{"id":177202,"date":"2025-12-01T13:28:32","date_gmt":"2025-12-01T13:28:32","guid":{"rendered":"https:\/\/ktromedia.com\/?p=177202"},"modified":"2025-12-01T13:28:32","modified_gmt":"2025-12-01T13:28:32","slug":"fine-tuning-a-bert-model-machinelearningmastery-com","status":"publish","type":"post","link":"http:\/\/ktromedia.com\/?p=177202","title":{"rendered":"Fine-Tuning a BERT Model – MachineLearningMastery.com"},"content":{"rendered":"
\n

import <\/span>collections<\/span><\/p>\n

import <\/span>dataclasses<\/span><\/p>\n

import <\/span>functools<\/span><\/p>\n

\u00a0<\/p>\n

import <\/span>torch<\/span><\/p>\n

import <\/span>torch<\/span>.<\/span>nn <\/span>as<\/span> <\/span>nn<\/span><\/p>\n

import <\/span>torch<\/span>.<\/span>optim <\/span>as<\/span> <\/span>optim<\/span><\/p>\n

import <\/span>tqdm<\/span><\/p>\n

from <\/span>datasets <\/span>import <\/span>load_dataset<\/span><\/p>\n

from <\/span>tokenizers <\/span>import <\/span>Tokenizer<\/span><\/p>\n

from <\/span>torch <\/span>import <\/span>Tensor<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0<\/p>\n

# BERT config and model defined previously<\/span><\/p>\n

@<\/span>dataclasses<\/span>.<\/span>dataclass<\/span><\/p>\n

class<\/span> <\/span>BertConfig<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>“”<\/span>“Configuration for BERT model.”<\/span>“”<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>vocab_size<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>30522<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>num_layers<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>12<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>hidden_size<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>768<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>num_heads<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>12<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>dropout_prob<\/span>:<\/span> <\/span>float<\/span> <\/span>=<\/span> <\/span>0.1<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>pad_id<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>0<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>max_seq_len<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>512<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>num_types<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>2<\/span><\/p>\n

\u00a0<\/p>\n

class<\/span> <\/span>BertBlock<\/span>(<\/span>nn<\/span>.<\/span>Module<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>“”<\/span>“One transformer block in BERT.”<\/span>“”<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>__init__<\/span>(<\/span>self<\/span>,<\/span> <\/span>hidden_size<\/span>:<\/span> <\/span>int<\/span>,<\/span> <\/span>num_heads<\/span>:<\/span> <\/span>int<\/span>,<\/span> <\/span>dropout_prob<\/span>:<\/span> <\/span>float<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>super<\/span>(<\/span>)<\/span>.<\/span>__init__<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>attention<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>MultiheadAttention<\/span>(<\/span>hidden_size<\/span>,<\/span> <\/span>num_heads<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 <\/span>dropout<\/span>=<\/span>dropout_prob<\/span>,<\/span> <\/span>batch_first<\/span>=<\/span>True<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>attn_norm<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>LayerNorm<\/span>(<\/span>hidden_size<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>ff_norm<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>LayerNorm<\/span>(<\/span>hidden_size<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>dropout<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Dropout<\/span>(<\/span>dropout_prob<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>feed_forward<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Sequential<\/span>(<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>nn<\/span>.<\/span>Linear<\/span>(<\/span>hidden_size<\/span>,<\/span> <\/span>4<\/span> <\/span>*<\/span> <\/span>hidden_size<\/span>)<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>nn<\/span>.<\/span>GELU<\/span>(<\/span>)<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>nn<\/span>.<\/span>Linear<\/span>(<\/span>4<\/span> <\/span>*<\/span> <\/span>hidden_size<\/span>,<\/span> <\/span>hidden_size<\/span>)<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>forward<\/span>(<\/span>self<\/span>,<\/span> <\/span>x<\/span>:<\/span> <\/span>Tensor<\/span>,<\/span> <\/span>pad_mask<\/span>:<\/span> <\/span>Tensor<\/span>)<\/span> <\/span>-><\/span> <\/span>Tensor<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># self-attention with padding mask and post-norm<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>attn_output<\/span>,<\/span> <\/span>_<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>attention<\/span>(<\/span>x<\/span>,<\/span> <\/span>x<\/span>,<\/span> <\/span>x<\/span>,<\/span> <\/span>key_padding_mask<\/span>=<\/span>pad_mask<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>attn_norm<\/span>(<\/span>x<\/span> <\/span>+<\/span> <\/span>attn_output<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># feed-forward with GeLU activation and post-norm<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>ff_output<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>feed_forward<\/span>(<\/span>x<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>ff_norm<\/span>(<\/span>x<\/span> <\/span>+<\/span> <\/span>self<\/span>.<\/span>dropout<\/span>(<\/span>ff_output<\/span>)<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>return<\/span> <\/span>x<\/span><\/p>\n

\u00a0<\/p>\n

class<\/span> <\/span>BertPooler<\/span>(<\/span>nn<\/span>.<\/span>Module<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>“”<\/span>“Pooler layer for BERT to process the [CLS] token output.”<\/span>“”<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>__init__<\/span>(<\/span>self<\/span>,<\/span> <\/span>hidden_size<\/span>:<\/span> <\/span>int<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>super<\/span>(<\/span>)<\/span>.<\/span>__init__<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>dense<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Linear<\/span>(<\/span>hidden_size<\/span>,<\/span> <\/span>hidden_size<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>activation<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Tanh<\/span>(<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>forward<\/span>(<\/span>self<\/span>,<\/span> <\/span>x<\/span>:<\/span> <\/span>Tensor<\/span>)<\/span> <\/span>-><\/span> <\/span>Tensor<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>dense<\/span>(<\/span>x<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>activation<\/span>(<\/span>x<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>return<\/span> <\/span>x<\/span><\/p>\n

