t-abe

from __future__ import print_function

import sys

import chainer
from chainer.utils import conv


class ComputationalCostHook(chainer.function.FunctionHook):
    name = 'ComputationalCostHook'

coneybeare

on idle
        # Change these to your VPN's IP Address, and the VPN's name in your Mac's Network System Prefernces panel.
	set vpnIPAddress to "111.222.333.444"
	set vpnServiceName to "AlgoVPN"
	
	set myIP to "127.0.0.1"
	set shellScriptCommands to {¬
		"dig +short myip.opendns.com @resolver1.opendns.com", ¬
		"dig TXT +short o-o.myaddr.l.google.com @ns1.google.com | awk -F'\"' '{ print $2}'", ¬
		"curl ifconfig.me", ¬

tupui

"""Halton low discrepancy sequence.

This snippet implements the Halton sequence following the generalization of
a sequence of *Van der Corput* in n-dimensions.

---------------------------

MIT License

Copyright (c) 2017 Pamphile Tupui ROY

mbinna

Effective Modern CMake

Getting Started

For a brief user-level introduction to CMake, watch C++ Weekly, Episode 78, Intro to CMake by Jason Turner. LLVM’s CMake Primer provides a good high-level introduction to the CMake syntax. Go read it now.

After that, watch Mathieu Ropert’s CppCon 2017 talk Using Modern CMake Patterns to Enforce a Good Modular Design (slides). It provides a thorough explanation of what modern CMake is and why it is so much better than “old school” CMake. The modular design ideas in this talk are based on the book [Large-Scale C++ Software Design](https://www.amazon.de/Large-Scale-Soft

afspies

# EDIT 10/04/2022 - This version was provided by @jayelm who fixed some bugs and made the function much more robust

import os
import subprocess
import time

def assign_free_gpus(threshold_vram_usage=1500, max_gpus=2, wait=False, sleep_time=10):
    """
    Assigns free gpus to the current process via the CUDA_AVAILABLE_DEVICES env variable
    This function should be called after all imports,

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