First decupled architecture, still working on integrating tractor gpu workers

pull/2/head v0.1a5
Guillermo Rodriguez 2022-12-10 18:18:03 -03:00
parent 91e0693e65
commit 74d2426793
No known key found for this signature in database
GPG Key ID: EC3AB66D5D83B392
29 changed files with 1266 additions and 591 deletions

3
.dockerignore 100644
View File

@ -0,0 +1,3 @@
hf_home
inputs
outputs

3
.gitignore vendored
View File

@ -1,2 +1,5 @@
.python-version
hf_home
outputs
**/__pycache__
*.egg-info

View File

@ -1,31 +0,0 @@
from pytorch/pytorch:latest
env DEBIAN_FRONTEND=noninteractive
run apt-get update && apt-get install -y git wget
run conda install xformers -c xformers/label/dev
run pip install --upgrade \
diffusers[torch] \
accelerate \
transformers \
huggingface_hub \
pyTelegramBotAPI \
pymongo \
scipy \
pdbpp
env NVIDIA_VISIBLE_DEVICES=all
run mkdir /scripts
run mkdir /outputs
run mkdir /inputs
env HF_HOME /hf_home
run mkdir /hf_home
workdir /scripts
env PYTORCH_CUDA_ALLOC_CONF max_split_size_mb:128

13
Dockerfile.runtime 100644
View File

@ -0,0 +1,13 @@
from python:3.10.0
env DEBIAN_FRONTEND=noninteractive
workdir /skynet
copy requirements.* ./
run pip install \
-r requirements.txt \
-r requirements.test.txt
workdir /scripts

View File

@ -0,0 +1,23 @@
from nvidia/cuda:11.7.0-devel-ubuntu20.04
from python:3.10.0
env DEBIAN_FRONTEND=noninteractive
workdir /skynet
copy requirements.* .
run pip install -U pip ninja
run pip install -r requirements.cuda.0.txt
run pip install -v -r requirements.cuda.1.txt
run pip install \
-r requirements.txt \
-r requirements.test.txt
env NVIDIA_VISIBLE_DEVICES=all
env HF_HOME /hf_home
env PYTORCH_CUDA_ALLOC_CONF max_split_size_mb:128
workdir /scripts

11
LICENSE 100644
View File

@ -0,0 +1,11 @@
A menos que sea especificamente indicado en el cabezal del archivo, se reservan
todos los derechos sobre este codigo por parte de:
Guillermo Rodriguez, guillermor@fing.edu.uy
ENGLISH LICENSE:
Unless specifically indicated in the file header, all rights to this code are
reserved by:
Guillermo Rodriguez, guillermor@.edu.uy

47
README.md 100644
View File

@ -0,0 +1,47 @@
create db in postgres:
```sql
CREATE USER skynet WITH PASSWORD 'password';
CREATE DATABASE skynet_art_bot;
GRANT ALL PRIVILEGES ON DATABASE skynet_art_bot TO skynet;
CREATE SCHEMA IF NOT EXISTS skynet;
CREATE TABLE IF NOT EXISTS skynet.user(
id SERIAL PRIMARY KEY NOT NULL,
tg_id INT,
wp_id VARCHAR(128),
mx_id VARCHAR(128),
ig_id VARCHAR(128),
generated INT NOT NULL,
joined DATE NOT NULL,
last_prompt TEXT,
role VARCHAR(128) NOT NULL
);
ALTER TABLE skynet.user
ADD CONSTRAINT tg_unique
UNIQUE (tg_id);
ALTER TABLE skynet.user
ADD CONSTRAINT wp_unique
UNIQUE (wp_id);
ALTER TABLE skynet.user
ADD CONSTRAINT mx_unique
UNIQUE (mx_id);
ALTER TABLE skynet.user
ADD CONSTRAINT ig_unique
UNIQUE (ig_id);
CREATE TABLE IF NOT EXISTS skynet.user_config(
id SERIAL NOT NULL,
algo VARCHAR(128) NOT NULL,
step INT NOT NULL,
width INT NOT NULL,
height INT NOT NULL,
seed INT,
guidance INT NOT NULL,
upscaler VARCHAR(128)
);
ALTER TABLE skynet.user_config
ADD FOREIGN KEY(id)
REFERENCES skynet.user(id);
```

7
build_docker.sh 100755
View File

@ -0,0 +1,7 @@
docker build \
-t skynet:runtime-cuda \
-f Dockerfile.runtime-cuda .
docker build \
-t skynet:runtime \
-f Dockerfile.runtime .

2
pytest.ini 100644
View File

@ -0,0 +1,2 @@
[pytest]
trio_mode = true

View File

@ -0,0 +1,8 @@
pdbpp
scipy
accelerate
transformers
huggingface_hub
diffusers[torch]
torch==1.13.0+cu117
--extra-index-url https://download.pytorch.org/whl/cu117

View File

@ -0,0 +1 @@
git+https://github.com/facebookresearch/xformers.git@main#egg=xformers

View File

@ -0,0 +1,2 @@
pytest
pytest-trio

8
requirements.txt 100644
View File

@ -0,0 +1,8 @@
trio
pynng
triopg
aiohttp
msgspec
trio_asyncio
git+https://github.com/goodboy/tractor.git@master#egg=tractor

View File

@ -1,14 +0,0 @@
mkdir -p outputs
mkdir -p hf_home
docker run \
-it \
--rm \
--gpus=all \
--env HF_TOKEN='' \
--env DB_USER='skynet' \
--env DB_PASS='nnf01nmf091d0i' \
--mount type=bind,source="$(pwd)"/outputs,target=/outputs \
--mount type=bind,source="$(pwd)"/hf_home,target=/hf_home \
--mount type=bind,source="$(pwd)"/scripts,target=/scripts \
skynet:dif python telegram-bot-dev.py

View File

@ -1,9 +0,0 @@
docker run
-d \
--rm \
-p 27017:27017 \
--name mongodb-skynet \
--mount type=bind,source="$(pwd)"/mongodb,target=/data/db \
-e MONGO_INITDB_ROOT_USERNAME="" \
-e MONGO_INITDB_ROOT_PASSWORD="" \
mongo

