VikParuchuri

1. Algorithm Design and Analysis

One important aspect of algorithm design is problem-solving strategies. This involves breaking down a complex problem into smaller, more manageable subproblems. By solving these subproblems, we can then combine their solutions to solve the original problem. This approach is known as the divide-and-conquer method.

Another important aspect of algorithm design is understanding the time and space complexity of an algorithm. Time complexity refers to the amount of time an algorithm takes to run, while space complexity refers to the amount of memory an algorithm requires. By analyzing the time and space complexity of an algorithm, we can determine its efficiency and scalability.

For example, let's consider the problem of finding the largest number in a list. One possible algorithm is to iterate through the list and keep track of the largest number encountered so far. This algorithm has a time complexity of O(n), where n is the size of the list. This means that the algorithm's

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

cedrickchee

Some remarks on Large Language Models

Yoav Goldberg, January 2023

Audience: I assume you heard of ChatGPT, maybe played with it a little, and was impressed by it (or tried very hard not to be). And that you also heard that it is "a large language model". And maybe that it "solved natural language understanding". Here is a short personal perspective of my thoughts of this (and similar) models, and where we stand with respect to language understanding.

Intro

Around 2014-2017, right within the rise of neural-network based methods for NLP, I was giving a semi-academic-semi-popsci lecture, revolving around the story that achieving perfect language modeling is equivalent to being as intelligent as a human. Somewhere around the same time I was also asked in an academic panel "what would you do if you were given infinite compute and no need to worry about labor costs" to which I cockily responded "I would train a really huge language model, just to show that it doesn't solve everything!". We

innat

import tensorflow as tf

# credit: https://stackoverflow.com/a/66524901/9215780
class CustomTrainStep(tf.keras.Model):
    def __init__(self, n_gradients, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.n_gradients = tf.constant(n_gradients, dtype=tf.int32)
        self.n_acum_step = tf.Variable(0, dtype=tf.int32, trainable=False)
        self.gradient_accumulation = [tf.Variable(tf.zeros_like(v, dtype=tf.float32), 
                                                  trainable=False) for v in self.trainable_variables]

rajkrrsingh

$wget https://github.com/google/protobuf/releases/download/v2.5.0/protobuf-2.5.0.tar.bz2
$tar xvf protobuf-2.5.0.tar.bz2
$cd protobuf-2.5.0
$./configure CC=clang CXX=clang++ CXXFLAGS='-std=c++11 -stdlib=libc++ -O3 -g' LDFLAGS='-stdlib=libc++' LIBS="-lc++ -lc++abi"
$make -j 4 
$sudo make install
$protoc --version

diegopacheco

sudo apt-get install libboost-dev libboost-test-dev libboost-program-options-dev libboost-system-dev libboost-filesystem-dev libevent-dev automake libtool flex bison pkg-config g++ libssl-dev 
wget http://www-eu.apache.org/dist/thrift/0.9.3/thrift-0.9.3.tar.gz
./configure 
make
sudo make install
thrift --help 

esfand

Java Serialization & final class attributes

The SmartWeb BusinessObject class defines a protected attribute named logger carrying the logger for subclasses. The BusinessObject class implements Serializable thus it needs to define the logger attribute as transient because Commons Logging loggers are non serializable.

The problem arises whenever you deserialize a BusinessObject subclass because the logger attribute will not be deserialized (it has not be serialized at all!)

lyoshenka

How not to rm yourself

Copied from https://github.com/sindresorhus/guides/blob/master/how-not-to-rm-yourself.md

The rm command is inherently dangerous and should not be used directly. It can at worst let you accidentally remove everything. Here's how you can protect you from yourself.

Use trash

The trash command-line tool will move stuff to the trash instead of permanently deleting it. You should not alias rm to trash as it will break external scripts relaying on the behavior of rm. Instead use it directly: trash image.jpg.

dropwhile

virtualenv --no-site-packages hodor
hodor/bin/pip install simplejson ujson cbor tnetstring msgpack-python
curl -s 'http://www.json-generator.com/api/json/get/cvfsLVmKiG?indent=2' > test.json
hodor/bin/python shootout.py

Gekkio

import com.esotericsoftware.Kryo
import com.estoericsoftware.serialize.SimpleSerializer
import java.nio.ByteBuffer

object OptionSerializers {
  def register(kryo: Kryo) {
    kryo.register(classOf[None$], new NoneSerializer)
    kryo.register(classOf[Some[_]], new SomeSerializer(kryo))    
  }
}