hf-llm-api / messagers

Commit History

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

    8697de8
  2. :zap: [Enhance] Add new model yi-1.5-34b, and remove unavailable models

    b9a642a

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

    214fb7b
  2. :zap: [Enhance] Quieter openai auth, use cffi to request hf-chat id, and console token count for exceeds

    0f710a2
  3. :gem: [Feature] Moduralize TokenChecker, and fix gated model repos with alternatives

    8df3985
  4. :gem: [Feature] MessageCompoer: decompose to system and input prompt

    0b6e0c9

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

    3b00a19

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

    3965e1f

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

    cd6b52a
  2. :recycle: [Refactor] Use tclogger, and configure envs better

    babcd78

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

    dc3b7a9
  2. :recycle: [Refactor] Use official chat_template to replace mannual parses for several models, and remove unused message split functions

    4a9e46f
  3. :recycle: [Refactor] Move AVAILABLE_MODELS(_DICTS) to constants

    40ba0ea
  4. :gem: [Feature] Support new model: openchat-3.5-0106

    9f5d69c

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

    77b5a47

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

    395e196
  2. :gem: [Feature] New model enabled: nous-mixtral-8-7b

    4a28ca5

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

    30421b7

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

    d2b20f2

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

    eb3e513
  2. :gem: [Feature] New MessageComposer: Merge and split messages in mixtral-instruct format

    9e62d8e