Felt the urge to write about computer vision for a change! Summarizing the main themes from CVPR:
1) Vision transformers
2) Diffusion models
3) NeRFs
4) Object detection & segmentation Recapping it in more detail along with some paper highlights here:
@rasbt
-
CVPR 2023 Main Themes: Vision Transformers, Diffusion Models, NeRFs
By
–
-
Recommendation for RLHF Technical Tutorial
By
–
Great collection! And after the technical tutorials, I can highly recommend “How RLHF actually works”
-
What’s wrong with miniconda/miniforge?
By
–
What's wrong with miniconda/miniforge? https://
github.com/rasbt/scipy202
3-deeplearning/tree/main/00-1_python-setup-guide
… -
Open-sourcing LLaMa weights constrained by dataset legal concerns
By
–
Yup, totally agree. That’s probably it. Afaik the reason why they couldn’t open-source the LLaMa weights was the dataset. So instead of scraping data from elsewhere and staying on shaky legal grounds, getting data from your own platform is probably easiest.
-
Sentiment Analysis Goes Beyond Binary Classification
By
–
Sentiment analysis is not just a binary classification task. It's kind of like saying [insert any disease here] research is just a binary classification problem.
-
New scikit-learn Version Released with PyTorch Support and Features
By
–
It doesn't get much attention these days (in both senses) but a new version of @scikit_learn
, my favorite machine learning library, is out! – PyTorch support for LinearDiscriminant Analysis
– Validation Curves
– Decision tree with N/A features
– and more -
Appreciation for MPT Article and FSDP Implementation Question
By
–
Thanks for writing & sharing your latest article on MPT. Really well written and has just the right level of detail .
One question though. You wrote > "The entire training framework is based upon PyTorch’s Fully Sharded Data Parallel (FSDP) package and uses no pipeline or -
Stable Diffusion Based on Latent Diffusion Models Knowledge
By
–
I haven't read the full article, yet, and I also don't know much about the person in question (not familiar with any statements besides the occasional tweets that pop up on my timeline). That being said, I thought it was known that Stable Diffusion is based on the Latent
-
Stage 3 Offloading and CPUAdam Performance Comparison with FSDP
By
–
Relatively similar. I think stage 3 with offloading and CPUAdam was even a tad better but I’d have to double check again on Wed when I am back at my computer. I usually use DeepSpeed but opted for FSDP here to reduce external dependencies.
-
Why Transformers and Self-Attention Over Convolutional or RNN Layers
By
–
Sure. But why specifically for transformer layers and self-attention, not say convolutional or RNN layers?