Fernando MrCordeiro

Maybe you've heard about this technique but you haven't completely understood it, especially the PPO part. This explanation might help.

We will focus on text-to-text language models 📝, such as GPT-3, BLOOM, and T5. Models like BERT, which are encoder-only, are not addressed.

Reinforcement Learning from Human Feedback (RLHF) has been successfully applied in ChatGPT, hence its major increase in popularity. 📈

RLHF is especially useful in two scenarios 🌟:

You can’t create a good loss function
- Example: how do you calculate a metric to measure if the model’s output was funny?
You want to train with production data, but you can’t easily label your production data

🤗 Stable Diffusion for AMD GPUs on Windows using DirectML

Requirements

This could be a problem cause by your Windows OS being a bottleneck and limit your speed. After some talks and discussion with a colleague, this is the solution that worked for my case!

From Windows

Go to your Start Menu.
Go to 'View Network Connections'.
Right-click on your Wi-Fi adapter.
Select the 'Properties' menu option.
Find the 'Internet Protocol Version 4' and select the 'Properties' menu option. (keep the IPv4 option checked)
Click on the 'Advanced' button.
Uncheck 'Automatic Metric' and enter 1 in the 'Interface metric' text box.

Steps:

Create private Feed in Azure DevOps
Create Personal Access Token (PAT) in Azure DevOps with Feed Read permission (details)
Navigate to the Azure DevOps Artifacts Feed page where you can see the details for the next steps (you'll need feed name, project name, organization name and later also package name):
Create Build pipeline in Azure DevOps to create package and push to private feed:

trigger:

Introductory resources

High level overview https://yogthos.github.io/ClojureDistilled.html
An Animated Introduction to Clojure https://markm208.github.io/cljbook/
Interactive tutorial in a browser https://tryclojure.org/
Interactive exercises http://clojurescriptkoans.com/
Clerk notebooks with introductory examples https://github.clerk.garden/anthonygalea/notes-on-clojure
More interactive exercises https://4clojure.oxal.org/
Lambda Island tutorials https://lambdaisland.com/
Functional Programming with Clojure resources https://practicalli.github.io/

Running jupyter notebook with nohup

This is helpful when running jupyter notebook on remote server (AWS) and tunneling into it, because with nohup the notebook process continues to run on the server even if you get disconnected from it (any running notebook code continues to run on the server without interruption, and just needs to be re-tunneled into).

Start jupyter notebook on remote server

In the remote server terminal, run:

nohup jupyter notebook &amp;

Note: this is a summary of different git workflows putting together to a small git bible. references are in between the text

How to Branch

try to keep your hacking out of the master and create feature branches. the [feature-branch workflow][4] is a good median between noobs (i have no idea how to branch) and git veterans (let's do some rocket sience with git branches!). everybody get the idea!

	pragma solidity ^0.5.0;

	contract Token {
	string public name = "DApp Token";
	string public symbol = "DAPP";
	uint256 public totalSupply = 1000000000000000000000000; // 1 million tokens
	uint8 public decimals = 18;

	event Transfer(
	address indexed _from,

-<html>
-<head>
-	<title>remove last row from a table</title>
-	<script language="JavaScript" type="text/javascript">
-	<!--
-	var rws;
-	function RemoveRow(obj){
-		obj=document.getElementById(obj);
-		rws=obj.getElementsByTagName('TR');
-		obj.removeChild(rws[rws.length-1]);

	# http://kb.detlus.com/articles/git/how-to-update-remote-branch-list-on-local-machine/

	git remote update origin --prune