VictorTaelin

A::B is a system with 4 tokens: `A#`, `#A`, `B#` and `#B`.

An A::B program is a sequence of tokens. Example:

    B# A# #B #A B#

To *compute* a program, we must rewrite neighbor tokens, using the rules:

    A# #A ... becomes ... nothing
    A# #B ... becomes ... #B A#

odellus

"""
Basically a carbon copy of
https://github.com/stanfordnlp/dspy/blob/main/examples/math/gsm8k/CoT.ipynb

I downloaded their optimized few shot prompt
https://github.com/stanfordnlp/dspy/blob/main/examples/math/gsm8k/turbo_8_8_10_gsm8k_200_300.json

This code also only works with this branch of DSPy
https://github.com/odellus/dspy/tree/quantize-hf-w-bnb
"""

hayden-donnelly

# Example usage: 
# python make_crop_labels.py --input_path data/images --output_path --data/cropped_images --csv_path data/crops.csv

# Controls:
# scroll to change crop size, mouse to aim the crop, left click to crop image and move to next, x to skip to the next image.

# The script is pretty messy since I quickly hacked it together with little regard for quality, but it works.

import pygame
import argparse

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

rain-1

Purpose

Bootstrap knowledge of LLMs ASAP. With a bias/focus to GPT.

Avoid being a link dump. Try to provide only valuable well tuned information.

Prelude

Neural network links before starting with transformers.

YukiSnowy

// g++ *.cpp -o vulkan -lSDL2main -lSDL2 -lvulkan-1
// https://vulkan-tutorial.com/

#include <iostream>
using namespace std;

#include <SDL2/SDL.h>
SDL_Window *window;
char* window_name = "example SDL2 Vulkan application";

seLain

# assume :
# - openproject installed in /opt/openproject
# - local port: 6000
# - external port: 6020

server {
   listen 6020;
   server_name SERVER_DOMAIN_NAME;
   root /opt/openproject/public;