v0.2.0

• Tiktoken-style tokenization via GenGPT3.tokenize(encoding, text)
• Support for OpenAI models using the cl100k_base encoding (babbage-002, davinci-002)

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Initial version of GenGPT3.jl, with support for:

• Calling OpenAI's original GPT-3 and GPT-3.5 series of models as generative functions
• Batched generation and mixture of prompt models, which reduce API overhead
• Importance sampling with a pair of GPT-3 generative functions
• Text embedding via the OpenAI embeddings endpoint
• Tokenization using the gpt2 BPE tokenizer

Since OpenAI has shutdown access to the original GPT-3 and GPT-3.5 series of models (as of Jan 4, 2024), this version no longer provides much in the way of useful functionality.