View File

@ -1,537 +0,0 @@
#!/usr/bin/python
import os
import logging
import random
from torch.multiprocessing import spawn
import telebot
from telebot.types import InputFile
import sys
import uuid
from pathlib import Path
import torch
from torch.multiprocessing.spawn import ProcessRaisedException
from diffusers import (
StableDiffusionPipeline,
EulerAncestralDiscreteScheduler
)
from huggingface_hub import login
from datetime import datetime
from pymongo import MongoClient
from typing import Tuple, Optional
db_user = os.environ['DB_USER']
db_pass = os.environ['DB_PASS']
logging.basicConfig(level=logging.INFO)
MEM_FRACTION = .33
ALGOS = {
'stable': 'runwayml/stable-diffusion-v1-5',
'midj': 'prompthero/openjourney',
'hdanime': 'Linaqruf/anything-v3.0',
'waifu': 'hakurei/waifu-diffusion',
'ghibli': 'nitrosocke/Ghibli-Diffusion',
'van-gogh': 'dallinmackay/Van-Gogh-diffusion',
'pokemon': 'lambdalabs/sd-pokemon-diffusers',
'ink': 'Envvi/Inkpunk-Diffusion',
'robot': 'nousr/robo-diffusion'
}
N = '\n'
HELP_TEXT = f'''
test art bot v0.1a4
commands work on a user per user basis!
config is individual to each user!
/txt2img TEXT - request an image based on a prompt
/redo - redo last prompt
/cool - list of cool words to use
/stats - user statistics
/donate - see donation info
/config algo NAME - select AI to use one of:
{N.join(ALGOS.keys())}
/config step NUMBER - set amount of iterations
/config seed NUMBER - set the seed, deterministic results!
/config size WIDTH HEIGHT - set size in pixels
/config guidance NUMBER - prompt text importance
'''
UNKNOWN_CMD_TEXT = 'unknown command! try sending \"/help\"'
DONATION_INFO = '0xf95335682DF281FFaB7E104EB87B69625d9622B6\ngoal: 25/650usd'
COOL_WORDS = [
'cyberpunk',
'soviet propaganda poster',
'rastafari',
'cannabis',
'art deco',
'H R Giger Necronom IV',
'dimethyltryptamine',
'lysergic',
'slut',
'psilocybin',
'trippy',
'lucy in the sky with diamonds',
'fractal',
'da vinci',
'pencil illustration',
'blueprint',
'internal diagram',
'baroque',
'the last judgment',
'michelangelo'
]
GROUP_ID = -1001541979235
MP_ENABLED_ROLES = ['god']
MIN_STEP = 1
MAX_STEP = 100
MAX_SIZE = (512, 656)
MAX_GUIDANCE = 20
DEFAULT_SIZE = (512, 512)
DEFAULT_GUIDANCE = 7.5
DEFAULT_STEP = 75
DEFAULT_CREDITS = 10
DEFAULT_ALGO = 'stable'
DEFAULT_ROLE = 'pleb'
DEFAULT_UPSCALER = None
rr_total = 1
rr_id = 0
request_counter = 0
def its_my_turn():
global request_counter, rr_total, rr_id
my_turn = request_counter % rr_total == rr_id
logging.info(f'new request {request_counter}, turn: {my_turn} rr_total: {rr_total}, rr_id {rr_id}')
request_counter += 1
return my_turn
def round_robined(func):
def rr_wrapper(*args, **kwargs):
if not its_my_turn():
return
func(*args, **kwargs)
return rr_wrapper
def generate_image(
i: int,
prompt: str,
name: str,
step: int,
size: Tuple[int, int],
guidance: int,
seed: int,
algo: str,
upscaler: Optional[str]
):
assert torch.cuda.is_available()
torch.cuda.empty_cache()
torch.cuda.set_per_process_memory_fraction(MEM_FRACTION)
with torch.no_grad():
if algo == 'stable':
pipe = StableDiffusionPipeline.from_pretrained(
'runwayml/stable-diffusion-v1-5',
torch_dtype=torch.float16,
revision="fp16",
safety_checker=None
)
else:
pipe = StableDiffusionPipeline.from_pretrained(
ALGOS[algo],
torch_dtype=torch.float16,
safety_checker=None
)
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
pipe = pipe.to("cuda")
w, h = size
print(f'generating image... of size {w, h}')
image = pipe(
prompt,
width=w,
height=h,
guidance_scale=guidance, num_inference_steps=step,
generator=torch.Generator("cuda").manual_seed(seed)
).images[0]
if upscaler == 'x4':
pipe = StableDiffusionPipeline.from_pretrained(
'stabilityai/stable-diffusion-x4-upscaler',
revision="fp16",
torch_dtype=torch.float16
)
image = pipe(prompt=prompt, image=image).images[0]
image.save(f'/outputs/{name}.png')
print('saved')
if __name__ == '__main__':
API_TOKEN = '5880619053:AAFge2UfObw1kCn9Kb7AAyqaIHs_HgM0Fx0'
bot = telebot.TeleBot(API_TOKEN)
db_client = MongoClient(
host=['ancap.tech:64000'],
username=db_user,
password=db_pass)
tgdb = db_client.get_database('telegram')
collections = tgdb.list_collection_names()
if 'users' in collections:
tg_users = tgdb.get_collection('users')
# tg_users.delete_many({})
else:
tg_users = tgdb.create_collection('users')
# db functions
def get_user(uid: int):
return tg_users.find_one({'uid': uid})
def new_user(uid: int):
if get_user(uid):
raise ValueError('User already present on db')
res = tg_users.insert_one({
'generated': 0,
'uid': uid,
'credits': DEFAULT_CREDITS,
'joined': datetime.utcnow().isoformat(),
'last_prompt': None,
'role': DEFAULT_ROLE,
'config': {
'algo': DEFAULT_ALGO,
'step': DEFAULT_STEP,
'size': DEFAULT_SIZE,
'seed': None,
'guidance': DEFAULT_GUIDANCE,
'upscaler': DEFAULT_UPSCALER
}
})
assert res.acknowledged
return get_user(uid)
def migrate_user(db_user):
# new: user roles
if 'role' not in db_user:
res = tg_users.find_one_and_update(
{'uid': db_user['uid']}, {'$set': {'role': DEFAULT_ROLE}})
# new: algo selection
if 'algo' not in db_user['config']:
res = tg_users.find_one_and_update(
{'uid': db_user['uid']}, {'$set': {'config.algo': DEFAULT_ALGO}})
# new: upscaler selection
if 'upscaler' not in db_user['config']:
res = tg_users.find_one_and_update(
{'uid': db_user['uid']}, {'$set': {'config.upscaler': DEFAULT_UPSCALER}})
return get_user(db_user['uid'])
def get_or_create_user(uid: int):
db_user = get_user(uid)
if not db_user:
db_user = new_user(uid)
logging.info(f'req from: {uid}')
return migrate_user(db_user)
def update_user(uid: int, updt_cmd: dict):
user = get_user(uid)
if not user:
raise ValueError('User not present on db')
return tg_users.find_one_and_update(
{'uid': uid}, updt_cmd)
# bot handler
def img_for_user_with_prompt(
uid: int,
prompt: str, step: int, size: Tuple[int, int], guidance: int, seed: int,
algo: str, upscaler: Optional[str]
):
name = uuid.uuid4()
