aCandidMind

Documentation: Explaining Chat Templates for Transformers

Starting from Transformers version 4.42, the library requires a chat template for chat-based models. This configuration, defined in the Jinja format, specifies constraints for chat message inputs. Understanding and implementing these templates is crucial for optimal performance across various Large Language Models (LLMs) such as OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini.

What is a Chat Template?

A chat template is a configuration file that defines how chat messages are structured and passed to the underlying model. This structure ensures that the model receives well-defined inputs for optimal performance. The template is written in the Jinja templating language, allowing for flexible and reusable configurations.

Key Features of Chat Templates:

Message Role Specification: Defines whether a message is from the system, user, or assistant.
Input Formatting: Ensures the input adheres to the expected format by the ch

Cloud-Init Debugging

Cloud-init combined with terraform can be a powerful tool to provision instances on startup. Debugging scripts that are run by cloud-init however are not the easiest to debug.

Logs

Usually on an Ubuntu machine a lot of what is happening can be found in the syslog

cat /var/log/syslog

Use nginx as Non-Transparent SSL Proxy

Introduction

Many mobile apps have back-end API servers. They usually rely on the API replies to determine whether certain information is supposed to be shown. If the API responses could be manipulated on the fly, we may easily fool an unmodified app to expose some private data.

This manual guides you to set up nginx as non-transparent SSL proxy, which just subsitutes strings in the server responses (i.e. man-in-the-middle attack ourself). For both server-side (their API servers) and client-side (your device), the whole process is almost transparent.

Server Price Breakdown: DigitalOcean, Amazon AWS LightSail, Vultr, Linode, OVH, Hetzner, Scaleway/Online.net:

Permalink: git.io/vps

$5/mo

Provider	Type	RAM	Cores	Storage	Transfer	Network	Price

DB OpenData

DB OpenData Portal:

http://data.deutschebahn.com/

DB OpenData on GitHub:

https://github.com/dbopendata/

#Start a machine use any method you like here, I'm doing it from the command line 40 cores, 120GB RAM, SSD with 8,500 IOPS (SAN) - Amazon linux (Centos)

ec2-run-instances ami-a10897d6 -t m4.10xlarge -g jlp -k john_page_demos -b "/dev/xvdb=:725:true:io1:8500"

#Log on

ssh -L 27017:localhost:27017 -i <your public key>.pem ec2-user@<your ip address>

#Set up disk

	// dm Klopapier Widget
	//
	// Copyright (C) 2020 by marco79 <[email protected]>
	//
	// Permission to use, copy, modify, and/or distribute this software for any purpose with or without fee is hereby granted.
	//
	// THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL
	// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
	// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER
	// IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE

	require 'webrick'
	require 'fileutils'

	BIND_ADDRESS = "127.0.0.1"
	PORT = 8080

	BACKUP_DIR = 'bak'

	if ARGV.length != 0
	root = ARGV.first.gsub('\\', '/')

	#!/usr/bin/env bash

	# Script to (selectively) save/load multiple Docker images to/from a directory.
	# Run ./save-load-docker-images.sh for help.

	set -e

	directory=$PWD
	filter=""
	compress=0

	#!/bin/bash
	/usr/sbin/usermod -aG docker <user>