The Morning Paper

JUNE 13, 2019

As we saw with the SOAP paper last time out, even with a fixed model variant and hardware there are a lot of different ways to map a training workload over the available hardware. Different hardware architectures (CPUs, GPUs, TPUs, FPGAs, ASICs, …) offer different performance and cost trade-offs.

Hardware 58

Hardware Latency Serverless Energy 58

The Morning Paper

JUNE 27, 2019

The thrust of the argument is that there’s a chain of inter-linked assumptions / dependencies from the hardware all the way to the programming model, and any time you step outside of the mainstream it’s sufficiently hard to get acceptable performance that researchers are discouraged from doing so. GPU and TPU).

Systems 87

Systems Tuning Programming Innovation 87

Sutter's Mill

NOVEMBER 11, 2023

This week’s meeting: Meeting #2 of C++26 At the previous meeting in June, the committee adopted the first 40 proposed changes for C++26, including many that had been ready for a couple of meetings and were just waiting for the C++26 train to open to be adopted. For those highlights, see the previous trip report. This is that library.

C++ 107

C++ Lambda Network Programming 107

Adrian Cockcroft

NOVEMBER 13, 2023

In the simplest case, you have a growing workload, and you optimize it to run more efficiently so that you don’t need to buy or rent additional hardware, so your carbon footprint stays the same, but the carbon per transaction or operation is going down. I’ve written before about how to tune out retry storms.

Energy 52

Energy Cloud Traffic Transportation 52

O'Reilly

JANUARY 16, 2024

software” rather than “hardware” in our brains). Most notably, ChatGPT is tuned to be overly-friendly and overly-agreeable (sounding like a customer service agent) so it will simply go along with whatever you assert. would do if I were on a noisy phone connection with someone and didn’t hear them clearly.

Entertainment 73

Entertainment Internet Internet Design 73

O'Reilly

NOVEMBER 28, 2023

Even with cloud-based foundation models like GPT-4, which eliminate the need to develop your own model or provide your own infrastructure, fine-tuning a model for any particular use case is still a major undertaking. Training models and developing complex applications on top of those models is becoming easier. of nonusers, 5.4%

Open Source 108

Open Source Artificial Intelligence Google Infrastructure 108

O'Reilly

SEPTEMBER 19, 2023

A company then needed to train up their ops team to manage the cluster, and their analysts to express their ideas in MapReduce. Doubly so as hardware improved, eating away at the lower end of Hadoop-worthy work. Google goes a step further in offering compute instances with its specialized TPU hardware.

Hardware 99

Hardware Storage Big Data Blockchain 99

All Things Distributed

MARCH 19, 2018

For example, training on more data means more accurate models. Last re:Invent, to make the problem of authoring, training, and hosting ML models easier, faster, and more reliable, we launched Amazon SageMaker. Machine learning models are usually trained tens or hundreds of times. In machine learning, more is usually more.

Scalability 159

Scalability Hardware AWS Tuning 159

The Netflix TechBlog

APRIL 29, 2019

Such applications track the inventory of our network gear: what devices, of which models, with which hardware components, located in which sites. Our Infrastructure Security team leverages Python to help with IAM permission tuning using Repokid. We leverage Python to protect our SSH resources using Bless.

Open Source 279

Open Source Network Infrastructure Big Data 279

O'Reilly

OCTOBER 11, 2022

Andrew Ng , Christopher Ré , and others have pointed out that in the past decade, we’ve made a lot of progress with algorithms and hardware for running AI. Our current set of AI algorithms are good enough, as is our hardware; the hard problems are all about data. Was it trained using biased, unfair data? Is retraining needed?

Games 116

Games Healthcare Processing Retail 116

Smashing Magazine

AUGUST 22, 2021

As such, tuning congestion logic is usually only done by a select few developers, and evolution is slow. Many network interface controllers (NICs) even have built-in hardware-offload features for TCP. We can also expect QUIC-specific hardware to become available. Finally convinced?

Performance 105

Performance Network Latency Servers 105

All Things Distributed

NOVEMBER 30, 2016

Effectively applying AI involves extensive manual effort to develop and tune many different types of machine learning and deep learning algorithms (e.g. automatic speech recognition, natural language understanding, image classification), collect and clean the training data, and train and tune the machine learning models.

AWS 165

AWS Lambda Artificial Intelligence Mobile 165

High Scalability

NOVEMBER 25, 2019

Once it's resolved our pipeline takes over and the ticket catches the next release train. The next release train is verified by automated REST API tests and some manual smoke tests. How are software and hardware upgrades rolled out? A merge request has to be approved by 2 engineers and only then JIRA ticket can be resolved.

Architecture 52

Architecture Cache Azure Storage 52