\u00a0<\/p>\n

class<\/span> <\/span>BertModel<\/span>(<\/span>nn<\/span>.<\/span>Module<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>“”<\/span>“Backbone of BERT model.”<\/span>“”<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>__init__<\/span>(<\/span>self<\/span>,<\/span> <\/span>config<\/span>:<\/span> <\/span>BertConfig<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>super<\/span>(<\/span>)<\/span>.<\/span>__init__<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># embedding layers<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>word_embeddings<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Embedding<\/span>(<\/span>config<\/span>.<\/span>vocab_size<\/span>,<\/span> <\/span>config<\/span>.<\/span>hidden_size<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>padding_idx<\/span>=<\/span>config<\/span>.<\/span>pad_id<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>type_embeddings<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Embedding<\/span>(<\/span>config<\/span>.<\/span>num_types<\/span>,<\/span> <\/span>config<\/span>.<\/span>hidden_size<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>position_embeddings<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Embedding<\/span>(<\/span>config<\/span>.<\/span>max_seq_len<\/span>,<\/span> <\/span>config<\/span>.<\/span>hidden_size<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>embeddings_norm<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>LayerNorm<\/span>(<\/span>config<\/span>.<\/span>hidden_size<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>embeddings_dropout<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Dropout<\/span>(<\/span>config<\/span>.<\/span>dropout_prob<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># transformer blocks<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>blocks<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>ModuleList<\/span>(<\/span>[<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>BertBlock<\/span>(<\/span>config<\/span>.<\/span>hidden_size<\/span>,<\/span> <\/span>config<\/span>.<\/span>num_heads<\/span>,<\/span> <\/span>config<\/span>.<\/span>dropout_prob<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>for<\/span> <\/span>_<\/span> <\/span>in<\/span> <\/span>range<\/span>(<\/span>config<\/span>.<\/span>num_layers<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>]<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># [CLS] pooler layer<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>pooler<\/span> <\/span>=<\/span> <\/span>BertPooler<\/span>(<\/span>config<\/span>.<\/span>hidden_size<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>forward<\/span>(<\/span>self<\/span>,<\/span> <\/span>input_ids<\/span>:<\/span> <\/span>Tensor<\/span>,<\/span> <\/span>token_type_ids<\/span>:<\/span> <\/span>Tensor<\/span>,<\/span> <\/span>pad_id<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>0<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>)<\/span> <\/span>-><\/span> <\/span>tuple<\/span>[<\/span>Tensor<\/span>,<\/span> <\/span>Tensor<\/span>]<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># create attention mask for padding tokens<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>pad_mask<\/span> <\/span>=<\/span> <\/span>input_ids<\/span> <\/span>==<\/span> <\/span>pad<\/span>_<\/span>id<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># convert integer tokens to embedding vectors<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>batch_size<\/span>,<\/span> <\/span>seq_len<\/span> <\/span>=<\/span> <\/span>input_ids<\/span>.<\/span>shape<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>position_ids<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>arange<\/span>(<\/span>seq_len<\/span>,<\/span> <\/span>device<\/span>=<\/span>input_ids<\/span>.<\/span>device<\/span>)<\/span>.<\/span>unsqueeze<\/span>(<\/span>0<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>position_embeddings<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>position_embeddings<\/span>(<\/span>position_ids<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>type_embeddings<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>type_embeddings<\/span>(<\/span>token_type_ids<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_embeddings<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>word_embeddings<\/span>(<\/span>input_ids<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>token_embeddings<\/span> <\/span>+<\/span> <\/span>type_embeddings<\/span> <\/span>+<\/span> <\/span>position<\/span>_<\/span>embeddings<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>embeddings_norm<\/span>(<\/span>x<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>embeddings_dropout<\/span>(<\/span>x<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># process the sequence with transformer blocks<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>for<\/span> <\/span>block <\/span>in<\/span> <\/span>self<\/span>.<\/span>blocks<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>x<\/span> <\/span>=<\/span> <\/span>block<\/span>(<\/span>x<\/span>,<\/span> <\/span>pad_mask<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># pool the hidden state of the `[CLS]` token<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>pooled_output<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>pooler<\/span>(<\/span>x<\/span>[<\/span>:<\/span>,<\/span> <\/span>0<\/span>,<\/span> <\/span>:<\/span>]<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>return<\/span> <\/span>x<\/span>,<\/span> <\/span>pooled<\/span>_<\/span>output<\/p>\n

\u00a0<\/p>\n

# Define new BERT model for question answering<\/span><\/p>\n

class<\/span> <\/span>BertForQuestionAnswering<\/span>(<\/span>nn<\/span>.<\/span>Module<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>“”<\/span>“BERT model for SQuAD question answering.”<\/span>“”<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>__init__<\/span>(<\/span>self<\/span>,<\/span> <\/span>config<\/span>:<\/span> <\/span>BertConfig<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>super<\/span>(<\/span>)<\/span>.<\/span>__init__<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>bert<\/span> <\/span>=<\/span> <\/span>BertModel<\/span>(<\/span>config<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># Two outputs: start and end position logits<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>self<\/span>.<\/span>qa_outputs<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>Linear<\/span>(<\/span>config<\/span>.<\/span>hidden_size<\/span>,<\/span> <\/span>2<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>def <\/span>forward<\/span>(<\/span>self<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span>:<\/span> <\/span>Tensor<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids<\/span>:<\/span> <\/span>Tensor<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>pad_id<\/span>:<\/span> <\/span>int<\/span> <\/span>=<\/span> <\/span>0<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>)<\/span> <\/span>-><\/span> <\/span>tuple<\/span>[<\/span>Tensor<\/span>,<\/span> <\/span>Tensor<\/span>]<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># Get sequence output from BERT (batch_size, seq_len, hidden_size)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>seq_output<\/span>,<\/span> <\/span>pooled_output<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>bert<\/span>(<\/span>input_ids<\/span>,<\/span> <\/span>token_type_ids<\/span>,<\/span> <\/span>pad_id<\/span>=<\/span>pad_id<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># Project to start and end logits<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>logits<\/span> <\/span>=<\/span> <\/span>self<\/span>.<\/span>qa_outputs<\/span>(<\/span>seq_output<\/span>)<\/span>\u00a0\u00a0<\/span># (batch_size, seq_len, 2)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_logits<\/span> <\/span>=<\/span> <\/span>logits<\/span>[<\/span>:<\/span>,<\/span> <\/span>:<\/span>,<\/span> <\/span>0<\/span>]<\/span>\u00a0\u00a0<\/span># (batch_size, seq_len)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>end_logits<\/span> <\/span>=<\/span> <\/span>logits<\/span>[<\/span>:<\/span>,<\/span> <\/span>:<\/span>,<\/span> <\/span>1<\/span>]<\/span>\u00a0\u00a0\u00a0\u00a0<\/span># (batch_size, seq_len)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>return<\/span> <\/span>start_logits<\/span>,<\/span> <\/span>end<\/span>_<\/span>logits<\/p>\n

