Discover gists

Anti-hype LLM reading list

Goals: Add links that are reasonable and good explanations of how stuff works. No hype and no vendor content if possible. Practical first-hand accounts of models in prod eagerly sought.

Foundational Concepts

Pre-Transformer Models

DeepSeek R1 Distill: Complete Tutorial for Deployment & Fine-Tuning

This guide shows how to deploy an uncensored DeepSeek R1 Distill model to Google Cloud Run with GPU support and how to perform a basic, functional fine-tuning process. The tutorial is split into:

Environment Setup
FastAPI Inference Server
Docker Configuration
Google Cloud Run Deployment
Fine-Tuning Pipeline (Cold Start, Reasoning RL, Data Collection, Final RL Phase)

NTLite Image Configuration Notes - Windows 11 - 23H2

April 2024
- updated with latest NTLite (2024.4.9880) and Windows 11 23H2 (Update 1) ISO (Feb 2024)
- Negate default Admin user password expiration
- apply latest security updates
- So far, no luck removing the onedrive backup nag message without completely disabling Onedrive notifications
Nov 2023
updated with latest NTLite and Windows 11 23H2 ISO (Oct 2023)

	# coding=utf-8
	# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
	#
	# Licensed under the Apache License, Version 2.0 (the "License");
	# you may not use this file except in compliance with the License.
	# You may obtain a copy of the License at
	#
	# http://www.apache.org/licenses/LICENSE-2.0
	#
	# Unless required by applicable law or agreed to in writing, software

	import asyncio
	import json
	from datetime import datetime

	from fasthtml.common import *
	from fasthtml.starlette import StreamingResponse
	from fasthtml.core import to_xml
	from datastar_py.sse import SSE_HEADERS, ServerSentEventGenerator

	import { integer, sqliteTableCreator, text } from 'drizzle-orm/sqlite-core'

	const sqliteTable = (tenant?: string) =>
	sqliteTableCreator(name => (tenant ? `${tenant}_${name}` : name))

	export const users = sqliteTable()('users', {
	id: integer('id', { mode: 'number' }).primaryKey({ autoIncrement: true }),
	name: text('name'),
	})

	/**
	* Configuration object for API settings.
	* Contains key constants used throughout the extraction process.
	*/
	const API_CONFIG = {
	key: "...", // Your Firecrawl API key
	baseUrl: 'https://api.firecrawl.dev/v1', // Base URL for Firecrawl API endpoints
	timeout: 30, // Timeout for HTTP requests in seconds
	maxAttempts: 10, // Maximum number of attempts to poll for job completion
	initialDelay: 1000 // Initial delay (in ms) before polling for job status

	# Check server process is using jemalloc
	# ref. [Japanese] https://tech.studyplus.co.jp/entry/2019/09/09/094140
	class AppController < ActionController::Base
	def show
	r = `strings /proc/#{Process.pid}/maps \| grep jemalloc`
	render plain: "<h1>#{ENV['LD_PRELOAD'].presence \|\| 'empty'}</h1><pre>#{r}</pre>"
	end
	end

	/*
	Collect and export a list of Twitter usernames you're following, along with follow-back status, to a CSV file every 80 seconds. Done client-side using Javascript running in the Console window of the web browser.

	Instructions:
	1) Modify the 'LANGUAGE' variable to match your Twitter interface language.
	2) Customize 'SKIP_USERS' with usernames you do not wish to include in the CSV.
	3) 'PRINT_FOLLOW_INFORMATION' determines whether follow-back status is included.
	4) 'SKIP_FOLLOWERS' and 'SKIP_NON_FOLLOWERS' settings filter out respective user groups.
	5) 'MS_PER_CYCLE' controls the frequency of data collection cycles.
	6) Load the Twitter page showing users you're following, open the console (F12), paste this code, and run it.