Decentralized compute layer

Go to file

Tyler Goodlet e0704e3787 Suggest `skynet.dgpu` docs, typing, pythonisms From the deep-ish dive drafting our first set of design/architecture diagrams in https://github.com/skygpu/cyberdyne/pull/2, this adds a buncha suggestions, typing, and styling adjustments. Namely the code tweaks include, - changing to multi-line import tuples where appropriate (since they're much handier to modify ;) - adding typing in many spots where it wasn't clear to me the types being returned/operated-with in various (internal) methods. - doc strings (in mostly random spots Xp ) where i had the need to remember the impl's purpose but didn't want to re-read the code in detail again. - ALOT of TODOs surrounding various potential style changes, re-factorings, naming and in some cases "modernization" according to the latest python3.12 feats/spec/stdlib.		2025-02-18 15:56:17 -03:00
docker	Switch to uv package manager	2025-01-22 01:18:21 -03:00
skynet	Suggest `skynet.dgpu` docs, typing, pythonisms	2025-02-18 15:56:17 -03:00
tests	Fix minor issues on compute daemon found thanks to tests, vastly improve pipeline_for function and support old diffuse method	2025-01-10 12:33:23 -03:00
.dockerignore	Add venv to dockerignore	2023-10-07 11:01:40 -03:00
.gitignore	Switch to uv package manager	2025-01-22 01:18:21 -03:00
LICENSE	Update LICENSE	2023-04-09 19:50:08 -03:00
README.md	Fix minor errors, update lockfile, add cli entrypoints in pyproject.toml fix README	2025-01-09 17:04:35 -03:00
pyproject.toml	Switch to uv package manager	2025-01-22 01:18:21 -03:00
pytest.ini	Add test for new ipfs async apis, fix cli entrypoints endpoint loading to new format	2023-09-24 15:23:25 -03:00
skynet.toml.example	Drop old ipfs stuff	2025-01-09 14:03:56 -03:00
uv.lock	Switch to uv package manager	2025-01-22 01:18:21 -03:00

README.md

skynet

<img src="https://explorer.skygpu.net/v2/explore/assets/logo.png" width=512 height=512>

decentralized compute platform

native install

system dependencies: - cuda 11.8 - llvm 10 - python 3.10+ - docker (for ipfs node)

# create and edit config from template
cp skynet.toml.example skynet.toml

# install poetry package manager
curl -sSL https://install.python-poetry.org | python3 -

# install
poetry install

# enable environment
poetry shell

# test you can run this command
skynet --help

# to launch worker
skynet run dgpu

dockerized install

frontend

system dependencies: - docker

# create and edit config from template
cp skynet.toml.example skynet.toml

# pull runtime container
docker pull guilledk/skynet:runtime-frontend

# run telegram bot
docker run \
    -it \
    --rm \
    --network host \
    --name skynet-telegram \
    --mount type=bind,source="$(pwd)",target=/root/target \
    guilledk/skynet:runtime-frontend \
    skynet run telegram --db-pass PASSWORD --db-user USER --db-host HOST

worker

system dependencies: - docker with gpu enabled

# create and edit config from template
cp skynet.toml.example skynet.toml

# pull runtime container
docker pull guilledk/skynet:runtime-cuda

# or build it (takes a bit of time)
./build_docker.sh

# run worker with all gpus
docker run \
    -it \
    --rm \
    --gpus all \
    --network host \
    --name skynet-worker \
    --mount type=bind,source="$(pwd)",target=/root/target \
    guilledk/skynet:runtime-cuda \
    skynet run dgpu

# run worker with specific gpu
docker run \
    -it \
    --rm \
    --gpus '"device=1"' \
    --network host \
    --name skynet-worker-1 \
    --mount type=bind,source="$(pwd)",target=/root/target \
    guilledk/skynet:runtime-cuda \
    skynet run dgpu