spawn(
generate_image,
args=(prompt, name, step, size, guidance, seed, algo, upscaler))
logging.info(f'done generating. got {name}, sending...')
if len(prompt) > 256:
reply_txt = f'prompt: \"{prompt[:256]}...\"\n(full prompt too big to show on reply...)\n'
else:
reply_txt = f'prompt: \"{prompt}\"\n'
reply_txt += f'seed: {seed}\n'
reply_txt += f'iterations: {step}\n'
reply_txt += f'size: {size}\n'
reply_txt += f'guidance: {guidance}\n'
reply_txt += f'algo: {ALGOS[algo]}\n'
reply_txt += f'euler ancestral discrete'
return reply_txt, name
@bot.message_handler(commands=['help'])
@round_robined
def send_help(message):
bot.reply_to(message, HELP_TEXT)
@bot.message_handler(commands=['cool'])
@round_robined
def send_cool_words(message):
bot.reply_to(message, '\n'.join(COOL_WORDS))
@bot.message_handler(commands=['txt2img'])
@round_robined
def send_txt2img(message):
chat = message.chat
user = message.from_user
db_user = get_or_create_user(user.id)
if ((chat.type != 'group' and chat.id != GROUP_ID) and
(db_user['role'] not in MP_ENABLED_ROLES)):
return
prompt = ' '.join(message.text.split(' ')[1:])
if len(prompt) == 0:
bot.reply_to(message, 'empty text prompt ignored.')
return
logging.info(f"{user.first_name} ({user.id}) on chat {chat.id} txt2img: {prompt}")
user_conf = db_user['config']
algo = user_conf['algo']
step = user_conf['step']
size = user_conf['size']
seed = user_conf['seed'] if user_conf['seed'] else random.randint(0, 999999999)
guidance = user_conf['guidance']
upscaler = user_conf['upscaler']
try:
reply_txt, name = img_for_user_with_prompt(
user.id, prompt, step, size, guidance, seed, algo, upscaler)
update_user(
user.id,
{'$set': {
'generated': db_user['generated'] + 1,
'last_prompt': prompt
}})
bot.send_photo(
chat.id,
caption=f'sent by: {user.first_name}\n' + reply_txt,
photo=InputFile(f'/outputs/{name}.png'))
except BaseException as e:
logging.error(e)
bot.reply_to(message, 'that command caused an error :(\nchange settings and try again (?')
@bot.message_handler(commands=['redo'])
@round_robined
def redo_txt2img(message):
# check msg comes from testing group
chat = message.chat
user = message.from_user
db_user = get_or_create_user(user.id)
if ((chat.type != 'group' and chat.id != GROUP_ID) and
(db_user['role'] not in MP_ENABLED_ROLES)):
return
prompt = db_user['last_prompt']
if not prompt:
bot.reply_to(message, 'do a /txt2img command first silly!')
return
user_conf = db_user['config']
algo = user_conf['algo']
step = user_conf['step']
size = user_conf['size']
seed = user_conf['seed'] if user_conf['seed'] else random.randint(0, 999999999)
guidance = user_conf['guidance']
upscaler = user_conf['upscaler']
logging.info(f"{user.first_name} ({user.id}) on chat {chat.id} redo: {prompt}")
try:
reply_txt, name = img_for_user_with_prompt(
user.id, prompt, step, size, guidance, seed, algo, upscaler)
update_user(
user.id,
{'$set': {
'generated': db_user['generated'] + 1,
}})
bot.send_photo(
chat.id,
caption=f'sent by: {user.first_name}\n' + reply_txt,
photo=InputFile(f'/outputs/{name}.png'))
except BaseException as e:
logging.error(e)
bot.reply_to(message, 'that command caused an error :(\nchange settings and try again (?')
@bot.message_handler(commands=['config'])
@round_robined
def set_config(message):
logging.info(f'config req on chat: {message.chat.id}')
params = message.text.split(' ')
if len(params) < 3:
bot.reply_to(message, 'wrong msg format')
else:
user = message.from_user
chat = message.chat
db_user = get_or_create_user(user.id)
try:
attr = params[1]
if attr == 'algo':
val = params[2]
assert val in ALGOS
res = update_user(user.id, {'$set': {'config.algo': val}})
elif attr == 'step':
val = int(params[2])
val = max(min(val, MAX_STEP), MIN_STEP)
res = update_user(user.id, {'$set': {'config.step': val}})
elif attr == 'size':
max_w, max_h = MAX_SIZE
w = max(min(int(params[2]), max_w), 16)
h = max(min(int(params[3]), max_h), 16)
val = (w, h)
if (w % 8 != 0) or (h % 8 != 0):
bot.reply_to(message, 'size must be divisible by 8!')
return
res = update_user(user.id, {'$set': {'config.size': val}})
elif attr == 'seed':
val = params[2]
if val == 'auto':
val = None
else:
val = int(params[2])
res = update_user(user.id, {'$set': {'config.seed': val}})
elif attr == 'guidance':
val = float(params[2])
val = max(min(val, MAX_GUIDANCE), 0)
res = update_user(user.id, {'$set': {'config.guidance': val}})
elif attr == 'upscaler':
val = params[2]
if val == 'off':
val = None
res = update_user(user.id, {'$set': {'config.upscaler': val}})
else:
bot.reply_to(message, f'\"{attr}\" not a parameter')
bot.reply_to(message, f'config updated! {attr} to {val}')
except ValueError:
bot.reply_to(message, f'\"{val}\" is not a number silly')
except AssertionError:
bot.reply_to(message, f'no algo named {val}')
@bot.message_handler(commands=['stats'])
@round_robined
def user_stats(message):
user = message.from_user
db_user = get_or_create_user(user.id)
migrate_user(db_user)
joined_date_str = datetime.fromisoformat(db_user['joined']).strftime('%B the %dth %Y, %H:%M:%S')
user_stats_str = f'generated: {db_user["generated"]}\n'
user_stats_str += f'joined: {joined_date_str}\n'
user_stats_str += f'credits: {db_user["credits"]}\n'
user_stats_str += f'role: {db_user["role"]}\n'
bot.reply_to(
message, user_stats_str)
@bot.message_handler(commands=['donate'])
@round_robined
def donation_info(message):
bot.reply_to(
message, DONATION_INFO)
@bot.message_handler(commands=['say'])
@round_robined
def say(message):
chat = message.chat
user = message.from_user
db_user = get_or_create_user(user.id)
if (chat.type == 'group') or (db_user['role'] not in MP_ENABLED_ROLES):
return
bot.send_message(GROUP_ID, message.text[4:])
@bot.message_handler(func=lambda message: True)
@round_robined
def echo_message(message):
if message.text[0] == '/':
bot.reply_to(message, UNKNOWN_CMD_TEXT)
login(token=os.environ['HF_TOKEN'])
bot.infinity_polling()

