🕯️ candle - Minimalist ML framework - for Ruby
require "candle"
x = Candle::Tensor.new([1, 2, 3, 4, 5, 6], :i64)
x = x.reshape([3, 2])
# [[1., 2.],
# [3., 4.],
# [5., 6.]]
# Tensor[[3, 2], f32]require 'candle'
model = Candle::Model.new
embedding = model.embedding("Hi there!")The Candle::Model defaults to the jinaai/jina-embeddings-v2-base-en model with the sentence-transformers/all-MiniLM-L6-v2 tokenizer (both from HuggingFace). With this configuration the model takes a little more than 3GB of memory running on my Mac. The memory stays with the instantiated Candle::Model class, if you instantiate more than one, you'll use more memory. Likewise, if you let it go out of scope and call the garbage collector, you'll free the memory. For example:
> require 'candle'
# Ruby memory = 25.9 MB
> model = Candle::Model.new
# Ruby memory = 3.50 GB
> model2 = Candle::Model.new
# Ruby memory = 7.04 GB
> model2 = nil
> GC.start
# Ruby memory = 3.56 GB
> model = nil
> GC.start
# Ruby memory = 55.2 MBThe code should match the same embeddings when generated from the python transformers library. For instance, locally I was able to generate the same embedding for the text "Hi there!" using the python code:
from transformers import AutoModel
model = AutoModel.from_pretrained('jinaai/jina-embeddings-v2-base-en', trust_remote_code=True)
sentence = ['Hi there!']
embedding = model.encode(sentence)
print(embedding)And the following ruby:
require 'candle'
model = Candle::Model.new
embedding = model.embedding("Hi there!")FORK IT!
git clone https://github.com/your_name/red-candle
cd red-candle
bundle
bundle exec rake compile
Implemented with Magnus, with reference to Polars Ruby
Pull requests are welcome.