\u00a0<\/p>\n

# Load SQuAD dataset for question answering<\/span><\/p>\n

dataset<\/span> <\/span>=<\/span> <\/span>load_dataset<\/span>(<\/span>“squad”<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

# Load the pretrained BERT tokenizer<\/span><\/p>\n

TOKENIZER_PATH<\/span> <\/span>=<\/span> <\/span>“wikitext-2_wordpiece.json”<\/span><\/p>\n

tokenizer<\/span> <\/span>=<\/span> <\/span>Tokenizer<\/span>.<\/span>from_file<\/span>(<\/span>TOKENIZER_PATH<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

# Setup collate function to tokenize question-context pairs for the model<\/span><\/p>\n

def <\/span>collate<\/span>(<\/span>batch<\/span>:<\/span> <\/span>list<\/span>[<\/span>dict<\/span>]<\/span>,<\/span> <\/span>tokenizer<\/span>:<\/span> <\/span>Tokenizer<\/span>,<\/span> <\/span>max_len<\/span>:<\/span> <\/span>int<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>)<\/span> <\/span>-><\/span> <\/span>tuple<\/span>[<\/span>Tensor<\/span>,<\/span> <\/span>Tensor<\/span>,<\/span> <\/span>Tensor<\/span>,<\/span> <\/span>Tensor<\/span>]<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>“”<\/span>“Collate question-context pairs for the model.”<\/span>“”<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>cls_id<\/span> <\/span>=<\/span> <\/span>tokenizer<\/span>.<\/span>token_to_id<\/span>(<\/span>“[CLS]”<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>sep_id<\/span> <\/span>=<\/span> <\/span>tokenizer<\/span>.<\/span>token_to_id<\/span>(<\/span>“[SEP]”<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>pad_id<\/span> <\/span>=<\/span> <\/span>tokenizer<\/span>.<\/span>token_to_id<\/span>(<\/span>“[PAD]”<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>input_ids_list<\/span> <\/span>=<\/span> <\/span>[<\/span>]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids_list<\/span> <\/span>=<\/span> <\/span>[<\/span>]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>start_positions<\/span> <\/span>=<\/span> <\/span>[<\/span>]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>end_positions<\/span> <\/span>=<\/span> <\/span>[<\/span>]<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>for<\/span> <\/span>item <\/span>in<\/span> <\/span>batch<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># Tokenize question and context<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>question<\/span>,<\/span> <\/span>context<\/span> <\/span>=<\/span> <\/span>item<\/span>[<\/span>“question”<\/span>]<\/span>,<\/span> <\/span>item<\/span>[<\/span>“context”<\/span>]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>question_ids<\/span> <\/span>=<\/span> <\/span>tokenizer<\/span>.<\/span>encode<\/span>(<\/span>question<\/span>)<\/span>.<\/span>ids<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>context_ids<\/span> <\/span>=<\/span> <\/span>tokenizer<\/span>.<\/span>encode<\/span>(<\/span>context<\/span>)<\/span>.<\/span>ids<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># Build input: [CLS] question [SEP] context [SEP]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span> <\/span>=<\/span> <\/span>[<\/span>cls_id<\/span>,<\/span> <\/span>*<\/span>question_ids<\/span>,<\/span> <\/span>sep_id<\/span>,<\/span> <\/span>*<\/span>context_ids<\/span>,<\/span> <\/span>sep_id<\/span>]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids<\/span> <\/span>=<\/span> <\/span>[<\/span>0<\/span>]<\/span> <\/span>*<\/span> <\/span>(<\/span>len<\/span>(<\/span>question_ids<\/span>)<\/span>+<\/span>2<\/span>)<\/span> <\/span>+<\/span> <\/span>[<\/span>1<\/span>]<\/span> <\/span>*<\/span> <\/span>(<\/span>len<\/span>(<\/span>context_ids<\/span>)<\/span>+<\/span>1<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># Truncate or pad to max length<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>if<\/span> <\/span>len<\/span>(<\/span>input_ids<\/span>)<\/span> <\/span>><\/span> <\/span>max_len<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span> <\/span>=<\/span> <\/span>input_ids<\/span>[<\/span>:<\/span>max_len<\/span>]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids<\/span> <\/span>=<\/span> <\/span>token_type_ids<\/span>[<\/span>:<\/span>max_len<\/span>]<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>else<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span>.<\/span>extend<\/span>(<\/span>[<\/span>pad_id<\/span>]<\/span> <\/span>*<\/span> <\/span>(<\/span>max_len<\/span> <\/span>–<\/span> <\/span>len<\/span>(<\/span>input_ids<\/span>)<\/span>)<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids<\/span>.<\/span>extend<\/span>(<\/span>[<\/span>1<\/span>]<\/span> <\/span>*<\/span> <\/span>(<\/span>max_len<\/span> <\/span>–<\/span> <\/span>len<\/span>(<\/span>token_type_ids<\/span>)<\/span>)<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># Find answer position in tokens: Answer may not be in the context<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_pos<\/span> <\/span>=<\/span> <\/span>end_pos<\/span> <\/span>=<\/span> <\/span>0<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>if<\/span> <\/span>len<\/span>(<\/span>item<\/span>[<\/span>“answers”<\/span>]<\/span>[<\/span>“text”<\/span>]<\/span>)<\/span> <\/span>><\/span> <\/span>0<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>answers<\/span> <\/span>=<\/span> <\/span>tokenizer<\/span>.<\/span>encode<\/span>(<\/span>item<\/span>[<\/span>“answers”<\/span>]<\/span>[<\/span>“text”<\/span>]<\/span>[<\/span>0<\/span>]<\/span>)<\/span>.<\/span>ids<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># find the context offset of the answer in context_ids<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>for<\/span> <\/span>i<\/span> <\/span>in<\/span> <\/span>range<\/span>(<\/span>len<\/span>(<\/span>context_ids<\/span>)<\/span> <\/span>–<\/span> <\/span>len<\/span>(<\/span>answers<\/span>)<\/span> <\/span>+<\/span> <\/span>1<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>if<\/span> <\/span>context_ids<\/span>[<\/span>i<\/span>:<\/span>i<\/span>+<\/span>len<\/span>(<\/span>answers<\/span>)<\/span>]<\/span> <\/span>==<\/span> <\/span>answers<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_pos<\/span> <\/span>=<\/span> <\/span>i<\/span> <\/span>+<\/span> <\/span>len<\/span>(<\/span>question_ids<\/span>)<\/span> <\/span>+<\/span> <\/span>2<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>end_pos<\/span> <\/span>=<\/span> <\/span>start_pos<\/span> <\/span>+<\/span> <\/span>len<\/span>(<\/span>answers<\/span>)<\/span> <\/span>–<\/span> <\/span>1<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>break<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>if<\/span> <\/span>end_pos<\/span> <\/span>>=<\/span> <\/span>max_len<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_pos<\/span> <\/span>=<\/span> <\/span>end_pos<\/span> <\/span>=<\/span> <\/span>0<\/span>\u00a0\u00a0<\/span># answer is clipped, hence no answer<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids_list<\/span>.<\/span>append<\/span>(<\/span>input_ids<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids_list<\/span>.<\/span>append<\/span>(<\/span>token_type_ids<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_positions<\/span>.<\/span>append<\/span>(<\/span>start_pos<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>end_positions<\/span>.<\/span>append<\/span>(<\/span>end_pos<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>input_ids_list<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>tensor<\/span>(<\/span>input_ids_list<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids_list<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>tensor<\/span>(<\/span>token_type_ids_list<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>start_positions<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>tensor<\/span>(<\/span>start_positions<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>end_positions<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>tensor<\/span>(<\/span>end_positions<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>return<\/span> <\/span>(<\/span>input_ids_list<\/span>,<\/span> <\/span>token_type_ids_list<\/span>,<\/span> <\/span>start_positions<\/span>,<\/span> <\/span>end_positions<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

