:boom: [Fix] slow tokenizer issue, and finish by stop_sequences

:zap: [Enhance] Add new model yi-1.5-34b, and remove unavailable models

:zap: [Enhance] Use nous-mixtral-8x7b as default model

:zap: [Enhance] Quieter openai auth, use cffi to request hf-chat id, and console token count for exceeds

:gem: [Feature] Moduralize TokenChecker, and fix gated model repos with alternatives

:gem: [Feature] MessageCompoer: decompose to system and input prompt

:zap: [Enhance] Upgrade gemma-7b to 1.1

:gem: [Feature] New model supported: command-r-plus (CohereForai/c4ai-command-r-plus)

:gem: [Feature] Enable gpt-3.5 in chat_api

:recycle: [Refactor] Use tclogger, and configure envs better

:boom: [Fix] Revert chat templates with mannual version as nous-mixtral-8x7b behaves weirdly

:recycle: [Refactor] Use official chat_template to replace mannual parses for several models, and remove unused message split functions

:recycle: [Refactor] Move AVAILABLE_MODELS(_DICTS) to constants

:gem: [Feature] Support new model: openchat-3.5-0106

:gem: [Feature] New model enabled: gemma-7b

:boom: [Fix] Missing system message for original mistral models

:gem: [Feature] New model enabled: nous-mixtral-8-7b

:gem: [Feature] MessageComposer: Support openchat-3.5, and generalize preprocessing stages

:gem: [Feature] Support no-stream mode with dict response

:gem: [Feature] New MessageOutputer: Support Openai Stream format

:gem: [Feature] New MessageComposer: Merge and split messages in mixtral-instruct format