GPTNeoXClient

gpt_neox_client is a simple client for GPT-NeoX.

Install the gem and add to the application's Gemfile by executing:

$ bundle add gpt_neox_client

If bundler is not being used to manage dependencies, install the gem by executing:

$ gem install gpt_neox_client

GPTNeoXClient supports ggml format model. Model files need to be converted to ggml format. The following is an example of converting the model published by LINE Corporation:

# Clone ggml.
$ git clone https://github.com/ggerganov/ggml.git
$ cd ggml

# Install python dependencies.
$ pip install -U protobuf~=3.20.0
$ python -m pip install -r requirements.txt

# Convert downloaded model to ggml format.
$ git clone https://huggingface.co/line-corporation/japanese-large-lm-3.6b-instruction-sft
$ python examples/gpt-neox/convert-h5-to-ggml.py japanese-large-lm-3.6b-instruction-sft 1
$ ls japanese-large-lm-3.6b-instruction-sft/ggml-model-f16.bin
japanese-large-lm-3.6b-instruction-sft/ggml-model-f16.bin

require 'gpt_neox_client'

client = GPTNeoXClient.new(path: 'japanese-large-lm-3.6b-instruction-sft/ggml-model-f16.bin', seed: 123456789, n_threads: 4)
puts client.completions(
  'ユーザー:四国の県名を全て列挙してください。<0x0A>システム:',
  top_p: 0.9,
  top_k: 1,
  temperature: 0.7
).gsub('<0x0A>', "\n").gsub('</s>', '')
#
# ユーザー:四国の県名を全て列挙してください。
# システム:徳島県、香川県、愛媛県、高知県

pp client.embeddings('こんにちは世界', normalize: true)
#
# [-0.00973147340118885,
#  0.007523993495851755,
# ...
#  -0.004352027550339699,
#  -0.024439003318548203]

Bug reports and pull requests are welcome on GitHub at https://github.com/yoshoku/gpt_neox_client. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the code of conduct.

The gem is available as open source under the terms of the MIT License.

Everyone interacting in the GPTNeoXClient project's codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.