batch_size<\/span> <\/span>=<\/span> <\/span>16<\/span><\/p>\n

max_len<\/span> <\/span>=<\/span> <\/span>384<\/span>\u00a0\u00a0<\/span># Longer for Q&A to accommodate context<\/span><\/p>\n

collate_fn<\/span> <\/span>=<\/span> <\/span>functools<\/span>.<\/span>partial<\/span>(<\/span>collate<\/span>,<\/span> <\/span>tokenizer<\/span>=<\/span>tokenizer<\/span>,<\/span> <\/span>max_len<\/span>=<\/span>max_len<\/span>)<\/span><\/p>\n

train_loader<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>utils<\/span>.<\/span>data<\/span>.<\/span>DataLoader<\/span>(<\/span>dataset<\/span>[<\/span>“train”<\/span>]<\/span>,<\/span> <\/span>batch_size<\/span>=<\/span>batch_size<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 <\/span>shuffle<\/span>=<\/span>True<\/span>,<\/span> <\/span>collate_fn<\/span>=<\/span>collate_fn<\/span>)<\/span><\/p>\n

val_loader<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>utils<\/span>.<\/span>data<\/span>.<\/span>DataLoader<\/span>(<\/span>dataset<\/span>[<\/span>“validation”<\/span>]<\/span>,<\/span> <\/span>batch_size<\/span>=<\/span>batch_size<\/span>,<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 <\/span>shuffle<\/span>=<\/span>False<\/span>,<\/span> <\/span>collate_fn<\/span>=<\/span>collate_fn<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

# Create Q&A model with a pretrained foundation BERT model<\/span><\/p>\n

device<\/span> <\/span>=<\/span> <\/span>torch<\/span>.<\/span>device<\/span>(<\/span>“cuda”<\/span> <\/span>if<\/span> <\/span>torch<\/span>.<\/span>cuda<\/span>.<\/span>is_available<\/span>(<\/span>)<\/span> <\/span>else<\/span> <\/span>“cpu”<\/span>)<\/span><\/p>\n

config<\/span> <\/span>=<\/span> <\/span>BertConfig<\/span>(<\/span>)<\/span><\/p>\n

model<\/span> <\/span>=<\/span> <\/span>BertForQuestionAnswering<\/span>(<\/span>config<\/span>)<\/span><\/p>\n

model<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

model<\/span>.<\/span>bert<\/span>.<\/span>load_state_dict<\/span>(<\/span>torch<\/span>.<\/span>load<\/span>(<\/span>“bert_model.pth”<\/span>,<\/span> <\/span>map_location<\/span>=<\/span>device<\/span>)<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

