Add venv to dockerignore

Improve readme
Improve dockerization as ipfs cli exec runs not needed anymore
Fix pyproject toml for gpu workers
Add more sections on example config
Drop and siomplify many cli commands, try to use config.ini for everything now
Use more dynamic imports on cli to speed up startup
Improve model pipelines to allow low mem cards to run big models
Add upscaler download to `skynet download` cmd

.dockerignore

@@ -7,3 +7,4 @@ outputs
 *.egg-info
 **/*.key
 **/*.cert
+.venv

README.md

@@ -2,29 +2,61 @@
 ### decentralized compute platform
 
 To launch a worker:
+
+### native install
+
+system dependencies:
+- `cuda` 11.8
+- `llvm` 10
+- `python` 3.10+
+- `docker` (for ipfs node)
+
 ```
 # create and edit config from template
 cp skynet.ini.example skynet.ini
 
-# create python virtual envoirment 3.10+
+# install poetry package manager
-python3 -m venv venv
+curl -sSL https://install.python-poetry.org | python3 -
 
-# enable envoirment
+# install
-source venv/bin/activate
+poetry install
 
-# install requirements
+# enable environment
-pip install -r requirements.txt
+poetry shell
-pip install -r requirements.cuda.0.txt
-pip install -r requirements.cuda.1.txt
-pip install -r requirements.cuda.2.txt
-
-# install skynet
-pip install -e .
 
 # test you can run this command
 skynet --help
 
+# launch ipfs node
+skynet run ipfs
+
 # to launch worker
 skynet run dgpu
 
 ```
+
+### dockerized install
+
+system dependencies:
+- `docker` with gpu enabled
+
+```
+# pull runtime container
+docker pull guilledk/skynet:runtime-cuda
+
+# or build it (takes a bit of time)
+./build_docker.sh
+
+# launch simple ipfs node
+./launch_ipfs.sh
+
+# run worker
+docker run \
+    -it \
+    --rm \
+    --gpus all \
+    --network host \
+    --mount type=bind,source="$(pwd)",target=/root/target \
+    guilledk/skynet:runtime-cuda \
+    skynet run dgpu
+```

build_docker.sh

@@ -1,7 +1,3 @@
 docker build \
-    -t skynet:runtime-cuda \
+    -t guilledk/skynet:runtime-cuda \
     -f docker/Dockerfile.runtime+cuda .
-
-docker build \
-    -t skynet:runtime \
-    -f docker/Dockerfile.runtime .

docker/Dockerfile.runtime

@@ -2,15 +2,14 @@ from python:3.11
 
 env DEBIAN_FRONTEND=noninteractive
 
+run apt-get update && apt-get install -y git
+
+run curl -sSL https://install.python-poetry.org | python3 -
+
+env PATH "/root/.local/bin:$PATH"
+
+copy . /skynet
+
 workdir /skynet
 
-copy requirements.txt requirements.txt
+run poetry install -v
-copy pytest.ini ./
-copy setup.py ./
-copy skynet ./skynet
-
-run pip install \
-    -e . \
-    -r requirements.txt
-
-copy tests ./

docker/Dockerfile.runtime+cuda

@@ -1,29 +1,46 @@
-from nvidia/cuda:11.7.0-devel-ubuntu20.04
+from nvidia/cuda:11.8.0-devel-ubuntu20.04
 from python:3.11
 
 env DEBIAN_FRONTEND=noninteractive
 
-run apt-get update && \
+run apt-get update && apt-get install -y \
-    apt-get install -y ffmpeg libsm6 libxext6
+    git \
+    clang \
+    cmake \
+    ffmpeg \
+    libsm6 \
+    libxext6 \
+    ninja-build
+
+env CC /usr/bin/clang
+env CXX /usr/bin/clang++
+
+# install llvm10 as required by llvm-lite
+run git clone https://github.com/llvm/llvm-project.git -b llvmorg-10.0.1
+workdir /llvm-project
+# this adds a commit from 12.0.0 that fixes build on newer compilers
+run git cherry-pick -n b498303066a63a203d24f739b2d2e0e56dca70d1
+run cmake -S llvm -B build -G Ninja -DCMAKE_BUILD_TYPE=Release
+run ninja -C build install  # -j8
+
+run curl -sSL https://install.python-poetry.org | python3 -
+
+env PATH "/root/.local/bin:$PATH"
+
+copy . /skynet
 
 workdir /skynet
 
-copy requirements.cuda* ./
+env POETRY_VIRTUALENVS_PATH /skynet/.venv
 
-run pip install -U pip ninja
+run poetry install --with=cuda -v
-run pip install -v -r requirements.cuda.0.txt
-run pip install -v -r requirements.cuda.1.txt
-run pip install -v -r requirements.cuda.2.txt
 
-copy requirements.txt requirements.txt
+workdir /root/target
-copy pytest.ini pytest.ini
-copy setup.py setup.py
-copy skynet skynet
-
-run pip install -e . -r requirements.txt
 
 env PYTORCH_CUDA_ALLOC_CONF max_split_size_mb:128
 env NVIDIA_VISIBLE_DEVICES=all
-env HF_HOME /hf_home
 
