nomic-ai / gpt4all.io

gpt4all.io

gpt4all.io's Issues

Set up CI deployments

just a small typo in models.json

Found a small typo in https://gpt4all.io/models/models.json

commericially != commercially

    "md5sum": "1cfa4958f489f0a0d1ffdf6b37322809",
    "filename": "ggml-mpt-7b-instruct.bin",
    "filesize": "4854401028",
    "requires": "2.4.1",
    "description": "A commericially licensable instruct model based on MPT and trained by Mosaic ML."

Ln:86 Col:36 Sel:13|1

Thanks for all the amazing work y'all do!

Error report: Linux memory GiB and GB

Linux does not calculate the system memory in GB but in GiB. So exactly in binary difinition not in the Microsoft/Apple way of decimal difinition.
With exactly the same system memory of 16GB under Windows, a Linux computer only has <15 GiB. However, the program does not differentiate between GiB and GB, but complains that there is not enough memory available for models with a 16GB requirement.
A quarter of a year ago, the models could still be used, but unfortunately the bug was exacerbated and now the program no longer allows the models with 16GB to be used on a Linux computer with 16GB in GiB 14.896.

GPT4All datalake in Atlas not rendering at gpt4all.io

Has been like this for a few weeks now. I use Mozilla Firefox 119.0.1. Reproducible with Microsoft Edge as well.

Performance benchmarks of website out of sync with current models

Options:

1. Leave the website as is, proudly presenting past glory.
2. Remove benchmarks altogether from the website to avoid maintenance.
3. Update benchmarks. Be honest and in sync with current developments.

The choice is yours.

Bonus option: Add link to documention how users can upload their models to the benchmark. Something like this:

I saw someone mention at huggingface that they upload their model results to gpt4all, so there probably must be documentation about that, but I cannot find that conversation anymore to ask this person. It probably was Teknum.