Inference in Python uses harmony [1] (for request and response format) which is written in Rust with Python bindings. Another OpenAI's Rust library is tiktoken

Inference in Python uses harmony [1] (for request and response format) which is written in Rust with Python bindings. Another OpenAI's Rust library is tiktoken [2], used for all tokenization and detokenization. OpenAI Codex [3] is also written in Rust. It looks like OpenAI is increasingly adopting Rust (at least for inference). [1] https://github.com/openai/harmony https://github.com/openai/harmony [2] https://github.com/openai/tiktoken https://github.com/openai/tiktoken [3] https://github.com/openai/codex https://github.com/openai/codex