-copy tests tests
+copy docker/entrypoint.sh /entrypoint.sh
+entrypoint ["/entrypoint.sh"]
+
+cmd ["skynet", "--help"]

docker/entrypoint.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+
+export VIRTUAL_ENV='/skynet/.venv'
+poetry env use $VIRTUAL_ENV/bin/python
+
+exec "$@"

launch_ipfs.sh

@@ -0,0 +1,36 @@
+#!/bin/bash
+
+name='skynet-ipfs'
+peers=("$@")
+
+data_dir="$(pwd)/ipfs-docker-data"
+data_target='/data/ipfs'
+
+# Create data directory if it doesn't exist
+mkdir -p "$data_dir"
+
+# Run the container
+docker run -d \
+  --name "$name" \
+  -p 8080:8080/tcp \
+  -p 4001:4001/tcp \
+  -p 127.0.0.1:5001:5001/tcp \
+  --mount type=bind,source="$data_dir",target="$data_target" \
+  --rm \
+  ipfs/go-ipfs:latest
+
+# Change ownership
+docker exec "$name" chown 1000:1000 -R "$data_target"
+
+# Wait for Daemon to be ready
+while read -r log; do
+  echo "$log"
+  if [[ "$log" == *"Daemon is ready"* ]]; then
+    break
+  fi
+done < <(docker logs -f "$name")
+
+# Connect to peers
+for peer in "${peers[@]}"; do
+  docker exec "$name" ipfs swarm connect "$peer" || echo "Error connecting to peer: $peer"
+done

pyproject.toml

@@ -11,10 +11,10 @@ python = '>=3.10,<3.12'
 pytz = '^2023.3.post1'
 trio = '^0.22.2'
 asks = '^3.0.0'
-numpy = '^1.26.0'
 Pillow = '^10.0.1'
 docker = '^6.1.3'
 py-leap = {git = 'https://github.com/guilledk/py-leap.git', rev = 'v0.1a14'}
+xformers = "^0.0.22"
 
 [tool.poetry.group.frontend]
 optional = true
@@ -37,12 +37,14 @@ pytest = {version = '^7.4.2'}
 optional = true
 
 [tool.poetry.group.cuda.dependencies]
-torch = {version = '1.13.0+cu117', source = 'torch'}
+torch = {version = '2.0.1+cu118', source = 'torch'}
 scipy = {version = '^1.11.2'}
-triton = {version = '^2.1.0'}
+numba = {version = '0.57.0'}
+triton = {version = '2.0.0', source = 'torch'}
 basicsr = {version = '^1.4.2'}
 diffusers = {version = '^0.21.2'}
 realesrgan = {version = '^0.3.0'}
+torchvision = {version = '0.15.2+cu118', source = 'torch'}
 accelerate = {version = '^0.23.0'}
 transformers = {version = '^4.33.2'}
 huggingface-hub = {version = '^0.17.3'}
@@ -51,9 +53,12 @@ invisible-watermark = {version = '^0.2.0'}
 
 [[tool.poetry.source]]
 name = 'torch'
-url = 'https://download.pytorch.org/whl/cu117'
+url = 'https://download.pytorch.org/whl/cu118'
 priority = 'explicit'
 
 [build-system]
-requires = ['poetry-core']
+requires = ['poetry-core', 'cython']
 build-backend = 'poetry.core.masonry.api'
+
+[tool.poetry.scripts]
+skynet = 'skynet.cli:skynet'

skynet.ini.example

@@ -1,3 +1,6 @@
+# config sections are optional, depending on which services
+# you wish to run
+
 [skynet.dgpu]
 account = testworkerX
 permission = active
@@ -36,3 +39,13 @@ ipfs_gateway_url = /ip4/169.197.140.154/tcp/4001/p2p/12D3KooWKWogLFNEcNNMKnzU7Sn
 ipfs_url = http://127.0.0.1:5001
 
 token = XXXXXXXXXX:xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+
+[skynet.pinner]
+hyperion_url = https://testnet.skygpu.net
+ipfs_url = http://127.0.0.1:5001
+
+[skynet.user]
+account = testuser
+permission = active
+key = 5Xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+node_url = https://testnet.skygpu.net

skynet/cli.py

@@ -6,25 +6,12 @@ import random
 
 from functools import partial
 
-import trio
-import asks
 import click
-import asyncio
-import requests
 
-from leap.cleos import CLEOS
+from leap.sugar import Name, asset_from_str
-from leap.sugar import collect_stdout, Name, asset_from_str
-from leap.hyperion import HyperionAPI
 
-from skynet.ipfs import AsyncIPFSHTTP
-
-
-from .db import open_new_database
 from .config import *
-from .nodeos import open_cleos, open_nodeos
 from .constants import *
-from .frontend.telegram import SkynetTelegramFrontend
-from .frontend.discord import SkynetDiscordFrontend
 
 
 @click.group()
@@ -44,7 +31,9 @@ def skynet(*args, **kwargs):
 @click.option('--seed', '-S', default=None)
 def txt2img(*args, **kwargs):
     from . import utils
-    _, hf_token, _, _ = init_env_from_config()
+
+    config = load_skynet_ini()
+    hf_token = load_key(config, 'skynet.dgpu', 'hf_token')
     utils.txt2img(hf_token, **kwargs)
 
 @click.command()
@@ -59,7 +48,8 @@ def txt2img(*args, **kwargs):
 @click.option('--seed', '-S', default=None)
 def img2img(model, prompt, input, output, strength, guidance, steps, seed):
     from . import utils
-    _, hf_token, _, _ = init_env_from_config()
+    config = load_skynet_ini()
+    hf_token = load_key(config, 'skynet.dgpu', 'hf_token')
     utils.img2img(
         hf_token,
         model=model,
@@ -87,47 +77,42 @@ def upscale(input, output, model):
 @skynet.command()
 def download():
     from . import utils
-    _, hf_token, _, _ = init_env_from_config()
+    config = load_skynet_ini()
+    hf_token = load_key(config, 'skynet.dgpu', 'hf_token')
     utils.download_all_models(hf_token)
 
