Mark Tenenholtz
Head of AI @PredeloHQ. XGBoost peddler, transformer purveyor. Prev. forecasting @Kroger
View on πThreads
LLM sizes, and when to use them: 100M-500M param, encoder-only: you have a straightforward classification/regression task, or you need local embeddings for reasonably sized datase...
Python is removing the GIL. The GIL (Global Interpreter Lock) prevents you from running multi-threaded code. That makes ML code, in particular, really hard to write in pure Pytho...
I built a ChatGPT app that lets you chat with any codebase! 99% of projects just copy/paste Langchain tutorials. This goes well beyond that. Here's how I built it:
I am at least 3-5x more productive using ChatGPT to code. Not only am I faster writing code I'm familiar with, but I've even shipped apps in tech stacks I'd never used before. He...
Thereβs now a Python library for RLHF called TRLX! (The same reinforcement learning strategy used in training ChatGPT) It works well with Hugging Face models, supports multiple R...
GitHub Copilot is a genuinely brilliant piece of engineering sitting on top of GPT-3. Here's how it works, courtesy of @parth007_96. Huge shoutout to @alexgraveley and the Copilo...
Over the last year, I've spent 100's of hours training transformers for NLP. I went back over my most successful projects and competitions and distilled them into a solid, repeata...
I've spent 100's of hours training computer vision models. I've revisited my code and notes from those projects and distilled them into a repeatable process that anyone can follow...
Most data scientists use linear/logistic regression to figure out which features are important in a dataset. I almost never do this. Instead, I generally use leave-one-out featur...
College ML usually have major holes. Static courses are rarely up-to-date in a field moving this fast, plus they usually donβt teach the practical skills industry expects you to h...
I've shared tons of free/inexpensive material for learning data science and ML (and not just the basics). Combined, their value is easily greater than a $100k+ data science degree...
20 simple ideas I would tell anyone learning data science: 1. Building intuition for how techniques work is the most powerful tool you have. 2. A great individual model is better...