None
Project description
pytorch-fast-elmo
Introduction
A fast ELMo implementation with features:
Lower execution overhead. The core components are reimplemented in Libtorch in order to reduce the Python execution overhead (45% speedup).
A more flexible design. By redesigning the workflow, the user could extend or change the ELMo behavior easily.
Benchmark
Hardware:
CPU: i7-7800X
GPU: 1080Ti
Options:
Batch size: 32
Warm up iterations: 20
Test iterations: 1000
Word length: [1, 20]
Sentence length: [1, 30]
Random seed: 10000
Item |
Mean Of Durations (ms) |
cumtime(synchronize)% |
---|---|---|
Fast ELMo (CUDA, no synchronize) |
31 |
N/A |
AllenNLP ELMo (CUDA, no synchronize) |
56 |
N/A |
Fast ELMo (CUDA, synchronize) |
47 |
26.13% |
AllenNLP ELMo (CUDA, synchronize) |
57 |
0.02% |
Fast ELMo (CPU) |
1277 |
N/A |
AllenNLP ELMo (CPU) |
1453 |
N/A |
Usage
Please install torch==1.0.0 first. Then, simply run this command to install.
pip install pytorch-fast-elmo
FastElmo should have the same behavior as AllenNLP’s ELMo.
from pytorch_fast_elmo import FastElmo, batch_to_char_ids
options_file = '/path/to/elmo_2x4096_512_2048cnn_2xhighway_options.json'
weight_file = '/path/to/elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5'
elmo = FastElmo(options_file, weight_file)
sentences = [['First', 'sentence', '.'], ['Another', '.']]
character_ids = batch_to_ids(sentences)
embeddings = elmo(character_ids)
Use FastElmoWordEmbedding if you have disabled char_cnn in bilm-tf, or have exported the Char CNN representation to a weight file.
from pytorch_fast_elmo import FastElmoWordEmbedding, load_and_build_vocab2id, batch_to_word_ids
options_file = '/path/to/elmo_2x4096_512_2048cnn_2xhighway_options.json'
weight_file = '/path/to/elmo_2x4096_512_2048cnn_2xhighway_weights.hdf5'
vocab_file = '/path/to/vocab.txt'
embedding_file = '/path/to/cached_elmo_embedding.hdf5'
elmo = FastElmoWordEmbedding(
options_file,
weight_file,
# Could be omitted if the embedding weight is in `weight_file`.
word_embedding_weight_file=embedding_file,
)
vocab2id = load_and_build_vocab2id(vocab_file)
sentences = [['First', 'sentence', '.'], ['Another', '.']]
word_ids = batch_to_word_ids(sentences, vocab2id)
embeddings = elmo(word_ids)
CLI commands:
# For exporting the Char CNN representation.
fast-elmo cache-char-cnn ./vocab.txt ./options.json ./lm_weights.hdf5 ./lm_ebd.hdf5
Credits
This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.
History
0.1.0 (2019-01-02)
First release on PyPI.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.