# Training setup<\/span><\/p>\n

loss_fn<\/span> <\/span>=<\/span> <\/span>nn<\/span>.<\/span>CrossEntropyLoss<\/span>(<\/span>)<\/span><\/p>\n

optimizer<\/span> <\/span>=<\/span> <\/span>optim<\/span>.<\/span>AdamW<\/span>(<\/span>model<\/span>.<\/span>parameters<\/span>(<\/span>)<\/span>,<\/span> <\/span>lr<\/span>=<\/span>2e<\/span>–<\/span>5<\/span>)<\/span><\/p>\n

num_epochs<\/span> <\/span>=<\/span> <\/span>3<\/span><\/p>\n

\u00a0<\/p>\n

for<\/span> <\/span>epoch <\/span>in<\/span> <\/span>range<\/span>(<\/span>num_epochs<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>model<\/span>.<\/span>train<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span># Training<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>with <\/span>tqdm<\/span>.<\/span>tqdm<\/span>(<\/span>train_loader<\/span>,<\/span> <\/span>desc<\/span>=<\/span>f<\/span>“Epoch {epoch+1}\/{num_epochs}”<\/span>)<\/span> <\/span>as<\/span> <\/span>pbar<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>for<\/span> <\/span>batch <\/span>in<\/span> <\/span>pbar<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># get batched data<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span>,<\/span> <\/span>token_type_ids<\/span>,<\/span> <\/span>start_positions<\/span>,<\/span> <\/span>end_positions<\/span> <\/span>=<\/span> <\/span>batch<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span> <\/span>=<\/span> <\/span>input_ids<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids<\/span> <\/span>=<\/span> <\/span>token_type_ids<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_positions<\/span> <\/span>=<\/span> <\/span>start_positions<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>end_positions<\/span> <\/span>=<\/span> <\/span>end_positions<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># forward pass<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_logits<\/span>,<\/span> <\/span>end_logits<\/span> <\/span>=<\/span> <\/span>model<\/span>(<\/span>input_ids<\/span>,<\/span> <\/span>token_type_ids<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># backward pass<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>optimizer<\/span>.<\/span>zero_grad<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_loss<\/span> <\/span>=<\/span> <\/span>loss_fn<\/span>(<\/span>start_logits<\/span>,<\/span> <\/span>start_positions<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>end_loss<\/span> <\/span>=<\/span> <\/span>loss_fn<\/span>(<\/span>end_logits<\/span>,<\/span> <\/span>end_positions<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>loss<\/span> <\/span>=<\/span> <\/span>start_loss<\/span> <\/span>+<\/span> <\/span>end_loss<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>loss<\/span>.<\/span>backward<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>optimizer<\/span>.<\/span>step<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># update progress bar<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>pbar<\/span>.<\/span>set_postfix<\/span>(<\/span>loss<\/span>=<\/span>float<\/span>(<\/span>loss<\/span>)<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>pbar<\/span>.<\/span>update<\/span>(<\/span>1<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span># Validation: Keep track of the average loss and accuracy<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>model<\/span>.<\/span>eval<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>val_loss<\/span>,<\/span> <\/span>num_matches<\/span>,<\/span> <\/span>num_batches<\/span>,<\/span> <\/span>num_samples<\/span> <\/span>=<\/span> <\/span>0<\/span>,<\/span> <\/span>0<\/span>,<\/span> <\/span>0<\/span>,<\/span> <\/span>0<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>with <\/span>torch<\/span>.<\/span>no_grad<\/span>(<\/span>)<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>for<\/span> <\/span>batch <\/span>in<\/span> <\/span>val_loader<\/span>:<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># get batched data<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span>,<\/span> <\/span>token_type_ids<\/span>,<\/span> <\/span>start_positions<\/span>,<\/span> <\/span>end_positions<\/span> <\/span>=<\/span> <\/span>batch<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>input_ids<\/span> <\/span>=<\/span> <\/span>input_ids<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>token_type_ids<\/span> <\/span>=<\/span> <\/span>token_type_ids<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_positions<\/span> <\/span>=<\/span> <\/span>start_positions<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>end_positions<\/span> <\/span>=<\/span> <\/span>end_positions<\/span>.<\/span>to<\/span>(<\/span>device<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># forward pass on validation data<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_logits<\/span>,<\/span> <\/span>end_logits<\/span> <\/span>=<\/span> <\/span>model<\/span>(<\/span>input_ids<\/span>,<\/span> <\/span>token_type_ids<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># compute loss<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>start_loss<\/span> <\/span>=<\/span> <\/span>loss_fn<\/span>(<\/span>start_logits<\/span>,<\/span> <\/span>start_positions<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>end_loss<\/span> <\/span>=<\/span> <\/span>loss_fn<\/span>(<\/span>end_logits<\/span>,<\/span> <\/span>end_positions<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>loss<\/span> <\/span>=<\/span> <\/span>start_loss<\/span> <\/span>+<\/span> <\/span>end_loss<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>val_loss<\/span> <\/span>+=<\/span> <\/span>loss<\/span>.<\/span>item<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>num_batches<\/span> <\/span>+=<\/span> <\/span>1<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span># compute accuracy<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>pred_start<\/span> <\/span>=<\/span> <\/span>start_logits<\/span>.<\/span>argmax<\/span>(<\/span>dim<\/span>=<\/span>–<\/span>1<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>pred_end<\/span> <\/span>=<\/span> <\/span>end_logits<\/span>.<\/span>argmax<\/span>(<\/span>dim<\/span>=<\/span>–<\/span>1<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>match<\/span> <\/span>=<\/span> <\/span>(<\/span>pred_start<\/span> <\/span>==<\/span> <\/span>start_positions<\/span>)<\/span> <\/span>&<\/span> <\/span>(<\/span>pred_end<\/span> <\/span>==<\/span> <\/span>end_positions<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>num_matches<\/span> <\/span>+=<\/span> <\/span>match<\/span>.<\/span>sum<\/span>(<\/span>)<\/span>.<\/span>item<\/span>(<\/span>)<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0<\/span>num_samples<\/span> <\/span>+=<\/span> <\/span>len<\/span>(<\/span>start_positions<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>avg_loss<\/span> <\/span>=<\/span> <\/span>val_loss<\/span> <\/span>\/<\/span> <\/span>num_batches<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>acc<\/span> <\/span>=<\/span> <\/span>num_matches<\/span> <\/span>\/<\/span> <\/span>num_samples<\/span><\/p>\n

\u00a0\u00a0\u00a0\u00a0<\/span>print<\/span>(<\/span>f<\/span>“Validation {epoch+1}\/{num_epochs}: acc {acc:.4f}, avg loss {avg_loss:.4f}”<\/span>)<\/span><\/p>\n

