Uploaded May 2023 | Updated September 2026, 2 weeks ago
In this video, I compare the cost/performance of AWS Trainium with the NVIDIA V100 GPU.
I first launch a trn1.32xlarge instance (16 Trainium chips) and a p3dn.24xlarge (8 V100s). Then, I run 3 benchmarks: language pretraining with GPT2, token classification with BERT Large, and image classification with the Vision Transformer
The results? Trainium is 2 to 5x faster, and 3 to 8x cheaper!
⭐️⭐️⭐️ Don't forget to subscribe to be notified of future videos ⭐️⭐️⭐️
- Amazon EC2 trn1: aws.amazon.com/ec2/instance-types/trn1
- Amazon EC2 p3: aws.amazon.com/ec2/instance-types/p3
- Training commands: gist.github.com/juliensimon/da64fc6d6a2fe39bd8c5af12389a227e
- Trainium with Optimum Neuron: youtu.be/FmjTWags__Q
- Trn1 vs G5 benchmark: youtu.be/2SquGhkld7k
In this video, I compare the cost/performance of AWS Trainium with the NVIDIA V100 GPU.
I first launch a trn1.32xlarge instance (16 Trainium chips) and a p3dn.24xlarge (8 V100s). Then, I run 3 benchmarks: language pretraining with GPT2, token classification with BERT Large, and image classification with the Vision Transformer
The results? Trainium is 2 to 5x faster, and 3 to 8x cheaper!
⭐️⭐️⭐️ Don't forget to subscribe to be notified of future videos ⭐️⭐️⭐️
- Amazon EC2 trn1: aws.amazon.com/ec2/instance-types/trn1
- Amazon EC2 p3: aws.amazon.com/ec2/instance-types/p3
- Training commands: gist.github.com/juliensimon/da64fc6d6a2fe39bd8c5af12389a227e
- Trainium with Optimum Neuron: youtu.be/FmjTWags__Q
- Trn1 vs G5 benchmark: youtu.be/2SquGhkld7k








![How Witty Works leverages Hugging Face to scale inclusive language
During this webinar, Elena Nazarenko, Lead Data Scientist at Witty Works, Lukas Kahwe Smith, CTO & Co-Founder at Witty Works and Julien Simon, Chief Evangelist at Hugging Face, discuss how Witty Works leverages Hugging Face to scale inclusive language.
[No HD version, sorry]
- The impact of Transformers on text classification use cases
- How Witty Works leverages Hugging Face to scale inclusive language
- How to perform domain-adaptive pretraining on a transformer model
Speakers
Elena Nazarenko - Lead Data Scientist at Witty Works
Lukas Kahwe Smith - CTO & Co-Founder at Witty Works
Julien Simon - Chief Evangelist at Hugging Face
About Witty Works
Witty is a Digital Writing Assistant for Inclusive Language that enables organizations to detect their own bias, in writing and in behavior, and fix it. Because language builds culture.
About Hugging Face
Hugging Face is a wildly popular community-based repository for open-source ML technology. It is a platform that stores, serves and manages the latest and greatest in open-sources ML models, including enabling customers to fine-tune these models and deploy them at scale.
Hugging Face is one of the most used platforms and is empowering 10,000 companies to integrate artificial intelligence into their products or workflows. How Witty Works leverages Hugging Face to scale inclusive language](https://i.ytimg.com/vi/Z_S1gfRFtgA/mqdefault.jpg)

