Val Kharitonov

Use tiktoken tokenizer for gpt-3.5 and gpt-4 models

2

The tokenizer seems to work, but something is definitely missing (it doesn't continue sampling after the first token, see screenshot) Fixes #23

Add support for logit_bias for OpenAI Chat models / support for cl100k_base tokenizer

2

Currently, lmql doesn't support logit bias-based constraints for `openai/gpt-3.5-turbo` and `openai/gpt-4`. However, the [OpenAI documentation](https://platform.openai.com/docs/api-reference/chat/create#chat/create-logit_bias) states that `logit_bias` is indeed supported by the API. The reason why it currently doesn't...

In gpt.yml: ```yml together_api_key: assistants: mixtral: model: together/mistralai/Mixtral-8x7B-Instruct-v0.1 temperature: 0.7 user_prefix: " [INST] " user_suffix: " [/INST]" stop_tokens: - "" ``` ```bash gpt mixtral ```

This is in a super draft state, but mostly works, including inline images (at least in iTerm2). Under the hood, this uses new [Function Calling API](https://platform.openai.com/docs/guides/gpt/function-calling). Added Docker container for...

### Environment Node version: v22.5.1 npm version: v10.8.2 Local ESLint version: v9.12.0 (Currently used) Global ESLint version: Not found Operating System: darwin 23.6.0 ### What parser are you using? Default...