Removing pretrained layers?

[officialpatterson]

❓ Questions & Help

I'm currently trying to use a pretrained BertModel for finetuning but I want to remove some of the layers from the model before fine-tuning.

How do I do this?

[officialpatterson]

If this is important to anyone, I have found a solution:

def deleteEncodingLayers(model, num_layers_to_keep):  # must pass in the full bert model
    oldModuleList = model.bert.encoder.layer
    newModuleList = nn.ModuleList()

    # Now iterate over all layers, only keepign only the relevant layers.
    for i in range(0, len(num_layers_to_keep)):
        newModuleList.append(oldModuleList[i])

    # create a copy of the model, modify it with the new list, and return
    copyOfModel = copy.deepcopy(model)
    copyOfModel.bert.encoder.layer = newModuleList

    return copyOfModel

[ZLKong]

Hi,

Thank you for your question and solution. I also want to try such kind of thing.

I have a question. If I remove some layers, do I need to do pre-train from scratch again?

How does the performance look if you only do finetuning on GLUE or Squad tasks? Does the accuracy go down dramatically?

Thanks, ZLK

[officialpatterson]

@ZLKong no, the remaining layers will remain trained. Not quite sure what you mean by only fine-tuning though.

[ZLKong]

Thank you for your reply!

I want to decrease the FLOPS by simply removing some layers from the model. I want to see if I remove some layers, how much will if effect the accuracy of SQUAD task.

(If the accuracy goes down a lot, that means I might have do the pretraining again?)

Do you have any experiments on this?

Best, ZLK

[officialpatterson]

I haven't, but I'm sure in the original paper they performed a test like that. If not, I guarantee there will be a paper out there that does given how much research has been chucked at bert :)

[ZLKong]

OK, I will look if there are any papers about it. I will run some testings, too. Thank you very much!

[fteufel]

If you're dealing with loading a pretrained model, there is an easier way to remove the top layer:

config = XLNetConfig.from_pretrained(checkpoint)
config.n_layer = 29 #was 30 layers, in my case
model = XLNetModel.from_pretrained(checkpoint, config = config)

This will produce a warning that there are unused weights in the checkpoint and you'll get a model with the top layer removed.

[MisterXY89]

@ZLKong have you found any papers yet?:D

EDIT: I found this paper from March 2021: On the Effect of Dropping Layers of Pre-trained Transformer Models

[Ricco48]

If this is important to anyone, I have found a solution:

def deleteEncodingLayers(model, num_layers_to_keep):  # must pass in the full bert model
    oldModuleList = model.bert.encoder.layer
    newModuleList = nn.ModuleList()

    # Now iterate over all layers, only keepign only the relevant layers.
    for i in range(0, len(num_layers_to_keep)):
        newModuleList.append(oldModuleList[i])

    # create a copy of the model, modify it with the new list, and return
    copyOfModel = copy.deepcopy(model)
    copyOfModel.bert.encoder.layer = newModuleList

    return copyOfModel

Hello there,

I still don't know how to implement this. Does this just need to call the pre-trained model, for example: BERT model from TensorFlow

or

I need the full code BERT model?

thank you

[bambry-teachable]

hi @officialpatterson, thanks for providing the solution! now I'm trying to implement it with the BertModel package which doesn't have the same attributes as yours, anyway I can adapt this code to my model?

Class BERTClass(torch.nn.Module):
    def __init__(self):
        super(BERTClass, self).__init__()
        self.bert_model = BertModel.from_pretrained('bert-base-cased')
        self.dropout = torch.nn.Dropout(0.5)
        self.linear = torch.nn.Linear(768, 9)

    def forward(self, input_ids, attn_mask, token_type_ids):
        output = self.bert_model(
            input_ids, 
            attention_mask=attn_mask, 
            token_type_ids=token_type_ids
        )
        output_dropout = self.dropout(output.pooler_output)
        output = self.linear(output_dropout)
        return output

[alvations]

Not sure if anyone is looking for a way to remove layers for EncoderDecoderModel e.g. for some models with unbalance layers. I've tried this, and it seems to work:

from transformers import EncoderDecoderModel, BertLMHeadModel
from transformers import BertConfig, EncoderDecoderConfig, EncoderDecoderModel

# Initializing a BERT bert-base-uncased style configuration
config_encoder = BertConfig.from_pretrained("bert-base-multilingual-uncased")
config_decoder = BertConfig.from_pretrained("bert-base-multilingual-uncased")

config_encoder.num_hidden_layers = 5
config_decoder.num_hidden_layers = 2

config = EncoderDecoderConfig.from_encoder_decoder_configs(config_encoder, config_decoder)

# Initializing a Bert2Bert model from the bert-base-uncased style configurations
model = EncoderDecoderModel(config=config)

model.decoder # Shows 2 layers, if `num_hidden_layers` was unchanged, it should show 6.

[out]:

BertLMHeadModel(
  (bert): BertModel(
    (embeddings): BertEmbeddings(
      (word_embeddings): Embedding(105879, 768, padding_idx=0)
      (position_embeddings): Embedding(512, 768)
      (token_type_embeddings): Embedding(2, 768)
      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
      (dropout): Dropout(p=0.1, inplace=False)
    )
    (encoder): BertEncoder(
      (layer): ModuleList(
        (0): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
            (intermediate_act_fn): GELUActivation()
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (1): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (crossattention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
            (intermediate_act_fn): GELUActivation()
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
      )
    )
  )
  (cls): BertOnlyMLMHead(
    (predictions): BertLMPredictionHead(
      (transform): BertPredictionHeadTransform(
        (dense): Linear(in_features=768, out_features=768, bias=True)
        (transform_act_fn): GELUActivation()
        (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
      )
      (decoder): Linear(in_features=768, out_features=105879, bias=True)
    )
  )
)

---

Similarly, if it's just an LM encoder model, something like this should work:

from transformers import BertConfig, BertLMHeadModel

config_encoder = BertConfig.from_pretrained("bert-base-multilingual-uncased")
config_encoder.num_hidden_layers = 3
model = BertLMHeadModel(config=config_encoder)

model

[out]:

BertLMHeadModel(
  (bert): BertModel(
    (embeddings): BertEmbeddings(
      (word_embeddings): Embedding(105879, 768, padding_idx=0)
      (position_embeddings): Embedding(512, 768)
      (token_type_embeddings): Embedding(2, 768)
      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
      (dropout): Dropout(p=0.1, inplace=False)
    )
    (encoder): BertEncoder(
      (layer): ModuleList(
        (0): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
            (intermediate_act_fn): GELUActivation()
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (1): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
            (intermediate_act_fn): GELUActivation()
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (2): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
            (intermediate_act_fn): GELUActivation()
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
      )
    )
  )
  (cls): BertOnlyMLMHead(
    (predictions): BertLMPredictionHead(
      (transform): BertPredictionHeadTransform(
        (dense): Linear(in_features=768, out_features=768, bias=True)
        (transform_act_fn): GELUActivation()
        (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
      )
      (decoder): Linear(in_features=768, out_features=105879, bias=True)
    )
  )
)