights from pretrained gpt: %s" % model_type)
model_hf = BertModel.from_pretrained(model_type)
sd_hf = model_hf.state_dict()
sd_keys_hf = sd_hf.keys()

sd = model.state_dict()
sd_keys = sd.keys()

# copy while ensuring all of the parameters are aligned and match in names and shapes
sd_keys_hf = sd_hf.keys()
# basically the openai checkpoints use a "Conv1D" module, but we only want to use a vanilla Linear
# this means that we have to transpose these weights when we import them
assert len(sd_keys_hf) == len(sd_keys), f"mismatched keys: {len(sd_keys_hf)} != {len(sd_keys)} "
for k in sd_keys_hf:
    assert sd_hf[k].shape == sd[k].shape
    with torch.no_grad():
        sd[k].copy_(sd_hf[k])```

dont forget to use model.eval() to disable dropout

awesome!

Loading script below also works fine for me.
Thanks for the awesome work!

from transformers import BertModel as HFBertModel

hf_model_name = "roberta-base"
hf_model = HFBertModel.from_pretrained(hf_model_name)
hf_state_dict = hf_model.state_dict()

nanoconfig = hf_model.config
nanobert = BertModel(nanoconfig)
nanobert.load_state_dict(hf_state_dict)