|
|
|
|
The AWS team responsible for building Inferentia and Trainium products has officially become MLOps Community sponsors. Thanks to their commitment, we can continue to foster the global community. In fact, the core team has grown to four people! We have big plans for the second half of the year, including community courses, industry reports, and in-person meetings around
the world. If you want to get involved, reach out to us. You might also want to start using AWS Neuron, the SDK for running deep learning workloads on AWS Inferentia and AWS Trainium-based instances. Check out their documentation here get started using Hugging Face Optimum Neuron.
|
|
|
|
|
|
|
“I spent years building News Article Classification models. Then, we were able to deprecate the whole thing and rewrite the whole system in a day.
And I'm not kidding. Like, in one day.
All by leveraging large language models like OpenAI.”
That’s one of my favorite moments from the most recent coffee session podcasts where we spoke with Thibaut Labarre, the engineering lead at AngelList.
We also discussed how they got around the dreaded OAI rate limit, augmenting existing products with AI, and how prompts are his secret sauce.
Check out the whole convo below
|
|
|
Job of the week
Senior Engineer (“battlefield
architect”) // DataJoint - Delivering data integrity, speed, and scalability for science labs engaged in the grand challenge of reverse engineering the brain.
Their platform defines and operates custom research pipelines: orchestrating computing, managing change, and safeguarding the reproducibility of results.
You’ll collaborate closely with their team of engineers, neuroscientists, and data scientists, developing robust and highly scalable solutions using modern technology stacks. As a technical mentor, you will guide the team in creating and extending orchestration and workflow systems to run scientific computing tasks in a stable, repeatable way.
|
|
|
|
|
|
|
TimeGPT So hot, it hasn’t been released yet. The first time series foundational model. What? The folks at Nixtla trained
the model on the largest collection of public time series, consisting of 100 billion data points from more than 10 domains. This is some crazy shit. With TimeGPT, users and practitioners can obtain accurate zero-shot forecasts in seconds through their API or Python SDK. Additionally, the model can provide prediction intervals and historical forecasts, can be fine-tuned to the user's data to further improve accuracy, and can incorporate exogenous information provided by the user such as prices and weather. Before the team releases it, they wanted to give a selected group of early supporters access to test it and provide them with feedback. I have been a fan of what Max and the team have been doing for some time now and wanted to see if the community could be on the beta tester list. If you are up for trying it, please give feedback, and refrain from publicly mentioning it until they officially release it. Follow The White Bunny
|
|
|
|
|
|
|
Accelerating ML Deployment with Orchestration Systems
Justin kicks off the episode by sharing insights from his experience at Etsy. Their ML journey involves pre-processing, modeling, serving, and ongoing maintenance. He sheds light on the time-consuming nature of ML projects and the challenges faced by data scientists when transitioning their
models into production. To illustrate these challenges, Justin presents a compelling case study from Etsy's risk team, where they encountered difficulties in putting a fraud detection model into production. He discusses the five bottlenecks they encountered, including latency, scaling, failures, and observability. Justin shares his experiences in solving similar problems at previous companies like BuzzFeed and Twitter, teasing the innovative solution they are currently developing at Etsy called Komodo. But what is Komodo? As Justin explains there are three components of Komodo: the model interpreter, feature fetching, and inference process. He highlights the importance of encapsulating each step as a single deployment unit to streamline the workflow. Watch Now
|
|
|
|
|
|
|
Finding Harmony in MLOps: Balancing Functional and Object-Oriented Approaches
Mederic does it again with his latest blog post on balancing OOP and Functional Programming methods in MLOps. He goes through both the pros and cons of Object-Oriented Programming and the pros and
cons of Functional Programming.
And... We don't stop there.
What are the MLOps application requirements? He breaks down his list in order of importance:
- Reproducibility
- Modularity
- Configurability
- Extensibility
- Keep It Simple (KISS)
Mederic drives it home with the selection of styles for your project and how to enhance OOP in Python.
If only it could just be straightforward…..
Read Now
|
|
|
|
|
Add your profile to our jobs board here
|
|
|
|
|
|
|
|
|
Thanks for reading. This issue was written by Demetrios and edited by Jessica Rudd. See you in Slack, Youtube, and podcast land. Oh yeah, and we are also on Twitter if you like chirping birds.
|
|
|
|
|