11
setup.py 100644
View File

@ -0,0 +1,11 @@
from setuptools import setup, find_packages
setup(
name='skynet-bot',
version='0.1.0a6',
description='Decentralized compute platform',
author='Guillermo Rodriguez',
author_email='guillermo@telos.net',
packages=find_packages(),
install_requires=[]
)

View File

@ -0,0 +1,2 @@
#!/usr/bin/python

246
skynet_bot/brain.py 100644
View File

@ -0,0 +1,246 @@
#!/usr/bin/python
import json
import uuid
import base64
import logging
from uuid import UUID
from functools import partial
from collections import OrderedDict
import trio
import pynng
import trio_asyncio
from .db import *
from .types import *
from .constants import *
class SkynetDGPUOffline(BaseException):
...
class SkynetDGPUOverloaded(BaseException):
...
async def rpc_service(sock, dgpu_bus, db_pool):
nodes = OrderedDict()
wip_reqs = {}
fin_reqs = {}
def are_all_workers_busy():
for nid, info in nodes.items():
if info['task'] == None:
return False
return True
next_worker = 0
def get_next_worker():
nonlocal next_worker
if len(nodes) == 0:
raise SkynetDGPUOffline
if are_all_workers_busy():
raise SkynetDGPUOverloaded
next_worker += 1
if next_worker >= len(nodes):
next_worker = 0
nid = list(nodes.keys())[next_worker]
return nid
async def dgpu_image_streamer():
nonlocal wip_reqs, fin_reqs
while True:
msg = await dgpu_bus.arecv_msg()
rid = UUID(bytes=msg.bytes[:16]).hex
img = msg.bytes[16:].hex()
fin_reqs[rid] = img
event = wip_reqs[rid]
event.set()
del wip_reqs[rid]
async def dgpu_stream_one_img(req: ImageGenRequest):
nonlocal wip_reqs, fin_reqs, next_worker
nid = get_next_worker()
logging.info(f'dgpu_stream_one_img {next_worker} {nid}')
rid = uuid.uuid4().hex
event = trio.Event()
wip_reqs[rid] = event
nodes[nid]['task'] = rid
dgpu_req = DGPUBusRequest(
rid=rid,
nid=nid,
task='diffuse',
params=req.to_dict())
logging.info(f'dgpu_bus req: {dgpu_req}')
await dgpu_bus.asend(
json.dumps(dgpu_req.to_dict()).encode())
await event.wait()
nodes[nid]['task'] = None
img = fin_reqs[rid]
del fin_reqs[rid]
logging.info(f'done streaming {img}')
return rid, img
async def handle_user_request(rpc_ctx, req):
try:
async with db_pool.acquire() as conn:
user = await get_or_create_user(conn, req.uid)
result = {}
match req.method:
case 'txt2img':
logging.info('txt2img')
user_config = {**(await get_user_config(conn, user))}
del user_config['id']
prompt = req.params['prompt']
req = ImageGenRequest(
prompt=prompt,
**user_config
)
rid, img = await dgpu_stream_one_img(req)
result = {
'id': rid,
'img': img
}
case 'redo':
logging.info('redo')
user_config = await get_user_config(conn, user)
prompt = await get_last_prompt_of(conn, user)
req = ImageGenRequest(
prompt=prompt,
**user_config
)
rid, img = await dgpu_stream_one_img(req)
result = {
'id': rid,
'img': img
}
case 'config':
logging.info('config')
if req.params['attr'] in CONFIG_ATTRS:
await update_user_config(
conn, user, req.params['attr'], req.params['val'])
case 'stats':
logging.info('stats')
generated, joined, role = await get_user_stats(conn, user)
result = {
'generated': generated,
'joined': joined.strftime(DATE_FORMAT),
'role': role
}
case _:
logging.warn('unknown method')
except SkynetDGPUOffline:
result = {
'error': 'skynet_dgpu_offline'
}
except SkynetDGPUOverloaded:
result = {
'error': 'skynet_dgpu_overloaded',
'nodes': len(nodes)
}
except BaseException as e:
logging.error(e)
raise e
# result = {
# 'error': 'skynet_internal_error'
# }
await rpc_ctx.asend(
json.dumps(
SkynetRPCResponse(result=result).to_dict()).encode())
async with trio.open_nursery() as n:
n.start_soon(dgpu_image_streamer)
while True:
ctx = sock.new_context()
msg = await ctx.arecv_msg()
content = msg.bytes.decode()
req = SkynetRPCRequest(**json.loads(content))
logging.info(req)
if req.method == 'dgpu_online':
nodes[req.uid] = {
'task': None
}
logging.info(f'dgpu online: {req.uid}')
elif req.method == 'dgpu_offline':
i = nodes.values().index(req.uid)
del nodes[req.uid]
if i < next_worker:
next_worker -= 1
logging.info(f'dgpu offline: {req.uid}')
else:
n.start_soon(
handle_user_request, ctx, req)
continue
await ctx.asend(
json.dumps(
SkynetRPCResponse(
result={'ok': {}}).to_dict()).encode())
async def run_skynet(
db_user: str,
db_pass: str,
db_host: str = DB_HOST,
rpc_address: str = DEFAULT_RPC_ADDR,
dgpu_address: str = DEFAULT_DGPU_ADDR,
task_status = trio.TASK_STATUS_IGNORED
):
logging.basicConfig(level=logging.INFO)
logging.info('skynet is starting')
async with (
trio.open_nursery() as n,
open_database_connection(
db_user, db_pass, db_host) as db_pool
):
logging.info('connected to db.')
with (
pynng.Rep0(listen=rpc_address) as rpc_sock,
pynng.Bus0(listen=dgpu_address) as dgpu_bus
):
n.start_soon(
rpc_service, rpc_sock, dgpu_bus, db_pool)
task_status.started()
try:
await trio.sleep_forever()
except KeyboardInterrupt:
...

