willccbb

# train_grpo.py
#
# See https://github.com/willccbb/verifiers for ongoing developments
#
import re
import torch
from datasets import load_dataset, Dataset
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import LoraConfig
from trl import GRPOConfig, GRPOTrainer

veekaybee

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

srikumarks

function mandelbrot_kernel(c, max_iter)
    z = c
    for i in 1:max_iter
        z = z * z + c
        if abs2(z) > 4
            return i-1
        end
    end
    return max_iter

eugeneyan

Mandelbrot in Mojo with Python plots

Not only Mojo is great for writing high-performance code, but it also allows us to leverage huge Python ecosystem of libraries and tools. With seamless Python interoperability, Mojo can use Python for what it's good at, especially GUIs, without sacrificing performance in critical code. Let's take the classic Mandelbrot set algorithm and implement it in Mojo.

We'll introduce a Complex type and use it in our implementation.

Mandelbrot in python

danielgross

import requests
import time
import os
import sys
import openai
import tiktoken
from termcolor import colored

openai.api_key = open(os.path.expanduser('~/.openai')).read().strip()

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

virattt

osher

# This demo uses an alpine sandbox in a docker container in 
# interactive mode. ran with:
#    docker run --rm -it alpine
#
# if you run it on your own system:
#  1. you should use your own package manager instead of `apk`
#  2. expect the following left overs:
#    - installed binaries (age, age-keygen, sops)
#    - $HOME/.config/sops/age/keys.txt 
#    - demo files: source.env, encrypted.env, decrypted.env

orihomie

BUCKET_NAME=terraform-your_company-remote-store # this should be unique, and by that I mean really UNIQUE
BUCKET_REGION=eu-central-1
USER_NAME=terraform-deployer
POLICY_FILE_NAME=$PWD/policy.json
AWS_PROFILE=your_company

aws s3api create-bucket \
    --profile $AWS_PROFILE \
    --bucket $BUCKET_NAME \
    --region $BUCKET_REGION \

Mishco

Setup HashiCorp Vault on docker

Vault secures, stores, and tightly controls access to tokens, passwords, certificates, API keys, and other secrets in modern computing. Vault is primarily used in production environments to manage secrets. Vault is a complex system that has many different pieces. There is a clear separation of components that are inside or outside of the security barrier. Only the storage backend and the HTTP API are outside, all other components are inside the barrier.

Figure 1: Architecture of Vault and Spring App (Click to enlarge)

The storage backend is untrusted and is used to durably store encrypted data. When the Vault server is started, it must be provided with a storage backend so that data is available across restarts. The HTTP API similarly must be started by the Vault server on start so that clients can interact with it.