I wanted to run autoresearch on my Mac, then on a cloud GPU. Spent more time finding the right instance type, patching batch sizes for the GPU's VRAM, setting up SSH, and making sure VMs get torn down, than on the actual research. So I built autoresearch-anycloud so that researchers and enthusiasts can run autoresearch on any public cloud or on Mac without the infrastructure headaches.
It provides three functionality.
1. Lets you run on cloud GPU or Mac with minimal infra knowledge and setup. init gcp + run. It picks a compatible GPU instance, patches batch sizes, sets up SSH, and tears down the VM automatically.
2. Estimates and enforces cost control. Estimates cost upfront and enforces a budget ceiling. One demo experiment: $0.00 (Mac), $0.12 (GCP), $0.17 (AWS).
3. Delivers comprehensive execution logs and results.
https://github.com/abcdedf/autoresearch-anycloud
I was working on a similar tool (autoresearch-anycloud) unaware of this tool. Autofoundry looks to be an impressive tool. Looking at the supported providers, the tools provide complementary coverage of the providers. In autoresearch-anycloud, we focused on Mac and public cloud providers and a variety of cloud GPU instances.
It provides three functionality.