\u00a0<\/p>\n

# Save the fine-tuned model<\/span><\/p>\n

torch<\/span>.<\/span>save<\/span>(<\/span>model<\/span>.<\/span>state_dict<\/span>(<\/span>)<\/span>,<\/span> <\/span>f<\/span>“bert_model_squad.pth”<\/span>)<\/span><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

import collections import dataclasses import functools \u00a0 import torch import torch.nn as nn import torch.optim as optim import tqdm from datasets import load_dataset from tokenizers import Tokenizer from torch import Tensor \u00a0 \u00a0 # BERT config and model defined previously @dataclasses.dataclass class BertConfig: \u00a0\u00a0\u00a0\u00a0“”“Configuration for BERT model.”“” \u00a0\u00a0\u00a0\u00a0vocab_size: int = 30522 \u00a0\u00a0\u00a0\u00a0num_layers: int = 12<\/p>\n","protected":false},"author":1,"featured_media":177203,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[42],"tags":[],"class_list":{"0":"post-177202","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-ai"},"yoast_head":"\nFine-Tuning a BERT Model - MachineLearningMastery.com - Ktromedia<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"http:\/\/ktromedia.com\/?p=177202\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Fine-Tuning a BERT Model - MachineLearningMastery.com - Ktromedia\" \/>\n<meta property=\"og:description\" content=\"import collections import dataclasses import functools \u00a0 import torch import torch.nn as nn import torch.optim as optim import tqdm from datasets import load_dataset from tokenizers import Tokenizer from torch import Tensor \u00a0 \u00a0 # BERT config and model defined previously @dataclasses.dataclass class BertConfig: \u00a0\u00a0\u00a0\u00a0“”“Configuration for BERT model.”“” \u00a0\u00a0\u00a0\u00a0vocab_size: int = 30522 \u00a0\u00a0\u00a0\u00a0num_layers: int = 12\" \/>\n<meta property=\"og:url\" content=\"http:\/\/ktromedia.com\/?p=177202\" \/>\n<meta property=\"og:site_name\" content=\"Ktromedia\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/KTROMedia\/\" \/>\n<meta property=\"article:published_time\" content=\"2025-12-01T13:28:32+00:00\" \/>\n<meta property=\"og:image\" content=\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"2560\" \/>\n\t<meta property=\"og:image:height\" content=\"1707\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"KTRO TEAM\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"KTRO TEAM\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"http:\/\/ktromedia.com\/?p=177202#article\",\"isPartOf\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177202\"},\"author\":{\"name\":\"KTRO TEAM\",\"@id\":\"http:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b\"},\"headline\":\"Fine-Tuning a BERT Model – MachineLearningMastery.com\",\"datePublished\":\"2025-12-01T13:28:32+00:00\",\"mainEntityOfPage\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177202\"},\"wordCount\":1387,\"commentCount\":0,\"publisher\":{\"@id\":\"http:\/\/ktromedia.com\/#organization\"},\"image\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177202#primaryimage\"},\"thumbnailUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg\",\"articleSection\":[\"\u4eba\u5de5\u667a\u80fd\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"http:\/\/ktromedia.com\/?p=177202#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"http:\/\/ktromedia.com\/?p=177202\",\"url\":\"http:\/\/ktromedia.com\/?p=177202\",\"name\":\"Fine-Tuning a BERT Model - MachineLearningMastery.com - Ktromedia\",\"isPartOf\":{\"@id\":\"http:\/\/ktromedia.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177202#primaryimage\"},\"image\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177202#primaryimage\"},\"thumbnailUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg\",\"datePublished\":\"2025-12-01T13:28:32+00:00\",\"breadcrumb\":{\"@id\":\"http:\/\/ktromedia.com\/?p=177202#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"http:\/\/ktromedia.com\/?p=177202\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"http:\/\/ktromedia.com\/?p=177202#primaryimage\",\"url\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg\",\"contentUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg\",\"width\":2560,\"height\":1707},{\"@type\":\"BreadcrumbList\",\"@id\":\"http:\/\/ktromedia.com\/?p=177202#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/ktromedia.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Fine-Tuning a BERT Model – MachineLearningMastery.com\"}]},{\"@type\":\"WebSite\",\"@id\":\"http:\/\/ktromedia.com\/#website\",\"url\":\"http:\/\/ktromedia.com\/\",\"name\":\"Ktromedia\",\"description\":\"KTRO MEDIA Crypto News\",\"publisher\":{\"@id\":\"http:\/\/ktromedia.com\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"http:\/\/ktromedia.