 @skynet.command()
-@click.option(
-    '--account', '-A', default=None)
-@click.option(
-    '--permission', '-P', default=None)
-@click.option(
-    '--key', '-k', default=None)
-@click.option(
-    '--node-url', '-n', default='https://skynet.ancap.tech')
 @click.option(
     '--reward', '-r', default='20.0000 GPU')
 @click.option('--jobs', '-j', default=1)
-@click.option('--model', '-m', default='prompthero/openjourney')
+@click.option('--model', '-m', default='stabilityai/stable-diffusion-xl-base-1.0')
 @click.option(
     '--prompt', '-p', default='a red old tractor in a sunny wheat field')
 @click.option('--output', '-o', default='output.png')
-@click.option('--width', '-w', default=512)
+@click.option('--width', '-w', default=1024)
-@click.option('--height', '-h', default=512)
+@click.option('--height', '-h', default=1024)
 @click.option('--guidance', '-g', default=10)
 @click.option('--step', '-s', default=26)
 @click.option('--seed', '-S', default=None)
 @click.option('--upscaler', '-U', default='x4')
 @click.option('--binary_data', '-b', default='')
 def enqueue(
-    account: str,
-    permission: str,
-    key: str | None,
-    node_url: str,
     reward: str,
     jobs: int,
     **kwargs
 ):
-    key, account, permission = load_account_info(
+    import trio
-        'user', key, account, permission)
+    from leap.cleos import CLEOS
 
-    node_url, _, _, _ = load_endpoint_info('user', node_url=node_url)
+    config = load_skynet_ini()
+
+    key = load_key(config, 'skynet.user', 'key')
+    account = load_key(config, 'skynet.user', 'account')
+    permission = load_key(config, 'skynet.user', 'permission')
+    node_url = load_key(config, 'skynet.user', 'node_url')
+
+    cleos = CLEOS(None, None, url=node_url, remote=node_url)
 
-    with open_cleos(node_url, key=key) as cleos:
     async def enqueue_n_jobs():
         for i in range(jobs):
             if not kwargs['seed']:
@@ -152,30 +137,23 @@ def enqueue(
                 account, key, permission,
             )
             print(res)
+
     trio.run(enqueue_n_jobs)
 
 
 @skynet.command()
 @click.option('--loglevel', '-l', default='INFO', help='Logging level')
-@click.option(
-    '--account', '-A', default='telos.gpu')
-@click.option(
-    '--permission', '-P', default='active')
-@click.option(
-    '--key', '-k', default=None)
-@click.option(
-    '--node-url', '-n', default='https://skynet.ancap.tech')
 def clean(
     loglevel: str,
-    account: str,
-    permission: str,
-    key: str | None,
-    node_url: str,
 ):
-    key, account, permission = load_account_info(
+    import trio
-        'user', key, account, permission)
+    from leap.cleos import CLEOS
 
-    node_url, _, _, _ = load_endpoint_info('user', node_url=node_url)
+    config = load_skynet_ini()
+    key = load_key(config, 'skynet.user', 'key')
+    account = load_key(config, 'skynet.user', 'account')
+    permission = load_key(config, 'skynet.user', 'permission')
+    node_url = load_key(config, 'skynet.user', 'node_url')
 
     logging.basicConfig(level=loglevel)
     cleos = CLEOS(None, None, url=node_url, remote=node_url)
@@ -190,10 +168,10 @@ def clean(
     )
 
 @skynet.command()
-@click.option(
+def queue():
-    '--node-url', '-n', default='https://skynet.ancap.tech')
+    import requests
-def queue(node_url: str):
+    config = load_skynet_ini()
-    node_url, _, _, _ = load_endpoint_info('user', node_url=node_url)
+    node_url = load_key(config, 'skynet.user', 'node_url')
     resp = requests.post(
         f'{node_url}/v1/chain/get_table_rows',
         json={
@@ -206,11 +184,11 @@ def queue(node_url: str):
     print(json.dumps(resp.json(), indent=4))
 
 @skynet.command()
-@click.option(
-    '--node-url', '-n', default='https://skynet.ancap.tech')
 @click.argument('request-id')
-def status(node_url: str, request_id: int):
+def status(request_id: int):
-    node_url, _, _, _ = load_endpoint_info('user', node_url=node_url)
+    import requests
+    config = load_skynet_ini()
+    node_url = load_key(config, 'skynet.user', 'node_url')
     resp = requests.post(
         f'{node_url}/v1/chain/get_table_rows',
         json={
@@ -223,26 +201,16 @@ def status(node_url: str, request_id: int):
     print(json.dumps(resp.json(), indent=4))
 
 @skynet.command()
-@click.option(
-    '--account', '-a', default='telegram')
-@click.option(
-    '--permission', '-p', default='active')
-@click.option(
-    '--key', '-k', default=None)
-@click.option(
-    '--node-url', '-n', default='https://skynet.ancap.tech')
 @click.argument('request-id')
-def dequeue(
+def dequeue(request_id: int):
-    account: str,
+    import trio
-    permission: str,
+    from leap.cleos import CLEOS
-    key: str | None,
-    node_url: str,
-    request_id: int
-):
-    key, account, permission = load_account_info(
-        'user', key, account, permission)
 
