Much of the challenge in machine learning engineering is fitting the complexity into your head. Best way to approach is with great patience & a desire to dig into any detail.
CODE
-
Understanding LLM Architecture: Transformers Tokenizers and Attention
By
–
Deepen your understanding of LLM architecture with an interview by AI educator, @jay . It covers generative aspects of transformers’ architecture, such as tokenizers, attention, and feed-forward networks. Ideal for beginners and enthusiasts.
-
GPT Monitoring: Token Optimization and Prompt Quality Strategies
By
–
Efficient GPT-based products require effective monitoring strategies. Keys to success: optimal token usage, constant prompt quality improvement, and averting user-facing disasters. Developers, remember these tips!
-
Google Colab Workshop: Browser-Based Python Development Tool
By
–
Check out these shots from the Colab workshop at Research@ NYC. Members from the Colab team provided attendees an opportunity to explore Colab, a Google Research tool that lets developers create & run Python code directly from their browser.
-
ML Progress Driven by Consensus on Truth
By
–
A lot of the rapid progress in ML can be attributed to how much easier it is to agree on the truth than in many disciplines. Still room for improvement with open data and code requirements for papers, but pretty good today!
-
Scaling Down AI Models: Optimization Strategies for 512×256
By
–
Really nice! Good idea scaling down to 512×256 too, I hadn’t thought to take it that low. Were there any other optimisations that could be pushed upstream?
-
Using function_call parameter to force function calling in requests
By
–
You have to use `function_call={"name": }` as a request param to force it to call a function.
-
Learn Machine Learning Online: Community Engagement and Endless Opportunities
By
–
Regardless of location or experience, Jay asserts the internet offers endless learning opportunities. Don't be put off if you're new to machine learning. Engage with online communities to share knowledge and gain feedback.
-
Jay Alammar Develops LLM University for Practical Applications
By
–
As @cohere
's director and engineering fellow, Jay is delving into Large Language Models (LLMs), making them accessible for practical use. He's developing LLM University to aid users in understanding and applying these models to address real-world issues. -
PyTorch Update Resolves Profiling Issues but CUDA 12.3 Installation Fails
By
–
The profiling problem eventually got mostly resolved by updating PyTorch, but I still get a failure to install nsight compute with the CUDA 12.3 package. Sigh. These are not the fun and rewarding parts of tech development.