View File

@ -0,0 +1,129 @@
#!/usr/bin/python
API_TOKEN = '5880619053:AAFge2UfObw1kCn9Kb7AAyqaIHs_HgM0Fx0'
DB_HOST = 'ancap.tech:34508'
ALGOS = {
'stable': 'runwayml/stable-diffusion-v1-5',
'midj': 'prompthero/openjourney',
'hdanime': 'Linaqruf/anything-v3.0',
'waifu': 'hakurei/waifu-diffusion',
'ghibli': 'nitrosocke/Ghibli-Diffusion',
'van-gogh': 'dallinmackay/Van-Gogh-diffusion',
'pokemon': 'lambdalabs/sd-pokemon-diffusers',
'ink': 'Envvi/Inkpunk-Diffusion',
'robot': 'nousr/robo-diffusion'
}
N = '\n'
HELP_TEXT = f'''
test art bot v0.1a4
commands work on a user per user basis!
config is individual to each user!
/txt2img TEXT - request an image based on a prompt
/redo - re ont
/help step - get info on step config option
/help guidance - get info on guidance config option
/cool - list of cool words to use
/stats - user statistics
/donate - see donation info
/config algo NAME - select AI to use one of:
{N.join(ALGOS.keys())}
/config step NUMBER - set amount of iterations
/config seed NUMBER - set the seed, deterministic results!
/config size WIDTH HEIGHT - set size in pixels
/config guidance NUMBER - prompt text importance
'''
UNKNOWN_CMD_TEXT = 'unknown command! try sending \"/help\"'
DONATION_INFO = '0xf95335682DF281FFaB7E104EB87B69625d9622B6\ngoal: 25/650usd'
COOL_WORDS = [
'cyberpunk',
'soviet propaganda poster',
'rastafari',
'cannabis',
'art deco',
'H R Giger Necronom IV',
'dimethyltryptamine',
'lysergic',
'slut',
'psilocybin',
'trippy',
'lucy in the sky with diamonds',
'fractal',
'da vinci',
'pencil illustration',
'blueprint',
'internal diagram',
'baroque',
'the last judgment',
'michelangelo'
]
HELP_STEP = '''
diffusion models are iterative processes a repeated cycle that starts with a\
random noise generated from text input. With each step, some noise is removed\
, resulting in a higher-quality image over time. The repetition stops when the\
desired number of steps completes.
around 25 sampling steps are usually enough to achieve high-quality images. Us\
ing more may produce a slightly different picture, but not necessarily better \
quality.
'''
HELP_GUIDANCE = '''
the guidance scale is a parameter that controls how much the image generation\
process follows the text prompt. The higher the value, the more image sticks\
to a given text input.
'''
HELP_UNKWNOWN_PARAM = 'don\'t have any info on that.'
GROUP_ID = -1001541979235
MP_ENABLED_ROLES = ['god']
MIN_STEP = 1
MAX_STEP = 100
MAX_WIDTH = 512
MAX_HEIGHT = 656
MAX_GUIDANCE = 20
DEFAULT_SEED = None
DEFAULT_WIDTH = 512
DEFAULT_HEIGHT = 512
DEFAULT_GUIDANCE = 7.5
DEFAULT_STEP = 35
DEFAULT_CREDITS = 10
DEFAULT_ALGO = 'midj'
DEFAULT_ROLE = 'pleb'
DEFAULT_UPSCALER = None
DEFAULT_RPC_ADDR = 'tcp://127.0.0.1:41000'
DEFAULT_DGPU_ADDR = 'tcp://127.0.0.1:41069'
DEFAULT_DGPU_MAX_TASKS = 3
DEFAULT_INITAL_ALGOS = ['midj', 'stable', 'ink']
DATE_FORMAT = '%B the %dth %Y, %H:%M:%S'
CONFIG_ATTRS = [
'algo',
'step',
'width',
'height',
'seed',
'guidance',
'upscaler'
]

