yoavg

Reinforcement Learning for Language Models

Yoav Goldberg, April 2023.

Why RL?

With the release of the ChatGPT model and followup large language models (LLMs), there was a lot of discussion of the importance of "RLHF training", that is, "reinforcement learning from human feedback". I was puzzled for a while as to why RL (Reinforcement Learning) is better than learning from demonstrations (a.k.a supervised learning) for training language models. Shouldn't learning from demonstrations (or, in language model terminology "instruction fine tuning", learning to immitate human written answers) be sufficient? I came up with a theoretical argument that was somewhat convincing. But I came to realize there is an additional argumment which not only supports the case of RL training, but also requires it, in particular for models like ChatGPT. This additional argument is spelled out in (the first half of) a talk by John Schulman from OpenAI. This post pretty much

joe-niland

Notes on using Geodesic with repos created by cloudposse/reference-architectures

How to Use this Repo

AWS Setup

1. Log in to console for root account as your IAM user name@example.com
2. Go to https://console.aws.amazon.com/iam/home
3. Add an MFA device to your user
4. Generate an access key and store it securely to be added to aws-vault

bjorkqvist

Export images and video from Google Photos and adjust metadata

-- download images from 
takeout.google.com/settings/takeout

-- unzip

- flatten to new dir takeout2
exiftool -q -q -r -d takeout2 "-directory<filemodifydate" "-directory<createdate" "-directory<datetimeoriginal" Takeout

arikfr

1. Make sure the domain you picked points at the IP of your Redash server.
2. Switch to the root user (sudo su).
3. Create a folder named nginx in /opt/redash.
4. Create in the nginx folder two additional folders: certs and certs-data.
5. Create the file /opt/redash/nginx/nginx.conf and place the following in it: (replace example.redashapp.com with your domain name)

   upstream redash {
       server redash:5000;
   }
   

krishpop

// code courtesy of Toby team

chrome.storage.local.get("state", o => (
((f, t) => {
let e = document.createElement("a");
e.setAttribute("href", `data:text/plain;charset=utf-8,${encodeURIComponent(t)}`);
e.setAttribute("download", f);
e.click();
})(`TobyBackup${Date.now()}.json`, o.state)
));

kellyrmilligan

#!/bin/bash

if [[ "$1" != "" ]]; then
    S3BUCKETNAME="$1"
else
    echo ERROR: Failed to supply S3 bucket name
    exit 1
fi

aws s3 sync build s3://$S3BUCKETNAME --delete --cache-control max-age=31536000,public

wearhere

#!/bin/bash
#
# This script will convert a Meteor application into a Node.js application
# then deploy it to AWS Elastic Beanstalk.
#
# Run like `deploy_aws.sh my_eb_app my_eb_app-production`.
#
# That will deploy the Meteor application containing this script
# to the `my_eb_app-production` environment of the `my_eb_app` EB application.

nbremer

height: 600
license: mit
acknowledgement: Please add "Nadieh Bremer | Visual Cinnamon" to your credit when re-using this code, thank you!

jperl

###Icons

Name	Size
iphone_2x	120x120
iphone_3x	180x180
ipad	76x76
ipad_2x	152x152
android_ldpi	36x36
android_mdpi	48x48

rjames86

Last updated: 2017-03-18

Searching for Files

Find images in a directory that don't have a DateTimeOriginal

exiftool -filename -filemodifydate -createdate -r -if '(not $datetimeoriginal) and $filetype eq "JPEG"' .

###Output photos that don't have datetimeoriginal to a CSV### Note this can take a long time if you have a lot of jpgs