-    node_url, _, _, _ = load_endpoint_info('user', node_url=node_url)
+    config = load_skynet_ini()
+    key = load_key(config, 'skynet.user', 'key')
+    account = load_key(config, 'skynet.user', 'account')
+    permission = load_key(config, 'skynet.user', 'permission')
+    node_url = load_key(config, 'skynet.user', 'node_url')
 
     cleos = CLEOS(None, None, url=node_url, remote=node_url)
     res = trio.run(
@@ -261,29 +229,24 @@ def dequeue(
 
 
 @skynet.command()
-@click.option(
-    '--account', '-a', default='telos.gpu')
-@click.option(
-    '--permission', '-p', default='active')
-@click.option(
-    '--key', '-k', default=None)
-@click.option(
-    '--node-url', '-n', default='https://skynet.ancap.tech')
 @click.option(
     '--token-contract', '-c', default='eosio.token')
 @click.option(
     '--token-symbol', '-S', default='4,GPU')
 def config(
-    account: str,
-    permission: str,
-    key: str | None,
-    node_url: str,
     token_contract: str,
     token_symbol: str
 ):
-    key, account, permission = load_account_info(
+    import trio
-        'user', key, account, permission)
+    from leap.cleos import CLEOS
-    node_url, _, _, _ = load_endpoint_info('user', node_url=node_url)
+
+    config = load_skynet_ini()
+
+    key = load_key(config, 'skynet.user', 'key')
+    account = load_key(config, 'skynet.user', 'account')
+    permission = load_key(config, 'skynet.user', 'permission')
+    node_url = load_key(config, 'skynet.user', 'node_url')
+
     cleos = CLEOS(None, None, url=node_url, remote=node_url)
     res = trio.run(
         partial(
@@ -301,26 +264,18 @@ def config(
 
 
 @skynet.command()
-@click.option(
-    '--account', '-a', default='telegram')
-@click.option(
-    '--permission', '-p', default='active')
-@click.option(
-    '--key', '-k', default=None)
-@click.option(
-    '--node-url', '-n', default='https://skynet.ancap.tech')
 @click.argument('quantity')
-def deposit(
+def deposit(quantity: str):
-    account: str,
+    import trio
-    permission: str,
+    from leap.cleos import CLEOS
-    key: str | None,
-    node_url: str,
-    quantity: str
-):
-    key, account, permission = load_account_info(
-        'user', key, account, permission)
 
-    node_url, _, _, _ = load_endpoint_info('user', node_url=node_url)
+    config = load_skynet_ini()
+
+    key = load_key(config, 'skynet.user', 'key')
+    account = load_key(config, 'skynet.user', 'account')
+    permission = load_key(config, 'skynet.user', 'permission')
+    node_url = load_key(config, 'skynet.user', 'node_url')
+    cleos = CLEOS(None, None, url=node_url, remote=node_url)
 
     res = trio.run(
         partial(
@@ -345,6 +300,8 @@ def run(*args, **kwargs):
 
 @run.command()
 def db():
+    from .db import open_new_database
+
     logging.basicConfig(level=logging.INFO)
     with open_new_database(cleanup=False) as db_params:
         container, passwd, host = db_params
@@ -352,6 +309,8 @@ def db():
 
 @run.command()
 def nodeos():
+    from .nodeos import open_nodeos
+
     logging.basicConfig(filename='skynet-nodeos.log', level=logging.INFO)
     with open_nodeos(cleanup=False):
         ...
@@ -364,6 +323,7 @@ def dgpu(
     loglevel: str,
     config_path: str
 ):
+    import trio
     from .dgpu import open_dgpu_node
 
     logging.basicConfig(level=loglevel)
@@ -377,20 +337,6 @@ def dgpu(
 
 @run.command()
 @click.option('--loglevel', '-l', default='INFO', help='logging level')
-@click.option(
-    '--account', '-a', default='telegram')
-@click.option(
-    '--permission', '-p', default='active')
-@click.option(
-    '--key', '-k', default=None)
-@click.option(
-    '--hyperion-url', '-y', default=f'https://testnet.{DEFAULT_DOMAIN}')
-@click.option(
-    '--node-url', '-n', default=f'https://testnet.{DEFAULT_DOMAIN}')
-@click.option(
-    '--ipfs-url', '-i', default=DEFAULT_IPFS_LOCAL)
-@click.option(
-    '--ipfs-gateway-url', '-I', default=None)
 @click.option(
     '--db-host', '-h', default='localhost:5432')
 @click.option(
@@ -399,26 +345,26 @@ def dgpu(
     '--db-pass', '-u', default='password')
 def telegram(
     loglevel: str,
-    account: str,
-    permission: str,
-    key: str | None,
-    hyperion_url: str,
-    ipfs_url: str,
-    ipfs_gateway_url: str,
-    node_url: str,
     db_host: str,
     db_user: str,
     db_pass: str
 ):
+    import asyncio
+    from .frontend.telegram import SkynetTelegramFrontend
+
     logging.basicConfig(level=loglevel)
 
-    _, _, tg_token, _ = init_env_from_config()
+    config = load_skynet_ini()
+    tg_token = load_key(config, 'skynet.telegram', 'tg_token')
 
-    key, account, permission = load_account_info(
+    key = load_key(config, 'skynet.user', 'key')
-        'telegram', key, account, permission)
+    account = load_key(config, 'skynet.user', 'account')
+    permission = load_key(config, 'skynet.user', 'permission')
+    node_url = load_key(config, 'skynet.user', 'node_url')
+    hyperion_url = load_key(config, 'skynet.telegram', 'hyperion_url')
 
