catwell

Computed on the validation set of ImageNet 256 (50000 images).

model	encoder	decoder	rFID	LPIPS	PSNR
SD 1.5	base	base	0.9275	0.0753	25.35
SD 1.5	TAE	base	2.5711	0.1109	23.89
SD 1.5	base	TAE	2.9040	0.0995	23.47
SD 1.5	TAE	TAE	3.8339	0.1092	23.52
sdxl	base	base	0.7690	0.0703	25.60
sdxl	TAE	base	1.4093	0.0888	24.83

sayakpaul

Running Flux under limited resources with Diffusers

Flux: https://blackforestlabs.ai/announcing-black-forest-labs/

• Run Flux with quantization by AmericanPresidentJimmyCarter.
• Run Flux on a 24GB 4090 by decoupling the different stages of the pipeline
• Running with torchao
• Running with NF4

The first resource even allows you to run the pipeline under 16GBs of GPU VRAM.

madebyollin

def add_profiling_markers(model):
    """Monkey-patch profiling markers into an nn.Module.
    
    Args:
        model: an nn.Module
    
    Effect:
        all model.named_module() forward calls get wrapped in their
        own profiling scope, making traces easier to understand.
    """

mrsteyk

Memes

UPD8: clean-up and it runs now, cba'd to actually test it demo below. Fixes are from gh:madebyollin, check out his gist which actually works

weights now with ready to use checkpoint (will be up in few minutes)

Confirmed working with this code:

import time

kconner

macOS Internals

Understand your Mac and iPhone more deeply by tracing the evolution of Mac OS X from prelease to Swift. John Siracusa delivers the details.

Starting Points

How to use this gist

You've got two main options:

tomhicks

/*
Copy this into the console of any web page that is interactive and doesn't
do hard reloads. You will hear your DOM changes as different pitches of
audio.

I have found this interesting for debugging, but also fun to hear web pages
render like UIs do in movies.
*/

const audioCtx = new (window.AudioContext || window.webkitAudioContext)()

darconeous

Tesla Key Card Protocol

Researched by Robert Quattlebaum darco@deepdarc.com.

Last updated 2020-02-03.

ryan-robeson

Building Resynthesizer for GIMP 2.10.10 on macOS - V2

Feedback is welcome and appreciated.

• If you'd prefer to do things by hand check out Version 1 of this guide.

The following documents the steps I took to compile the latest version of Resynthesizer on macOS Sierra (10.12.6) for GIMP 2.10.10.

Homebrew was used to fulfill as many dependencies as possible.

TravisDunlop

'''
Freezing a stable-baselines to a frozen protocol buffer file to be served.
https://github.com/hill-a/stable-baselines

Some code taken from this lovely blog series
https://blog.metaflow.fr/tensorflow-how-to-freeze-a-model-and-serve-it-with-a-python-api-d4f3596b3adc
'''

import tensorflow as tf
import os

jdarpinian

///$(which true);FLAGS="-g -Wall -Wextra --std=c17 -O1 -fsanitize=address,undefined";THIS_FILE="$(cd "$(dirname "$0")"; pwd -P)/$(basename "$0")";OUT_FILE="/tmp/build-cache/$THIS_FILE";mkdir -p "$(dirname "$OUT_FILE")";test "$THIS_FILE" -ot "$OUT_FILE" || $(which clang || which gcc) $FLAGS "$THIS_FILE" -o "$OUT_FILE" || exit $?;exec bash -c "exec -a \"$0\" \"$OUT_FILE\" $([ $# -eq 0 ] || printf ' "%s"' "$@")"
#include <stdio.h>

int main() {
  printf("Hello world!\n");
  return 0;
}