rom1504

import json
import pandas as pd
import subprocess
import sys

def get_msg(
    backticks=True   # whether to add backticks for Discord formatting or not
    ):
    "gets a list of cluster usage from squeue and creates a text message from it"
    a = json.loads(subprocess.check_output(['squeue','--json']).decode("utf8"))

rom1504

Change paths then run keep_evaling.sh

Will keep evaluating openclip during training

to send to wandb, can also run while [ 1 ]; do python3 eval_to_wandb.py; sleep 300; done

rom1504

create an env:

python3.8 -m venv .env
source .env/bin/activate
pip install -U pip
pip3 install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu113

fix the paths in simple.sh

rom1504

Install

git clone https://github.com/mlfoundations/open_clip.git
cd open_clip
python3.8 -m venv .env
source .env/bin/activate
pip install -U pip
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
pip install -e .

rom1504

class WDSDataset(data.IterableDataset):
    def __init__(self, min_size, transform=None, target_transform=None):
        self.min_size = min_size
        self.transform = transform if transform is not None else nn.Identity()
        self.target_transform = target_transform if target_transform is not None else nn.Identity()
        self.kv = OnDiskKV(file='/home/ubuntu/laion5B-watermark-safety-ordered', key_format='q', value_format='ee')
        self.kv_aesthetic = OnDiskKV(file='/home/ubuntu/laion5B-aesthetic-tags-kv', key_format='q', value_format='e')
        self.pwatermark_threshold = 0.8
        self.punsafe_threshold = 0.5
        self.aesthetic_threshold = 5.

rom1504

https://wandb.ai/rom1504/dalle2_train_decoder/runs/mic5buox/files/decoder_config.json

get dalle2

get the config file

get these 2 .sh

run sbatch start_big.sh

rom1504

import pandas as pd

df = pd.read_parquet("aethetic_multi/0000.parquet")

buckets = [(i, i+1) for i in range(10)]

html= "<h1>Aesthetic subsets in Laion2B-multi</h1>"
for [a,b] in buckets:
    total_part = df[(df["prediction"] >= a) & (df["prediction"] <= b)]
    count_part = len(total_part) / len(df) * 100

rom1504

A red colored car.
A black colored car.
A pink colored car.
A black colored dog.
A red colored dog.
A blue colored dog.
A green colored banana.
A red colored banana.
A black colored banana.
A white colored sandwich.

rom1504

[
  "t-shirt, size M",
  "flower dress, size M",
  "a t-shirt of an avocado",
  "a rainbow hat",
  "white snow covered mountain under blue sky during daytime",
  "aerial view of the beach during daytime",
  "aerial view of the beach at night",
  "double rainbow over a lake",
  "a beautiful sunset at a beach with a shell on the shore",

rom1504

from glob import glob
import time
import datetime

def get_read_bytes():
    return sum([int(open(f"{p}/ports/1/hw_counters/rdma_read_bytes", "r").read().strip()) for p in glob("/sys/class/infiniband/*")])
from os.path import expanduser
home = expanduser("~")