-    node_url, _, ipfs_gateway_url, ipfs_url = load_endpoint_info(
+    ipfs_gateway_url = load_key(config, 'skynet.telegram', 'ipfs_gateway_url')
-        'telegram', node_url=node_url, ipfs_gateway_url=ipfs_gateway_url)
+    ipfs_url = load_key(config, 'skynet.telegram', 'ipfs_url')
 
     async def _async_main():
         frontend = SkynetTelegramFrontend(
@@ -442,20 +388,6 @@ def telegram(
 
 @run.command()
 @click.option('--loglevel', '-l', default='INFO', help='logging level')
-@click.option(
-    '--account', '-a', default='discord')
-@click.option(
-    '--permission', '-p', default='active')
-@click.option(
-    '--key', '-k', default=None)
-@click.option(
-    '--hyperion-url', '-y', default=f'https://testnet.{DEFAULT_DOMAIN}')
-@click.option(
-    '--node-url', '-n', default=f'https://testnet.{DEFAULT_DOMAIN}')
-@click.option(
-    '--ipfs-url', '-i', default=DEFAULT_IPFS_LOCAL)
-@click.option(
-    '--ipfs-gateway-url', '-I', default=DEFAULT_IPFS_REMOTE)
 @click.option(
     '--db-host', '-h', default='localhost:5432')
 @click.option(
@@ -464,26 +396,26 @@ def telegram(
     '--db-pass', '-u', default='password')
 def discord(
     loglevel: str,
-    account: str,
-    permission: str,
-    key: str | None,
-    hyperion_url: str,
-    ipfs_url: str,
-    ipfs_gateway_url: str,
-    node_url: str,
     db_host: str,
     db_user: str,
     db_pass: str
 ):
+    import asyncio
+    from .frontend.discord import SkynetDiscordFrontend
+
     logging.basicConfig(level=loglevel)
 
-    _, _, _, dc_token = init_env_from_config()
+    config = load_skynet_ini()
+    dc_token = load_key(config, 'skynet.discord', 'dc_token')
 
-    key, account, permission = load_account_info(
+    key = load_key(config, 'skynet.discord', 'key')
-        'discord', key, account, permission)
+    account = load_key(config, 'skynet.discord', 'account')
+    permission = load_key(config, 'skynet.discord', 'permission')
+    node_url = load_key(config, 'skynet.discord', 'node_url')
+    hyperion_url = load_key(config, 'skynet.discord', 'hyperion_url')
 
-    node_url, _, ipfs_gateway_url, ipfs_url = load_endpoint_info(
+    ipfs_gateway_url = load_key(config, 'skynet.discord', 'ipfs_gateway_url')
-        'discord', node_url=node_url, ipfs_gateway_url=ipfs_gateway_url)
+    ipfs_url = load_key(config, 'skynet.discord', 'ipfs_url')
 
     async def _async_main():
         frontend = SkynetDiscordFrontend(
@@ -507,24 +439,28 @@ def discord(
 @run.command()
 @click.option('--loglevel', '-l', default='INFO', help='logging level')
 @click.option('--name', '-n', default='skynet-ipfs', help='container name')
-def ipfs(loglevel, name):
+@click.option('--peer', '-p', default=(), help='connect to peer', multiple=True, type=str)
+def ipfs(loglevel, name, peer):
     from skynet.ipfs.docker import open_ipfs_node
 
     logging.basicConfig(level=loglevel)
-    with open_ipfs_node(name=name):
+    with open_ipfs_node(name=name, peers=peer):
         ...
 
 @run.command()
 @click.option('--loglevel', '-l', default='INFO', help='logging level')
-@click.option(
+def pinner(loglevel):
-    '--ipfs-rpc', '-i', default='http://127.0.0.1:5001')
+    import trio
-@click.option(
+    from leap.hyperion import HyperionAPI
-    '--hyperion-url', '-y', default='http://127.0.0.1:42001')
+    from .ipfs import AsyncIPFSHTTP
-def pinner(loglevel, ipfs_rpc, hyperion_url):
     from .ipfs.pinner import SkynetPinner
 
+    config = load_skynet_ini()
+    hyperion_url = load_key(config, 'skynet.pinner', 'hyperion_url')
+    ipfs_url = load_key(config, 'skynet.pinner', 'ipfs_url')
+
     logging.basicConfig(level=loglevel)
-    ipfs_node = AsyncIPFSHTTP(ipfs_rpc)
+    ipfs_node = AsyncIPFSHTTP(ipfs_url)
     hyperion = HyperionAPI(hyperion_url)
 
     pinner = SkynetPinner(hyperion, ipfs_node)

skynet/config.py

@@ -1,117 +1,30 @@
 #!/usr/bin/python
 
-import os
-import json
-
-from pathlib import Path
 from configparser import ConfigParser
-from re import sub
 
 from .constants import DEFAULT_CONFIG_PATH
 
 
+class ConfigParsingError(BaseException):
+    ...
+
+
 def load_skynet_ini(
     file_path=DEFAULT_CONFIG_PATH
-):
+) -> ConfigParser:
     config = ConfigParser()
     config.read(file_path)
 
     return config
 
 
-def init_env_from_config(
+def load_key(config: ConfigParser, section: str, key: str) -> str:
-    hf_token: str | None = None,
+    if section not in config:
-    hf_home: str | None = None,
+        conf_sections = [s for s in config]
-    tg_token: str | None = None,
+        raise ConfigParsingError(f'section \"{section}\" not in {conf_sections}')
-    dc_token: str | None = None,
-    file_path=DEFAULT_CONFIG_PATH
-):
-    config = load_skynet_ini(file_path=file_path)
 