146
skynet_bot/db.py 100644
View File

@ -0,0 +1,146 @@
#!/usr/bin/python
import logging
from datetime import datetime
from contextlib import asynccontextmanager as acm
import trio
import triopg
from .constants import *
def try_decode_uid(uid: str):
try:
proto, uid = uid.split('+')
uid = int(uid)
return proto, uid
except ValueError:
logging.warning(f'got non numeric uid?: {uid}')
return None, None
@acm
async def open_database_connection(
db_user: str,
db_pass: str,
db_host: str = DB_HOST,
):
async with triopg.create_pool(
dsn=f'postgres://{db_user}:{db_pass}@{db_host}/skynet_art_bot'
) as conn:
yield conn
async def get_user(conn, uid: str):
if isinstance(uid, str):
proto, uid = try_decode_uid(uid)
match proto:
case 'tg':
stmt = await conn.prepare(
'SELECT * FROM skynet.user WHERE tg_id = $1')
user = await stmt.fetchval(uid)
case _:
user = None
return user
else: # asumme is our uid
stmt = await conn.prepare(
'SELECT * FROM skynet.user WHERE id = $1')
return await stmt.fetchval(uid)
async def get_user_config(conn, user: int):
stmt = await conn.prepare(
'SELECT * FROM skynet.user_config WHERE id = $1')
return (await stmt.fetch(user))[0]
async def get_last_prompt_of(conn, user: int):
stms = await conn.prepare(
'SELECT last_prompt FROM skynet.user WHERE id = $1')
return await stmt.fetchval(user)
async def new_user(conn, uid: str):
if await get_user(conn, uid):
raise ValueError('User already present on db')
logging.info(f'new user! {uid}')
tg_id = None
date = datetime.utcnow()
proto, pid = try_decode_uid(uid)
match proto:
case 'tg':
tg_id = pid
async with conn.transaction():
stmt = await conn.prepare('''
INSERT INTO skynet.user(
tg_id, generated, joined, last_prompt, role)
VALUES($1, $2, $3, $4, $5)
''')
await stmt.fetch(
tg_id, 0, date, None, DEFAULT_ROLE
)
new_uid = await get_user(conn, uid)
stmt = await conn.prepare('''
INSERT INTO skynet.user_config(
id, algo, step, width, height, seed, guidance, upscaler)
VALUES($1, $2, $3, $4, $5, $6, $7, $8)
''')
user = await stmt.fetch(
new_uid,
DEFAULT_ALGO,
DEFAULT_STEP,
DEFAULT_WIDTH,
DEFAULT_HEIGHT,
DEFAULT_SEED,
DEFAULT_GUIDANCE,
DEFAULT_UPSCALER
)
return new_uid
async def get_or_create_user(conn, uid: str):
user = await get_user(conn, uid)
if not user:
user = await new_user(conn, uid)
return user
async def update_user(conn, user: int, attr: str, val):
...
async def update_user_config(conn, user: int, attr: str, val):
stmt = await conn.prepare(f'''
UPDATE skynet.user_config
SET {attr} = $2
WHERE id = $1
''')
await stmt.fetch(user, val)
async def get_user_stats(conn, user: int):
stmt = await conn.prepare('''
SELECT generated,joined,role FROM skynet.user
WHERE id = $1
''')
records = await stmt.fetch(user)
assert len(records) == 1
record = records[0]
return record

121
skynet_bot/dgpu.py 100644
View File

@ -0,0 +1,121 @@
#!/usr/bin/python
import trio
import json
import uuid
import logging
import pynng
import tractor
from . import gpu
from .gpu import open_gpu_worker
from .types import *
from .constants import *
from .frontend import rpc_call
async def open_dgpu_node(
rpc_address: str = DEFAULT_RPC_ADDR,
dgpu_address: str = DEFAULT_DGPU_ADDR,
dgpu_max_tasks: int = DEFAULT_DGPU_MAX_TASKS,
initial_algos: str = DEFAULT_INITAL_ALGOS
):
logging.basicConfig(level=logging.INFO)
name = uuid.uuid4()
workers = initial_algos.copy()
tasks = [None for _ in range(dgpu_max_tasks)]
portal_map: dict[int, tractor.Portal]
contexts: dict[int, tractor.Context]
def get_next_worker(need_algo: str):
nonlocal workers, tasks
for task, algo in zip(workers, tasks):
if need_algo == algo and not task:
return workers.index(need_algo)
return tasks.index(None)
async def gpu_streamer(
ctx: tractor.Context,
nid: int
):
nonlocal tasks
async with ctx.open_stream() as stream:
async for img in stream:
tasks[nid]['res'] = img
tasks[nid]['event'].set()
async def gpu_compute_one(ireq: ImageGenRequest):
wid = get_next_worker(ireq.algo)
event = trio.Event()
workers[wid] = ireq.algo
tasks[wid] = {
'res': None, 'event': event}
await contexts[i].send(ireq)
await event.wait()
img = tasks[wid]['res']
tasks[wid] = None
return img
with (
pynng.Req0(dial=rpc_address) as rpc_sock,
pynng.Bus0(dial=dgpu_address) as dgpu_sock
):
async def _rpc_call(*args, **kwargs):
return await rpc_call(rpc_sock, *args, **kwargs)
async def _process_dgpu_req(req: DGPUBusRequest):
img = await gpu_compute_one(
ImageGenRequest(**req.params))
await dgpu_sock.asend(
bytes.fromhex(req.rid) + img)
res = await _rpc_call(
name.hex, 'dgpu_online', {'max_tasks': dgpu_max_tasks})
logging.info(res)
assert 'ok' in res.result
async with (
tractor.open_actor_cluster(
modules=['skynet_bot.gpu'],
count=dgpu_max_tasks,
names=[i for i in range(dgpu_max_tasks)]
) as portal_map,
trio.open_nursery() as n
):
logging.info(f'starting {dgpu_max_tasks} gpu workers')
async with tractor.gather_contexts((
ctx.open_context(
open_gpu_worker, algo, 1.0 / dgpu_max_tasks)
)) as contexts:
contexts = {i: ctx for i, ctx in enumerate(contexts)}
for i, ctx in contexts.items():
n.start_soon(
gpu_streamer, ctx, i)
try:
while True:
msg = await dgpu_sock.arecv()
req = DGPUBusRequest(
**json.loads(msg.decode()))
if req.nid != name.hex:
continue
logging.info(f'dgpu: {name}, req: {req}')
n.start_soon(
_process_dgpu_req, req)
except KeyboardInterrupt:
...
res = await _rpc_call(name.hex, 'dgpu_offline')
logging.info(res)
assert 'ok' in res.result

