Open-Assistant.yaml

---
# Thank you for contributing!
# In filling out this yaml file, please follow the criteria as described here: 
# https://github.com/opening-up-chatgpt/opening-up-chatgpt.github.io/tree/main/projects#criteria

# You're free to build on this work and reuse the data. It is licensed under CC-BY 4.0, with the
# stipulation that attribution should come in the form of a link to http://opening-up-chatgpt.github.io
# and a citation to the paper in which the initial dataset & criteria were published:

# Liesenfeld, Andreas, Alianda Lopez, and Mark Dingemanse. 2023. “Opening up ChatGPT: Tracking Openness, Transparency, and Accountability in Instruction-Tuned Text Generators.” In CUI '23: Proceedings of the 5th International Conference on Conversational User Interfaces. July 19-21, Eindhoven. doi: 10.1145/3571884.3604316 

system:
    name: Open Assistant
    link: https://open-assistant.io/
    type: text 
    performanceclass: full
    basemodelname: Pythia 12B
    endmodelname: OpenAssistant Conversations
    endmodellicense: Apache 2.0
    releasedate: 2023-02
    notes:

org:
    name: LAION-AI
    link: https://open-assistant.io/
    notes:

# availability:
datasources_basemodel:
    class: open
    link: https://github.com/LAION-AI/Open-Assistant/tree/main/data/datasets
    notes: Datasets documented in detail and recipes for cleaning up and downloading provided in code notebooks.

datasources_endmodel:
    class: open
    link: https://huggingface.co/datasets/OpenAssistant/oasst1
    notes: OpenAssistant Conversations is 'a human-generated, human-annotated assistant-style conversation corpus consisting of 161443 messages distributed across 66497 conversation trees, in 35 different languages, annotated with 461292 quality ratings' (preprint)

weights_basemodel:
    class: open
    link: https://huggingface.co/OpenAssistant
    notes: Model weights in several variants downloadable through HuggingFace
    
weights_endmodel:
    class: closed
    link:
    notes: RLHF weights not separately released

trainingcode:
    class: open
    link: https://github.com/LAION-AI/Open-Assistant
    notes: Code includes guide for developers

# documentation
code:
    class: open
    link: https://projects.laion.ai/Open-Assistant/docs/intro
    notes: Separate website provides entry point to comprehensive documentation

architecture:
    class: open
    link: https://github.com/LAION-AI/Open-Assistant/tree/main/model
    notes: Instructions to tune the pipeline on training data

preprint:
    class: partial
    link: https://arxiv.org/abs//2304.07327
    notes: Preprint describes creation of OpenAssistant Conversations corpus for instruction tuning, but not the base LLM, hence partial.

paper:
    class: partial
    link: https://proceedings.neurips.cc/paper_files/paper/2023/hash/949f0f8f32267d297c2d4e3ee10a2e7e-Abstract-Datasets_and_Benchmarks.html
    notes: Preprint was published in NeurIPS

modelcard:
    class: partial
    link: https://huggingface.co/OpenAssistant
    notes: Various model cards exist

datasheet:
    class: partial
    link: https://docs.google.com/spreadsheets/d/1NYYa6vHiRnk5kwnyYaCT0cBO62--Tm3w4ihdBtp4ISk/edit?pli=1&gid=1537161081#gid=1537161081
    notes: Most data sets are linked and some contain a data sheet.

# access
package:
    class: open
    link:
    notes:

api:
    class: open
    link: https://projects.laion.ai/Open-Assistant/api
    notes:
    metaprompt: closed

licenses:
    class: open
    link: https://projects.laion.ai/Open-Assistant/docs/faq#what-license-does-open-assistant-use
    notes: Apache 2.0