-    if 'HF_TOKEN' in os.environ:
+    if key not in config[section]:
-        hf_token = os.environ['HF_TOKEN']
+        conf_keys = [k for k in config[section]]
+        raise ConfigParsingError(f'key \"{key}\" not in {conf_keys}')
 
-    elif 'skynet.dgpu' in config:
+    return str(config[section][key])
-        sub_config = config['skynet.dgpu']
-        if 'hf_token' in sub_config:
-            hf_token = sub_config['hf_token']
-            os.environ['HF_TOKEN'] = hf_token
-
-    if 'HF_HOME' in os.environ:
-        hf_home = os.environ['HF_HOME']
-
-    elif 'skynet.dgpu' in config:
-        sub_config = config['skynet.dgpu']
-        if 'hf_home' in sub_config:
-            hf_home = sub_config['hf_home']
-            os.environ['HF_HOME'] = hf_home
-
-    if 'TG_TOKEN' in os.environ:
-        tg_token = os.environ['TG_TOKEN']
-    elif 'skynet.telegram' in config:
-        sub_config = config['skynet.telegram']
-        if 'token' in sub_config:
-            tg_token = sub_config['token']
-
-    if 'DC_TOKEN' in os.environ:
-        dc_token = os.environ['DC_TOKEN']
-    elif 'skynet.discord' in config:
-        sub_config = config['skynet.discord']
-        if 'token' in sub_config:
-            dc_token = sub_config['token']
-
-    return hf_home, hf_token, tg_token, dc_token
-
-
-def load_account_info(
-    _type: str,
-    key: str | None = None,
-    account: str | None = None,
-    permission: str | None = None,
-    file_path=DEFAULT_CONFIG_PATH
-):
-    config = load_skynet_ini(file_path=file_path)
-
-    type_key = f'skynet.{_type}'
-
-    if type_key in config:
-        sub_config = config[type_key]
-        if not key and 'key' in sub_config:
-            key = sub_config['key']
-
-        if not account and 'account' in sub_config:
-            account = sub_config['account']
-
-        if not permission and 'permission' in sub_config:
-            permission = sub_config['permission']
-
-    return key, account, permission
-
-
-def load_endpoint_info(
-    _type: str,
-    node_url: str | None = None,
-    hyperion_url: str | None = None,
-    ipfs_url: str | None = None,
-    ipfs_gateway_url: str | None = None,
-    file_path=DEFAULT_CONFIG_PATH
-):
-    config = load_skynet_ini(file_path=file_path)
-
-    type_key = f'skynet.{_type}'
-
-    if type_key in config:
-        sub_config = config[type_key]
-        if not node_url and 'node_url' in sub_config:
-            node_url = sub_config['node_url']
-
-        if not hyperion_url and 'hyperion_url' in sub_config:
-            hyperion_url = sub_config['hyperion_url']
-
-        if not ipfs_url and 'ipfs_url' in sub_config:
-            ipfs_url = sub_config['ipfs_url']
-
-        if not ipfs_gateway_url and 'ipfs_gateway_url' in sub_config:
-            ipfs_gateway_url = sub_config['ipfs_gateway_url']
-
-    return node_url, hyperion_url, ipfs_gateway_url, ipfs_url

skynet/constants.py

@@ -5,18 +5,18 @@ VERSION = '0.1a11'
 DOCKER_RUNTIME_CUDA = 'skynet:runtime-cuda'
 
 MODELS = {
-    'prompthero/openjourney':                   { 'short': 'midj'},
+    'prompthero/openjourney':                   {'short': 'midj',        'mem': 8},
-    'runwayml/stable-diffusion-v1-5':           { 'short': 'stable'},
+    'runwayml/stable-diffusion-v1-5':           {'short': 'stable',      'mem': 8},
-    'stabilityai/stable-diffusion-2-1-base':    { 'short': 'stable2'},
+    'stabilityai/stable-diffusion-2-1-base':    {'short': 'stable2',     'mem': 8},
-    'snowkidy/stable-diffusion-xl-base-0.9':    { 'short': 'stablexl0.9'},
+    'snowkidy/stable-diffusion-xl-base-0.9':    {'short': 'stablexl0.9', 'mem': 24},
-    'stabilityai/stable-diffusion-xl-base-1.0': { 'short': 'stablexl'},
+    'stabilityai/stable-diffusion-xl-base-1.0': {'short': 'stablexl',    'mem': 24},
-    'Linaqruf/anything-v3.0':                   { 'short': 'hdanime'},
+    'Linaqruf/anything-v3.0':                   {'short': 'hdanime',     'mem': 8},
-    'hakurei/waifu-diffusion':                  { 'short': 'waifu'},
+    'hakurei/waifu-diffusion':                  {'short': 'waifu',       'mem': 8},
-    'nitrosocke/Ghibli-Diffusion':              { 'short': 'ghibli'},
+    'nitrosocke/Ghibli-Diffusion':              {'short': 'ghibli',      'mem': 8},
-    'dallinmackay/Van-Gogh-diffusion':          { 'short': 'van-gogh'},
+    'dallinmackay/Van-Gogh-diffusion':          {'short': 'van-gogh',    'mem': 8},
-    'lambdalabs/sd-pokemon-diffusers':          { 'short': 'pokemon'},
+    'lambdalabs/sd-pokemon-diffusers':          {'short': 'pokemon',     'mem': 8},
-    'Envvi/Inkpunk-Diffusion':                  { 'short': 'ink'},
+    'Envvi/Inkpunk-Diffusion':                  {'short': 'ink',         'mem': 8},
-    'nousr/robo-diffusion':                     { 'short': 'robot'}
+    'nousr/robo-diffusion':                     {'short': 'robot',       'mem': 8}
 }
 
