Julia Implementation of Transformer models
Author chengchingwen
82 Stars
Updated Last
10 Months Ago
Started In
December 2018

Build Status codecov

Julia implementation of transformer-based models, with Flux.jl.


In the Julia REPL:

]add Transformers

For using GPU, install & build:

]add CuArrays


julia> using CuArrays

julia> using Transformers

#run the model below


Using pretrained Bert with Transformers.jl.

using Transformers
using Transformers.Basic
using Transformers.Pretrain


bert_model, wordpiece, tokenizer = pretrain"bert-uncased_L-12_H-768_A-12"
vocab = Vocabulary(wordpiece)

text1 = "Peter Piper picked a peck of pickled peppers" |> tokenizer |> wordpiece
text2 = "Fuzzy Wuzzy was a bear" |> tokenizer |> wordpiece

text = ["[CLS]"; text1; "[SEP]"; text2; "[SEP]"]
@assert text == [
    "[CLS]", "peter", "piper", "picked", "a", "peck", "of", "pick", "##led", "peppers", "[SEP]", 
    "fuzzy", "wu", "##zzy",  "was", "a", "bear", "[SEP]"

token_indices = vocab(text)
segment_indices = [fill(1, length(text1)+2); fill(2, length(text2)+1)]

sample = (tok = token_indices, segment = segment_indices)

bert_embedding = sample |> bert_model.embed
feature_tensors = bert_embedding |> bert_model.transformers

See example folder for the complete example.


What we have in v0.1.0

  • Transformer and TransformerDecoder support for both 2d & 3d data.
  • PositionEmbedding implementation.
  • Positionwise for handling 2d & 3d input.
  • docstring for most of the functions.
  • runable examples (see example folder)

What we will have in v0.2.0

  • The BERT model (JSoC 2019)
  • tutorials
  • complete GPT APIs
  • GPT-2 model
  • docs site for this project
  • benchmarks
  • more examples

What we might have in v0.2.0 (If we are lucky)

  • TPU support with XLA.jl
  • complete docs for datasets
  • more datasets support

Messy checklist

  • refactor code
  • optimize performance
  • better dataset API
  • more datasets
  • openai gpt model
  • google bert model
  • openai gpt-2 model
  • TPU support
  • openai sparse transformer
  • benchmarks