View File

@ -0,0 +1,50 @@
#!/usr/bin/python
import json
from typing import Union
from contextlib import contextmanager as cm
import pynng
from ..types import SkynetRPCRequest, SkynetRPCResponse
from ..constants import DEFAULT_RPC_ADDR
class ConfigUnknownAttribute(BaseException):
...
class ConfigUnknownAlgorithm(BaseException):
...
class ConfigUnknownUpscaler(BaseException):
...
class ConfigSizeDivisionByEight(BaseException):
...
async def rpc_call(
sock,
uid: Union[int, str],
method: str,
params: dict = {}
):
req = SkynetRPCRequest(
uid=uid,
method=method,
params=params
)
await sock.asend(
json.dumps(
req.to_dict()).encode())
return SkynetRPCResponse(
**json.loads(
(await sock.arecv_msg()).bytes.decode()))
@cm
def open_skynet_rpc(rpc_address: str = DEFAULT_RPC_ADDR):
with pynng.Req0(dial=rpc_address) as rpc_sock:
yield rpc_sock

View File

@ -0,0 +1,164 @@
#!/usr/bin/python
import logging
from datetime import datetime
import pynng
from telebot.async_telebot import AsyncTeleBot
from trio_asyncio import aio_as_trio
from ..constants import *
from . import *
PREFIX = 'tg'
async def run_skynet_telegram(tg_token: str):
logging.basicConfig(level=logging.INFO)
bot = AsyncTeleBot(tg_token)
with open_skynet_rpc() as rpc_sock:
async def _rpc_call(
uid: int,
method: str,
params: dict
):
return await rpc_call(
rpc_sock, f'{PREFIX}+{uid}', method, params)
@bot.message_handler(commands=['help'])
async def send_help(message):
await bot.reply_to(message, HELP_TEXT)
@bot.message_handler(commands=['cool'])
async def send_cool_words(message):
await bot.reply_to(message, '\n'.join(COOL_WORDS))
@bot.message_handler(commands=['txt2img'])
async def send_txt2img(message):
resp = await _rpc_call(
message.from_user.id,
'txt2img',
{}
)
@bot.message_handler(commands=['redo'])
async def redo_txt2img(message):
resp = await _rpc_call(
message.from_user.id,
'redo',
{}
)
@bot.message_handler(commands=['config'])
async def set_config(message):
params = message.text.split(' ')
rpc_params = {}
if len(params) < 3:
bot.reply_to(message, 'wrong msg format')
else:
try:
attr = params[1]
if attr == 'algo':
val = params[2]
if val not in ALGOS:
raise ConfigUnknownAlgorithm
elif attr == 'step':
val = int(params[2])
val = max(min(val, MAX_STEP), MIN_STEP)
elif attr == 'width':
val = max(min(int(params[2]), MAX_WIDTH), 16)
if val % 8 != 0:
raise ConfigSizeDivisionByEight
elif attr == 'height':
val = max(min(int(params[2]), MAX_HEIGHT), 16)
if val % 8 != 0:
raise ConfigSizeDivisionByEight
elif attr == 'seed':
val = params[2]
if val == 'auto':
val = None
else:
val = int(params[2])
elif attr == 'guidance':
val = float(params[2])
val = max(min(val, MAX_GUIDANCE), 0)
elif attr == 'upscaler':
val = params[2]
if val == 'off':
val = None
elif val != 'x4':
raise ConfigUnknownUpscaler
else:
raise ConfigUnknownAttribute
resp = await _rpc_call(
message.from_user.id,
'config', {'attr': attr, 'val': val})
reply_txt = f'config updated! {attr} to {val}'
except ConfigUnknownAlgorithm:
reply_txt = f'no algo named {val}'
except ConfigUnknownAttribute:
reply_txt = f'\"{attr}\" not a configurable parameter'
except ConfigUnknownUpscaler:
reply_txt = f'\"{val}\" is not a valid upscaler'
except ConfigSizeDivisionByEight:
reply_txt = 'size must be divisible by 8!'
except ValueError:
reply_txt = f'\"{val}\" is not a number silly'
await bot.reply_to(message, reply_txt)
@bot.message_handler(commands=['stats'])
async def user_stats(message):
resp = await _rpc_call(
message.from_user.id,
'stats',
{}
)
stats = resp.result
stats_str = f'generated: {stats["generated"]}\n'
stats_str += f'joined: {stats["joined"]}\n'
stats_str += f'role: {stats["role"]}\n'
await bot.reply_to(
message, stats_str)
@bot.message_handler(commands=['donate'])
async def donation_info(message):
await bot.reply_to(
message, DONATION_INFO)
@bot.message_handler(func=lambda message: True)
async def echo_message(message):
if message.text[0] == '/':
await bot.reply_to(message, UNKNOWN_CMD_TEXT)
await aio_as_trio(bot.infinity_polling())