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"http:\/\/ktromedia.com\/#organization\",\"name\":\"Ktromedia\",\"url\":\"http:\/\/ktromedia.com\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"http:\/\/ktromedia.com\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png\",\"contentUrl\":\"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png\",\"width\":250,\"height\":250,\"caption\":\"Ktromedia\"},\"image\":{\"@id\":\"http:\/\/ktromedia.com\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/KTROMedia\/\",\"https:\/\/www.linkedin.com\/company\/ktro-media\/\",\"https:\/\/t.me\/ktrogroup\"]},{\"@type\":\"Person\",\"@id\":\"http:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b\",\"name\":\"KTRO TEAM\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"http:\/\/ktromedia.com\/#\/schema\/person\/image\/\",\"url\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png\",\"contentUrl\":\"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png\",\"caption\":\"KTRO TEAM\"},\"description\":\"KTRO MEDIA \u662f\u4e00\u5bb6\u5168\u7403\u6027\u7684\u534e\u6587WEB3\u5a92\u4f53\u516c\u53f8\u3002\u6211\u4eec\u81f4\u529b\u4e8e\u4e3a\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u9886\u57df\u63d0\u4f9b\u6700\u65b0\u7684\u65b0\u95fb\u3001\u89c1\u89e3\u548c\u8d8b\u52bf\u5206\u6790\u3002\u6211\u4eec\u7684\u5b97\u65e8\u662f\u4e3a\u5168\u7403\u7528\u6237\u63d0\u4f9b\u9ad8\u8d28\u91cf\u3001\u5168\u9762\u7684\u8d44\u8baf\u670d\u52a1\uff0c\u8ba9\u4ed6\u4eec\u66f4\u597d\u5730\u4e86\u89e3\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6700\u65b0\u52a8\u6001\u3002\u6211\u4eec\u4e5f\u5e0c\u671b\u80fd\u5e2e\u5230\u66f4\u591a\u4f18\u79c0\u7684WEB3\u4ea7\u54c1\u627e\u5230\u66f4\u591a\u66f4\u597d\u7684\u8d44\u6e90\u597d\u8ba9\u8fd9\u9886\u57df\u53d8\u5f97\u66f4\u6210\u719f\u3002 \u6211\u4eec\u7684\u62a5\u9053\u8303\u56f4\u6db5\u76d6\u4e86\u533a\u5757\u94fe\u3001\u52a0\u5bc6\u8d27\u5e01\u3001\u667a\u80fd\u5408\u7ea6\u3001DeFi\u3001NFT \u548c Web3 \u751f\u6001\u7cfb\u7edf\u7b49\u9886\u57df\u3002\u6211\u4eec\u7684\u62a5\u9053\u4e0d\u4ec5\u6765\u81ea\u884c\u4e1a\u5185\u7684\u4e13\u5bb6\uff0c\u5148\u950b\u8005\u4e5f\u5305\u62ec\u4e86\u6211\u4eec\u81ea\u5df1\u7684\u5206\u6790\u548c\u89c2\u70b9\u3002\u6211\u4eec\u5728\u5404\u4e2a\u56fd\u5bb6\u548c\u5730\u533a\u90fd\u8bbe\u6709\u56e2\u961f\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u672c\u5730\u5316\u7684\u62a5\u9053\u548c\u5206\u6790\u3002 \u9664\u4e86\u65b0\u95fb\u62a5\u9053\uff0c\u6211\u4eec\u8fd8\u63d0\u4f9b\u5e02\u573a\u7814\u7a76\u548c\u54a8\u8be2\u670d\u52a1\u3002\u6211\u4eec\u7684\u4e13\u4e1a\u56e2\u961f\u53ef\u4ee5\u4e3a\u60a8\u63d0\u4f9b\u6709\u5173\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6df1\u5165\u5206\u6790\u548c\u5e02\u573a\u8d8b\u52bf\uff0c\u5e2e\u52a9\u60a8\u505a\u51fa\u66f4\u660e\u667a\u7684\u6295\u8d44\u51b3\u7b56\u3002 \u6211\u4eec\u7684\u4f7f\u547d\u662f\u6210\u4e3a\u5168\u7403\u534e\u6587\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u6700\u53d7\u4fe1\u8d56\u7684\u4fe1\u606f\u6765\u6e90\u4e4b\u4e00\u3002\u6211\u4eec\u5c06\u7ee7\u7eed\u4e0d\u65ad\u52aa\u529b\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u6700\u65b0\u3001\u6700\u5168\u9762\u3001\u6700\u53ef\u9760\u7684\u4fe1\u606f\u670d\u52a1\u3002\",\"sameAs\":[\"https:\/\/ktromedia.com\"],\"url\":\"http:\/\/ktromedia.com\/?author=1\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Fine-Tuning a BERT Model - MachineLearningMastery.com - Ktromedia","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"http:\/\/ktromedia.com\/?p=177202","og_locale":"en_US","og_type":"article","og_title":"Fine-Tuning a BERT Model - MachineLearningMastery.com - Ktromedia","og_description":"import collections import dataclasses import functools \u00a0 import torch import torch.nn as nn import torch.optim as optim import tqdm from datasets import load_dataset from tokenizers import Tokenizer from torch import Tensor \u00a0 \u00a0 # BERT config and model defined previously @dataclasses.dataclass class BertConfig: \u00a0\u00a0\u00a0\u00a0“”“Configuration for BERT model.”“” \u00a0\u00a0\u00a0\u00a0vocab_size: int = 30522 \u00a0\u00a0\u00a0\u00a0num_layers: int = 12","og_url":"http:\/\/ktromedia.com\/?p=177202","og_site_name":"Ktromedia","article_publisher":"https:\/\/www.facebook.com\/KTROMedia\/","article_published_time":"2025-12-01T13:28:32+00:00","og_image":[{"width":2560,"height":1707,"url":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg","type":"image\/jpeg"}],"author":"KTRO TEAM","twitter_card":"summary_large_image","twitter_misc":{"Written by":"KTRO TEAM","Est. reading time":"7 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"http:\/\/ktromedia.com\/?p=177202#article","isPartOf":{"@id":"http:\/\/ktromedia.com\/?p=177202"},"author":{"name":"KTRO TEAM","@id":"http:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b"},"headline":"Fine-Tuning a BERT Model – MachineLearningMastery.com","datePublished":"2025-12-01T13:28:32+00:00","mainEntityOfPage":{"@id":"http:\/\/ktromedia.com\/?p=177202"},"wordCount":1387,"commentCount":0,"publisher":{"@id":"http:\/\/ktromedia.com\/#organization"},"image":{"@id":"http:\/\/ktromedia.com\/?p=177202#primaryimage"},"thumbnailUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg","articleSection":["\u4eba\u5de5\u667a\u80fd"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["http:\/\/ktromedia.com\/?p=177202#respond"]}]},{"@type":"WebPage","@id":"http:\/\/ktromedia.com\/?p=177202","url":"http:\/\/ktromedia.com\/?p=177202","name":"Fine-Tuning a BERT Model - MachineLearningMastery.com - Ktromedia","isPartOf":{"@id":"http:\/\/ktromedia.com\/#website"},"primaryImageOfPage":{"@id":"http:\/\/ktromedia.com\/?p=177202#primaryimage"},"image":{"@id":"http:\/\/ktromedia.com\/?p=177202#primaryimage"},"thumbnailUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg","datePublished":"2025-12-01T13:28:32+00:00","breadcrumb":{"@id":"http:\/\/ktromedia.com\/?p=177202#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["http:\/\/ktromedia.com\/?p=177202"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"http:\/\/ktromedia.com\/?p=177202#primaryimage","url":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg","contentUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/12\/Fine-Tuning-a-BERT-Model-MachineLearningMasterycom.jpg","width":2560,"height":1707},{"@type":"BreadcrumbList","@id":"http:\/\/ktromedia.com\/?p=177202#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/ktromedia.com\/"},{"@type":"ListItem","position":2,"name":"Fine-Tuning a BERT Model – MachineLearningMastery.com"}]},{"@type":"WebSite","@id":"http:\/\/ktromedia.com\/#website","url":"http:\/\/ktromedia.com\/","name":"Ktromedia","description":"KTRO MEDIA Crypto News","publisher":{"@id":"http:\/\/ktromedia.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"http:\/\/ktromedia.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"http:\/\/ktromedia.com\/#organization","name":"Ktromedia","url":"http:\/\/ktromedia.com\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"http:\/\/ktromedia.com\/#\/schema\/logo\/image\/","url":"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png","contentUrl":"https:\/\/ktromedia.com\/wp-content\/uploads\/2025\/11\/ktroicon.png","width":250,"height":250,"caption":"Ktromedia"},"image":{"@id":"http:\/\/ktromedia.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/KTROMedia\/","https:\/\/www.linkedin.com\/company\/ktro-media\/","https:\/\/t.me\/ktrogroup"]},{"@type":"Person","@id":"http:\/\/ktromedia.com\/#\/schema\/person\/612bf2fbac107722ea365932cdd35f5b","name":"KTRO TEAM","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"http:\/\/ktromedia.com\/#\/schema\/person\/image\/","url":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png","contentUrl":"http:\/\/ktromedia.com\/wp-content\/uploads\/2025\/10\/cropped-Untitled-design-7-1-150x150.png","caption":"KTRO TEAM"},"description":"KTRO MEDIA \u662f\u4e00\u5bb6\u5168\u7403\u6027\u7684\u534e\u6587WEB3\u5a92\u4f53\u516c\u53f8\u3002\u6211\u4eec\u81f4\u529b\u4e8e\u4e3a\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u9886\u57df\u63d0\u4f9b\u6700\u65b0\u7684\u65b0\u95fb\u3001\u89c1\u89e3\u548c\u8d8b\u52bf\u5206\u6790\u3002\u6211\u4eec\u7684\u5b97\u65e8\u662f\u4e3a\u5168\u7403\u7528\u6237\u63d0\u4f9b\u9ad8\u8d28\u91cf\u3001\u5168\u9762\u7684\u8d44\u8baf\u670d\u52a1\uff0c\u8ba9\u4ed6\u4eec\u66f4\u597d\u5730\u4e86\u89e3\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6700\u65b0\u52a8\u6001\u3002\u6211\u4eec\u4e5f\u5e0c\u671b\u80fd\u5e2e\u5230\u66f4\u591a\u4f18\u79c0\u7684WEB3\u4ea7\u54c1\u627e\u5230\u66f4\u591a\u66f4\u597d\u7684\u8d44\u6e90\u597d\u8ba9\u8fd9\u9886\u57df\u53d8\u5f97\u66f4\u6210\u719f\u3002 \u6211\u4eec\u7684\u62a5\u9053\u8303\u56f4\u6db5\u76d6\u4e86\u533a\u5757\u94fe\u3001\u52a0\u5bc6\u8d27\u5e01\u3001\u667a\u80fd\u5408\u7ea6\u3001DeFi\u3001NFT \u548c Web3 \u751f\u6001\u7cfb\u7edf\u7b49\u9886\u57df\u3002\u6211\u4eec\u7684\u62a5\u9053\u4e0d\u4ec5\u6765\u81ea\u884c\u4e1a\u5185\u7684\u4e13\u5bb6\uff0c\u5148\u950b\u8005\u4e5f\u5305\u62ec\u4e86\u6211\u4eec\u81ea\u5df1\u7684\u5206\u6790\u548c\u89c2\u70b9\u3002\u6211\u4eec\u5728\u5404\u4e2a\u56fd\u5bb6\u548c\u5730\u533a\u90fd\u8bbe\u6709\u56e2\u961f\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u672c\u5730\u5316\u7684\u62a5\u9053\u548c\u5206\u6790\u3002 \u9664\u4e86\u65b0\u95fb\u62a5\u9053\uff0c\u6211\u4eec\u8fd8\u63d0\u4f9b\u5e02\u573a\u7814\u7a76\u548c\u54a8\u8be2\u670d\u52a1\u3002\u6211\u4eec\u7684\u4e13\u4e1a\u56e2\u961f\u53ef\u4ee5\u4e3a\u60a8\u63d0\u4f9b\u6709\u5173\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u7684\u6df1\u5165\u5206\u6790\u548c\u5e02\u573a\u8d8b\u52bf\uff0c\u5e2e\u52a9\u60a8\u505a\u51fa\u66f4\u660e\u667a\u7684\u6295\u8d44\u51b3\u7b56\u3002 \u6211\u4eec\u7684\u4f7f\u547d\u662f\u6210\u4e3a\u5168\u7403\u534e\u6587\u533a\u5757\u94fe\u548c\u91d1\u878d\u79d1\u6280\u884c\u4e1a\u6700\u53d7\u4fe1\u8d56\u7684\u4fe1\u606f\u6765\u6e90\u4e4b\u4e00\u3002\u6211\u4eec\u5c06\u7ee7\u7eed\u4e0d\u65ad\u52aa\u529b\uff0c\u4e3a\u8bfb\u8005\u63d0\u4f9b\u6700\u65b0\u3001\u6700\u5168\u9762\u3001\u6700\u53ef\u9760\u7684\u4fe1\u606f\u670d\u52a1\u3002","sameAs":["https:\/\/ktromedia.com"],"url":"http:\/\/ktromedia.com\/?author=1"}]}},"_links":{"self":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts\/177202","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=177202"}],"version-history":[{"count":1,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts\/177202\/revisions"}],"predecessor-version":[{"id":177204,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/posts\/177202\/revisions\/177204"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=\/wp\/v2\/media\/177203"}],"wp:attachment":[{"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=177202"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=177202"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/ktromedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=177202"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}