Recent Advances in Language Model Fine-tuning
— Sebastian Ruder (@seb_ruder) February 24, 2021
New blog post that takes a closer look at fine-tuning, the most common way large pre-trained language models are used in practice.https://t.co/A5KYoq5zuw
Recent Advances in Language Model Fine-tuning
— Sebastian Ruder (@seb_ruder) February 24, 2021
New blog post that takes a closer look at fine-tuning, the most common way large pre-trained language models are used in practice.https://t.co/A5KYoq5zuw
20 hours of new lectures on Deep Learning and Reinforcement Learning with lots of examples https://t.co/3cXv2E0rcd
— /MachineLearning (@slashML) February 23, 2021
I made a silly game: try to guess if a C/C++ code snippet is real or GPT2-generated: https://t.co/Rgfk7Hw0h0
— Brendan Dolan-Gavitt (@moyix) February 23, 2021
Excited to present Clockwork VAEs for video prediction!
— Danijar Hafner (@danijarh) February 23, 2021
Clockwork VAEs (CW-VAEs) leverage hierarchies of latent sequences, where higher levels tick slower. They learn long-term deps across 1000 frames, semantically separate content, and outperform strong video models.
👇 Thread pic.twitter.com/Cn11MAKjIB
This is the AutoML debate all over again. No, you can't replace your data scientists with AutoML code -- what are you going to do when it doesn't work?
— Hilary Mason (@hmason) February 22, 2021
Same for prompt engineering vs ML engineering. If you're building these systems you need to understand them. https://t.co/nj1kQTetiw
The new SOTA is in Transformers! DeBERTa-v2 beats the human baseline on SuperGLUE and up to a crazy 91.7% dev accuracy on MNLI task.
— Hugging Face (@huggingface) February 22, 2021
Beats T5 while 10x smaller!
DeBERTa-v2 contributed by @Pengcheng2020 from @MSFTResearch
Try it directly on the hub: https://t.co/HhlL5WrJxp pic.twitter.com/fcUUCiKE0z
Want to use Hugo for your website but hate that you can’t get the themes to look *exactly* how you want them? Check out my new blog post: Hugo for Fussy People! In it I go through how to turn a blank theme into precisely what you want. Fuss away!https://t.co/hEGZA7rgoH
— Dr. Jacqueline Nolis (@skyetetra) February 22, 2021
For best results, fall in the love with the process, not the result
— François Chollet (@fchollet) February 22, 2021
Frequency of letters in English words and where they occur in the word. #dataviz
— Randy Olson (@randal_olson) February 21, 2021
Source: https://t.co/2rPRLDI1Om pic.twitter.com/T1hI4CLqD3
Taming Transformers for High-Resolution Image Synthesis https://t.co/6zdyT0HaR0 impressive work/results! (also fun to see a shoutout and my minGPT code used for the transformer :)) pic.twitter.com/cApDT7Yf67
— Andrej Karpathy (@karpathy) February 21, 2021
8 reasons machine learning projects fail - by @elenasamuylova
— Alexey Grigorev (@Al_Grigor) February 21, 2021
🔸 Doing ML for wrong reasons
🔸 ML not needed
🔸 Bad data
🔸 Poor problem framing
🔸 Model ≠ product
🔸 Bad infrastructure
🔸 No trust from stakeholders
🔸 Production failures
Solution? 👉 https://t.co/mvs7sJyxDe pic.twitter.com/poTAzwWT4b
.@GradioML is a neat library for interacting with a trained model. It's useful for debugging and for giving collaborators the an easy way to interact with the model.
— Anthony Goldbloom (@antgoldbloom) February 21, 2021
Here's a notebook to try it:https://t.co/cwzzdpCTbZ
(Hit "Copy and Edit" and then run the notebook.) pic.twitter.com/FsVKULbLaV