 SHORT_NAMES = [
@@ -165,8 +165,7 @@ DEFAULT_UPSCALER = None
 DEFAULT_CONFIG_PATH = 'skynet.ini'
 
 DEFAULT_INITAL_MODELS = [
-    'prompthero/openjourney',
+    'stabilityai/stable-diffusion-xl-base-1.0'
-    'runwayml/stable-diffusion-v1-5'
 ]
 
 DATE_FORMAT = '%B the %dth %Y, %H:%M:%S'
@@ -189,3 +188,5 @@ DEFAULT_IPFS_LOCAL = 'http://127.0.0.1:5001'
 
 TG_MAX_WIDTH = 1280
 TG_MAX_HEIGHT = 1280
+
+DEFAULT_SINGLE_CARD_MAP = 'cuda:0'

skynet/dgpu/__init__.py

@@ -1,7 +1,5 @@
 #!/usr/bin/python
 
-import trio
-
 from skynet.dgpu.compute import SkynetMM
 from skynet.dgpu.daemon import SkynetDGPUDaemon
 from skynet.dgpu.network import SkynetGPUConnector
@@ -11,6 +9,4 @@ async def open_dgpu_node(config: dict):
     conn = SkynetGPUConnector(config)
     mm = SkynetMM(config)
 
-    async with conn.open() as conn:
+    await SkynetDGPUDaemon(mm, conn, config).serve_forever()
-        await (SkynetDGPUDaemon(mm, conn, config)
-            .serve_forever())

skynet/dgpu/network.py

@@ -10,8 +10,6 @@ import logging
 import asks
 from PIL import Image
 
-from contextlib import asynccontextmanager as acm
-
 from leap.cleos import CLEOS
 from leap.sugar import Checksum256, Name, asset_from_str
 from skynet.constants import DEFAULT_DOMAIN
@@ -26,7 +24,9 @@ async def failable(fn: partial, ret_fail=None):
 
     except (
         asks.errors.RequestTimeout,
-        json.JSONDecodeError
+        asks.errors.BadHttpResponse,
+        json.JSONDecodeError,
+        OSError
     ):
         return ret_fail
 
@@ -193,11 +193,11 @@ class SkynetGPUConnector:
         )
 
     # IPFS helpers
-
     async def publish_on_ipfs(self, raw_img: bytes):
+        Path('ipfs-staging').mkdir(exist_ok=True)
         logging.info('publish_on_ipfs')
         img = Image.open(io.BytesIO(raw_img))
-        img.save('ipfs-docker-staging/image.png')
+        img.save('ipfs-staging/image.png')
 
         # check peer connections, reconnect to skynet gateway if not
         gateway_id = Path(self.ipfs_gateway_url).name
@@ -205,7 +205,7 @@ class SkynetGPUConnector:
         if gateway_id not in [p['Peer'] for p in peers]:
             await self.ipfs_client.connect(self.ipfs_gateway_url)
 
-        file_info = await self.ipfs_client.add(Path('ipfs-docker-staging/image.png'))
+        file_info = await self.ipfs_client.add(Path('ipfs-staging/image.png'))
         file_cid = file_info['Hash']
 
         await self.ipfs_client.pin(file_cid)

skynet/ipfs/docker.py

@@ -1,6 +1,5 @@
 #!/usr/bin/python
 
-import os
 import sys
 import logging
 
@@ -10,48 +9,14 @@ from contextlib import contextmanager as cm
 import docker
 
 from docker.types import Mount
-from docker.models.containers import Container
-
-
-class IPFSDocker:
-
-    def __init__(self, container: Container):
-        self._container = container
-
-    def add(self, file: str) -> str:
-        ec, out = self._container.exec_run(
-            ['ipfs', 'add', '-w', f'/export/{file}', '-Q'])
-        if ec != 0:
-            logging.error(out)
-        assert ec == 0
-
-        return out.decode().rstrip()
-
-    def pin(self, ipfs_hash: str):
-        ec, _ = self._container.exec_run(
-            ['ipfs', 'pin', 'add', ipfs_hash])
-        assert ec == 0
-
-    def connect(self, remote_node: str):
-        ec, out = self._container.exec_run(
-            ['ipfs', 'swarm', 'connect', remote_node])
-        if ec != 0:
-            logging.error(out)
-
-        assert ec == 0
-
-    def check_connect(self):
-        ec, out = self._container.exec_run(
-            ['ipfs', 'swarm', 'peers'])
-        if ec != 0:
-            logging.error(out)
-        assert ec == 0
-
-        return out.splitlines()
 
 
 @cm
-def open_ipfs_node(name='skynet-ipfs', teardown=False):
+def open_ipfs_node(
+    name: str = 'skynet-ipfs',
+    teardown: bool = False,
+    peers: list[str] = []
+):
     dclient = docker.from_env()
 
     container = None
@@ -59,13 +24,9 @@ def open_ipfs_node(name='skynet-ipfs', teardown=False):
         container = dclient.containers.get(name)
 
     except docker.errors.NotFound:
-        staging_dir = Path().resolve() / 'ipfs-docker-staging'
-        staging_dir.mkdir(parents=True, exist_ok=True)
-
         data_dir = Path().resolve() / 'ipfs-docker-data'
         data_dir.mkdir(parents=True, exist_ok=True)
 
