Speaker: Neha Sengupta

Built by G42’s Inception, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), and Cerebras, Jais is a bilingual LLM that has been trained on Arabic and English data, plus code in different programming languages.

This Tech Talk will be led by Inception’s Principal Applied Scientist, Neha Sengupta

Agenda:

• Why build an Arabic-Centric LLM?
• We know Arabic data is limited, how did you overcome this challenge?
• Other LLMs support Arabic. What makes Jais special?
• How was Jais trained and finetuned?
• What can users do with Jais?
• How can users try out Jais-chat?

Speaker: Cerebras Team

On Wednesday, August 9th, at 11 am PST, we will host a 30-minute tech talkwhere we will share details about Condor Galaxy 1 (CG-1), the first of nine interconnected 4 exaFLOPS AI supercomputers to be built in partnership with G42.

Agenda:

• Why are Cerebras and G42 building AI Supercomputers?
• Who needs this much compute? (hint: most organizations!)
• What is the problem with using a smaller machine?
• How can I get access?

Visit us to explore our leading-edge research into models, datasets and training strategies.

Speaker: Darius Lam, ML Product Manager

We will host a 30-minute tech talk where we will discuss large images, convolutional receptive fields, and their impact on medicine. This talk will be great for those interested in machine learning, life science, and their intersection!

Tech Talk Topics:

• What are receptive fields in deep learning?
• How do they work jointly with larger input sizes?
• How does this improve vision tasks on larger images?

This interactive Q&A will be hosted by Cerebras ML Product Manager Darius Lam.

Speaker: Udai Mody, Cerebras Hardware Product Manager

Tech Talk Topics:

• Why data pre-processing is important
• An analysis of RedPajama and SlimPajama
• What Cerebras did to pre-process this massive dataset
• How this impacts organizations

Speaker: Richard Kuzma, Senior Product Manager, NLP

Tech Talk Topics:

• Why proprietary data enables enterprises to build higher quality large language models
• Should your organization fine-tune pre-trained models or train from scratch
• What are the steps that need to be considered before training your first model

Speaker: Nick della Cioppa, HW/Systems Specialist

Tech Talk Topics:

• Why we built a big chip
• Supporting hardware that enables us to train 100B+ parameter models on a single system
• Cluster design that enables us to scale beyond 64 nodes with a few clicks

The Thirty-Sixth Annual Conference on Neural Information Processing Systems (NeurIPS 2022) is an interdisciplinary conference that brings together researchers in machine learning, neuroscience, statistics, optimization, computer vision, natural language processing, life sciences, natural sciences, social sciences, and other adjacent fields.

Cerebras will have a booth available for attendees to visit!