Blog of Brev.dev, a cloud GPU service, with guides on fine-tuning models, Stable Diffusion and Go development.
HN commenters mostly mention Brev as a way to rent cloud GPUs and use cloud dev environments, with one noting no affiliation.
I feel your pain,
I try to mitigate this as much as possible with templating projects. I use https://gitlab.com/reedrichards/cookiecutter-goapp for small webservers and this generates the infra i need for aws, and go has been a relatively simple development and packaging platform for me.
I am also working at https://brev.dev and have been using their product at work and on personal projects to help standardize my development environment well. I often found myself at other companies constantly fire fighting developer environments falling into entropy and wishing for something like this so we can not having to worry about the different configurations of N developers * M machines.
*edited to make link clickable
This is a huge pain, mostly stemming from using the same computer for different projects, because dependencies start to conflict. Big companies (fb, google, workday) solve this with cloud-hosted dev environments. Cloud environments let me contribute to our code base as a technical founder without having to have the technical expertise of a CTO. I'd rather not go back because while I know how to code, I'd rather not spend time configuring dev environments if someone can just share theirs with me. (disclaimer: I work at https://brev.dev)
> What about alternatives?
Custom models? Apart from the Big 3 (in no particular order):
- https://together.ai/
- https://replicate.com/
- https://anyscale.com/
- https://baseten.co/
- https://modal.com/
- https://banana.dev/
- https://runpod.io/
- https://bentoml.com/
- https://brev.dev/
- https://octo.ai/
- https://cerebrium.ai/
...
R.I.P. battery.
Personally I've been using Brev [1] to do my cloud training, you get a cloud GPU instance that you can upgrade/downgrade on the fly, and makes supports VS Code out of the box.
[1] https://brev.dev/
(I'm not affiliated with Brev)
The approach I see used is axolotl with QLoRA using cloud GPUs which can be quite cheap.
https://github.com/OpenAccess-AI-Collective/axolotl
Someone from one of the cloud GPU vendors wrote a guide: https://brev.dev/blog/fine-tuning-mistral
You should check out https://brev.dev. You can rent GPUs, pause instances, use your own AWS/GCP accounts to make use of your credits, and the CLI lets you use your GPU as if it’s on your local machine.