75
skynet_bot/gpu.py 100644
View File

@ -0,0 +1,75 @@
#!/usr/bin/python
import io
import random
import logging
import torch
import tractor
from diffusers import (
StableDiffusionPipeline,
EulerAncestralDiscreteScheduler
)
from .types import ImageGenRequest
from .constants import ALGOS
def pipeline_for(algo: str, mem_fraction: float):
assert torch.cuda.is_available()
torch.cuda.empty_cache()
torch.cuda.set_per_process_memory_fraction(mem_fraction)
torch.backends.cuda.matmul.allow_tf32 = True
torch.backends.cudnn.allow_tf32 = True
params = {
'torch_dtype': torch.float16,
'safety_checker': None
}
if algo == 'stable':
params['revision'] = 'fp16'
pipe = StableDiffusionPipeline.from_pretrained(
ALGOS[algo], **params)
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(
pipe.scheduler.config)
return pipe.to("cuda")
@tractor.context
async def open_gpu_worker(
ctx: tractor.Context,
start_algo: str,
mem_fraction: float
):
current_algo = start_algo
with torch.no_grad():
pipe = pipeline_for(current_algo, mem_fraction)
await ctx.started()
async with ctx.open_stream() as bus:
async for ireq in bus:
if ireq.algo != current_algo:
current_algo = ireq.algo
pipe = pipeline_for(current_algo, mem_fraction)
seed = ireq.seed if ireq.seed else random.randint(0, 2 ** 64)
image = pipe(
ireq.prompt,
width=ireq.width,
height=ireq.height,
guidance_scale=ireq.guidance,
num_inference_steps=ireq.step,
generator=torch.Generator("cuda").manual_seed(seed)
).images[0]
torch.cuda.empty_cache()
# convert PIL.Image to BytesIO
img_bytes = io.BytesIO()
image.save(img_bytes, format='PNG')
await bus.send(img_bytes.getvalue())

109
skynet_bot/types.py 100644
View File

@ -0,0 +1,109 @@
# piker: trading gear for hackers
# Copyright (C) Guillermo Rodriguez (in stewardship for piker0)
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
"""
Built-in (extension) types.
"""
import sys
from typing import Optional, Union
from pprint import pformat
import msgspec
class Struct(
msgspec.Struct,
# https://jcristharif.com/msgspec/structs.html#tagged-unions
# tag='pikerstruct',
# tag=True,
):
'''
A "human friendlier" (aka repl buddy) struct subtype.
'''
def to_dict(self) -> dict:
return {
f: getattr(self, f)
for f in self.__struct_fields__
}
def __repr__(self):
# only turn on pprint when we detect a python REPL
# at runtime B)
if (
hasattr(sys, 'ps1')
# TODO: check if we're in pdb
):
return self.pformat()
return super().__repr__()
def pformat(self) -> str:
return f'Struct({pformat(self.to_dict())})'
def copy(
self,
update: Optional[dict] = None,
) -> msgspec.Struct:
'''
Validate-typecast all self defined fields, return a copy of us
with all such fields.
This is kinda like the default behaviour in `pydantic.BaseModel`.
'''
if update:
for k, v in update.items():
setattr(self, k, v)
# roundtrip serialize to validate
return msgspec.msgpack.Decoder(
type=type(self)
).decode(
msgspec.msgpack.Encoder().encode(self)
)
def typecast(
self,
# fields: Optional[list[str]] = None,
) -> None:
for fname, ftype in self.__annotations__.items():
setattr(self, fname, ftype(getattr(self, fname)))
# proto
class SkynetRPCRequest(Struct):
uid: Union[str, int] # user unique id
method: str # rpc method name
params: dict # variable params
class SkynetRPCResponse(Struct):
result: dict
class ImageGenRequest(Struct):
prompt: str
step: int
width: int
height: int
guidance: int
seed: Optional[int]
algo: str
upscaler: Optional[str]
class DGPUBusRequest(Struct):
rid: str # req id
nid: str # node id
task: str
params: dict

8
test.sh 100755
View File

@ -0,0 +1,8 @@
docker run \
-it \
--rm \
--mount type=bind,source="$(pwd)",target=/skynet \
skynet:runtime-cuda \
bash -c \
"cd /skynet && pip install -e . && \
pytest tests/test_dgpu.py --log-cli-level=info"

55
tests/test_dgpu.py 100644
View File

@ -0,0 +1,55 @@
#!/usr/bin/python
import time
import json
import logging
import trio
import pynng
import trio_asyncio
from skynet_bot.dgpu import open_dgpu_node
from skynet_bot.types import *
from skynet_bot.brain import run_skynet
from skynet_bot.constants import *
from skynet_bot.frontend import open_skynet_rpc, rpc_call
def test_dgpu_simple():
async def main():
async with trio.open_nursery() as n:
await n.start(
run_skynet,
'skynet', '3GbZd6UojbD8V7zWpeFn', 'ancap.tech:34508')
await trio.sleep(2)
for i in range(3):
n.start_soon(open_dgpu_node)
await trio.sleep(1)
start = time.time()
async def request_img():
with pynng.Req0(dial=DEFAULT_RPC_ADDR) as rpc_sock:
res = await rpc_call(
rpc_sock, 'tg+1', 'txt2img', {
'prompt': 'test',
'step': 28,
'width': 512, 'height': 512,
'guidance': 7.5,
'seed': None,
'algo': 'stable',
'upscaler': None
})
logging.info(res)
async with trio.open_nursery() as inner_n:
for i in range(3):
inner_n.start_soon(request_img)
logging.info(f'time elapsed: {time.time() - start}')
n.cancel_scope.cancel()
trio_asyncio.run(main)

View File

@ -0,0 +1,22 @@
#!/usr/bin/python
import trio
import trio_asyncio
from skynet_bot.brain import run_skynet
from skynet_bot.frontend import open_skynet_rpc
from skynet_bot.frontend.telegram import run_skynet_telegram
def test_run_tg_bot():
async def main():
async with trio.open_nursery() as n:
await n.start(
run_skynet,
'skynet', '3GbZd6UojbD8V7zWpeFn', 'ancap.tech:34508')
n.start_soon(
run_skynet_telegram,
'5853245787:AAFEmv3EjJ_qJ8d_vmOpi6o6HFHUf8a0uCQ')
trio_asyncio.run(main)