-        export_target = '/export'
         data_target = '/data/ipfs'
 
         container = dclient.containers.run(
@@ -77,19 +38,15 @@ def open_ipfs_node(name='skynet-ipfs', teardown=False):
                 '5001/tcp': ('127.0.0.1', 5001)
             },
             mounts=[
-                Mount(export_target, str(staging_dir), 'bind'),
                 Mount(data_target, str(data_dir), 'bind')
             ],
             detach=True,
             remove=True
         )
 
+        uid, gid = 1000, 1000
+
         if sys.platform != 'win32':
-            uid = os.getuid()
-            gid = os.getgid()
-            ec, out = container.exec_run(['chown', f'{uid}:{gid}', '-R', export_target])
-            logging.info(out)
-            assert ec == 0
             ec, out = container.exec_run(['chown', f'{uid}:{gid}', '-R', data_target])
             logging.info(out)
             assert ec == 0
@@ -100,7 +57,13 @@ def open_ipfs_node(name='skynet-ipfs', teardown=False):
             if 'Daemon is ready' in log:
                 break
 
-    yield IPFSDocker(container)
+    for peer in peers:
+        ec, out = container.exec_run(
+            ['ipfs', 'swarm', 'connect', peer])
+        if ec != 0:
+            logging.error(out)
+
+    yield
 
     if teardown and container:
         container.stop()

skynet/nodeos.py

@@ -4,44 +4,12 @@ import json
 import time
 import logging
 
-from datetime import datetime
 from contextlib import contextmanager as cm
 
 import docker
 
-from pytz import timezone
+from leap.cleos import CLEOS
-from leap.cleos import CLEOS, default_nodeos_image
+from leap.sugar import get_container, Symbol
-from leap.sugar import get_container, Symbol, random_string
-
-
-@cm
-def open_cleos(
-    node_url: str,
-    key: str | None
-):
-    vtestnet = None
-    try:
-        dclient = docker.from_env()
-        vtestnet = get_container(
-            dclient,
-            default_nodeos_image(),
-            name=f'skynet-wallet-{random_string(size=8)}',
-            force_unique=True,
-            detach=True,
-            network='host',
-            remove=True)
-
-        cleos = CLEOS(dclient, vtestnet, url=node_url, remote=node_url)
-
-        if key:
-            cleos.setup_wallet(key)
-
-        yield cleos
-
-    finally:
-        if vtestnet:
-            vtestnet.stop()
-
 
 
 @cm

skynet/utils.py

@@ -1,12 +1,14 @@
 #!/usr/bin/python
 
 import io
+import logging
 import os
 import time
 import random
 
 from typing import Optional
 from pathlib import Path
+import asks
 
 import torch
 import numpy as np
@@ -23,6 +25,8 @@ from diffusers import (
 )
 from realesrgan import RealESRGANer
 from huggingface_hub import login
+from torch.distributions import weibull
+import trio
 
 from .constants import MODELS
 
@@ -74,16 +78,25 @@ def pipeline_for(model: str, mem_fraction: float = 1.0, image=False) -> Diffusio
     torch.backends.cudnn.benchmark = False
     torch.use_deterministic_algorithms(True)
 
+    model_info = MODELS[model]
+
+    req_mem = model_info['mem']
+    mem_gb = torch.cuda.mem_get_info()[1] / (10**9)
+    over_mem = mem_gb < req_mem
+    if over_mem:
+        logging.warn(f'model requires {req_mem} but card has {mem_gb}, model will run slower..')
+
+    shortname = model_info['short']
+
     params = {
         'torch_dtype': torch.float16,
         'safety_checker': None
     }
 
-    if model == 'runwayml/stable-diffusion-v1-5':
+    if shortname == 'stable':
         params['revision'] = 'fp16'
 
-    if (model == 'stabilityai/stable-diffusion-xl-base-1.0' or
+    if 'xl' in shortname:
-        model == 'snowkidy/stable-diffusion-xl-base-0.9'):
         if image:
             pipe_class = StableDiffusionXLImg2ImgPipeline
         else:
@@ -100,10 +113,16 @@ def pipeline_for(model: str, mem_fraction: float = 1.0, image=False) -> Diffusio
     pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(
         pipe.scheduler.config)
 
+    if over_mem:
         if not image:
             pipe.enable_vae_slicing()
+            pipe.enable_vae_tiling()
 
-    return pipe.to('cuda')
+        pipe.enable_model_cpu_offload()
+
+    pipe.enable_xformers_memory_efficient_attention()
+
+    return pipe
 
 
 def txt2img(
@@ -209,14 +228,25 @@ def upscale(
         convert_from_image_to_cv2(input_img), outscale=4)
 
     image = convert_from_cv2_to_image(up_img)
-
-
     image.save(output)
 
 
+async def download_upscaler():
+    print('downloading upscaler...')
+    weights_path = Path('weights')
+    weights_path.mkdir(exist_ok=True)
+    upscaler_url = 'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth'
+    save_path = weights_path / 'RealESRGAN_x4plus.pth'
+    response = await asks.get(upscaler_url)
+    with open(save_path, 'wb') as f:
+        f.write(response.content)
+    print('done')
+
 def download_all_models(hf_token: str):
     assert torch.cuda.is_available()
 
+    trio.run(download_upscaler)
+
     login(token=hf_token)
     for model in MODELS:
         print(f'DOWNLOADING {model.upper()}')