Research Open-Source LLMs

[grayJiaaoLi]

User story

1. As a software developer
2. I need to make a list of potential open-source LLMs for the project
3. So that we can identify the best one for our project from comparison.

Acceptance criteria

• A list of 3-5 open-source LLMs
• The candidate LLMs are generally suitable for the project's requirements
  • Answer natural language questions about CNCF projects and their documentation
  • Explain relationships between projects
  • Recommend projects based on use cases or constraints
  • Each LLM is available under a permissive license (MIT, Apache 2.0 etc.)
  • (Do not have to generate code)
• The Gemma model is part of the research
• Link the relevant resources and documentation for each LLM

DoD general criteria

• Feature has been fully implemented
• Feature has been merged into the mainline
• All acceptance criteria were met
• Product owner approved features
• All tests are passing
• Developers agreed to release

[julioc-p]

Gemma, License Apache 2.0 Mixtral-8x22b-Instruct-v0.1 (Apache 2.0) Zephyr-ORPO-141b-A35b-v0.1 (Apache 2.0) Snowflake Arctic Instruct (Apache 2.0)

Documentation:

Gemma: https://ai.google.dev/gemma Mixtral: https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1 Zephyr: https://huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1 Snowflake Arctic Instruct: https://www.snowflake.com/en/data-cloud/arctic/

Resources: https://towardsdatascience.com/choosing-the-right-language-model-for-your-nlp-use-case-1288ef3c4929

https://chat.lmsys.org/?leaderboard

https://github.com/eugeneyan/open-llms

https://storage.googleapis.com/deepmind-media/gemma/gemma-report.pdf