The cat and mouse games with large language models are going to be fascinating to watch. A recent example (of many) https://t.co/u2lhjwuvLZ if offense is sufficiently advantaged/strong (which I think is likely) then maybe we can't have nice things
— Andrej Karpathy (@karpathy) November 5, 2020
In the last transformers release, we teamed up with @raydistributed @anyscalecompute to provide a simple yet powerful integration for hyperparameter tuning.
— Hugging Face (@huggingface) November 2, 2020
To demonstrate it, @richliaw shows you how to fine tune BERT on MRPC leveraging multiple GPUs. ⤵️https://t.co/TFaVKcRlme
New ML benchmark/competition for receipt OCR: "The dataset contains ~20,000 labeled receipts from political television ads bought in the 2012, 2014, and 2020 US elections" sample code & dataset here: https://t.co/zdBIFlizwO https://t.co/T9xziblSq1
— Pete Skomoroch (@peteskomoroch) October 29, 2020
Our work CheXbert, for automated radiology report labeling, is now accepted to #EMNLP 🥳
— Pranav Rajpurkar (@pranavrajpurkar) October 28, 2020
We just released the pre-trained model+code, with an option for companies to license:
Code: https://t.co/ISj5Oem8wa
License: https://t.co/zdTfYEgiDV@AkshaySmit @saahil9jain @stanfordnlp
DATASET RELEASE: "CC100", the CommonCrawl dataset of 2.5TB of clean unsupervised text from 100 languages (used to train XLM-R) is now publicly available.
— Alexis Conneau (@alex_conneau) October 28, 2020
You can find below the
Data: https://t.co/KDnynkH6hX
Script: https://t.co/BY906YXEHg
By @VishravC et al. pic.twitter.com/KrjcdeRG6P
My @GlobAICommunity talk on @spacy_io v3.0 is now live! In it, I'm walking through some of the most exciting new features & the ideas behind them.
— Ines Montani 〰️ (@_inesmontani) October 26, 2020
📺 Video: https://t.co/dRQbyZKlnq
🌙 spaCy v3.0 nightly: https://t.co/cpUa14hZTr
💥 Blog post: https://t.co/OfC6AYlJPU pic.twitter.com/ioYD1hfjvT
Check out Performer, a generalized attention framework based on the Transformer architecture, which implements the novel FAVOR+ algorithm to provide linearly scalable, low-variance and unbiased estimation of attention mechanisms. Read more at https://t.co/lLzhXJRIbh pic.twitter.com/k153ltoqMD
— Google AI (@GoogleAI) October 23, 2020
Our API now includes a brand new pipeline: zero-shot text classification
— Hugging Face (@huggingface) October 23, 2020
This feature lets you classify sequences into the specified class names out-of-the-box w/o any additional training in a few lines of code! 🚀
Try it out (and share screenshots 📷): https://t.co/C7na3deHKd pic.twitter.com/kFPolnnL85
We are releasing mT5: A massively-multilingual version of T5 that supports over 💯 languages! mT5 was pre-trained on a multilingual version of C4 and achieves SoTA on many cross-lingual NLP tasks.
— Adam Roberts (@ada_rob) October 23, 2020
📜Pre-print: https://t.co/lfQd9CeF7b
💾Code/models: https://t.co/1MauV6etQY pic.twitter.com/LUJIFaMxF5
We just published a new video by @ClementDelangue who walks you through fine-tuning of a language model with #huggingtweets by @borisdayma & how to use the model hub, zero-shot classification & open-domain question answering! https://t.co/qPUdW94BZv
— Hugging Face (@huggingface) October 22, 2020
🚨New model in town!!🚨
— Hugging Face (@huggingface) October 20, 2020
ProphetNet is a new pretrained seq2seq model. We just added the model in 🤗 Transformers! Both English and multilingual pretrained models are available from 🤗 Model Hub!
👉 Doc: https://t.co/k5UKScrgO9@MSFTResearch @QiWeizhen @gyynlpanda @ruofeibruce pic.twitter.com/06UEJdBhwQ
Training a transformer model for text classification has never been easier. Pick a model checkpoint from the 🤗Transformers library, a dataset from the dataset library and fine-tune your model on the task with the built-in Trainer!https://t.co/ahz6sHc0i4
— Sylvain Gugger (@GuggerSylvain) October 19, 2020
