59befe6c07 | ||
---|---|---|
img | ||
.gitattributes | ||
.gitignore | ||
CHANGELOG.txt | ||
LICENSE.txt | ||
README.md | ||
models.ods |
README.md
Parrot Models
Models for Parrot Libre AI IDE.
Libre Models
Parrot is a Libre AI, so it needs a good free content ("open source") data model.
There are endless AI models, only some suitable for this application. Of the available models, a smaller subset claims to be "open source". Of the "open source" subset, only a much smaller subset actually uses an "open source" license. Of the "open source" subset that actually uses "open source" licenses, only a smaller subset uses exclusively "open source" training data.
AFAICT, there are exactly ZERO (0) AI instruct models are actually "open source" trained on "open source" datasets.
Parrot Model Licensing
Preferably, there is some truly libre instruct model suitable for Parrot. Likely, one will have to be created from scratch.
The model may use data that is under a license that appears on one of these three lists as an acceptable free/open license:
Models to Evaluate
Models freely available to download, but may not have suitable license. Determine which, if any, are ok.
- DeepSeek
- EleutherAI
- EverythingLM
- GPT-NeoX
- Llemma
- Mistral
- Orca-mini
- Pythia
- StarCoder
Suitable Models
The following models may be suitable, but need further evalution.
Suitable Licenses
Models that appear to have libre licenses.
Apache 2.0
The following models use the Apache 2.0 license:
- Falcon 7B
- Falcon 40B
CC-BY-SA
The following models use the CC by SA license:
- Open-Assistant StableLM
- SQLCoder
- StableLM
Suitable Model, Unsuitable Dataset
The following have libre licenses, but are built from non-libre datasets.
- Falcon 7B --- Refinedweb (Common Crawl).
- Falcon 40B --- Refinedweb (Common Crawl).
Suitable Model, Suitable Dataset
None.
Unsuitable Licenses
Licenses that are not free, libre, open, even if they may claim to be "open source".
These are not "Wikipedia Commons compatible", for example:
- BigCode OpenRAIL-M.
- Creative Commons Non-commercial (NC).
- Llama 2 Community License Agreement.
- OpenRAIL.
- OpenRAIL++.
- StableCode Research License.
- Falcon 180B TII License.
- Proprietary licenses.
- Any "custom" license that hasn't been reviewed by the general community.
Unsuitable Models
Models that are not free, libre, open, even if they may claim to be "open source".
Unsuitable Model License
The following models are unsuitable due to using an unsuitable license.
Llama 2
The following are unsuitable due to Llama 2 license:
- CodeLlama
- Llama 2
- NexusRaven
- Phind-CodeLlama
- WizardCoder
- Vicuna
- WizardLM
- Wizard-Math
OpenRAIL
The following are unsuitable due to OpenRAIL license:
- CodeUp-Llama
- StarCoder
- WizardCoder
Falcon 180B TII License
The following are unsuitable due to Falcon 180B TII license:
- Falcon-180B
Proprietary
The following are unsuitable due to proprietary licenses:
- Claude
- GPT
- PaLM
- Qwen
Model Table
See models.ods
LibreOffice spreadsheet.
License
Creative Commons Attribution-ShareAlike 4.0 International
Copyright © 2023, Jeff Moe.