Fun experiment: they tested GPT-3's ability to perform simple arithmetic problems in natural language (without explicitly training it to do arithmetic) pic.twitter.com/KETICaNwxB
— hardmaru (@hardmaru) May 29, 2020
Fun experiment: they tested GPT-3's ability to perform simple arithmetic problems in natural language (without explicitly training it to do arithmetic) pic.twitter.com/KETICaNwxB
— hardmaru (@hardmaru) May 29, 2020
GPT-3 has 175 billion parameters, trained on 300 billion tokenshttps://t.co/rE97CQclwl pic.twitter.com/5tJgwwmABN
— Mark Riedl (@mark_riedl) May 29, 2020
Our new #GPU-based SubWord Tokenizer in @rapidsai CLX feeds BERT models 270x faster than other wordpiece tokenizers. Supports non-truncation of logs and stride, and pipes directly to @PyTorch. Give it a shot! https://t.co/DPe4TpMzYq
— RAPIDS AI (@rapidsai) May 28, 2020
A collection of 481 NLP datasets: https://t.co/xYh6imk0CT
— Denny Britz (@dennybritz) May 28, 2020
This great news! With so many datasets, there must be at least one where your model beats the others and gets SOTA!
I've been working on #huggingtweets, a fun project to generate tweets based on your favorite twitter account using @huggingface. You can fine-tune a neural network and log the predictions automatically into @weights_biases.
— Boris Dayma (@borisdayma) May 27, 2020
The demo runs in 2-3mn: https://t.co/mlfULqtzjR pic.twitter.com/vY48nnKFqp
Check out BLEURT, a novel automatic metric for #NaturalLanguageProcessing that builds upon recent advances in #TransferLearning to capture widespread linguistic phenomena, delivering robust ratings of unprecedented quality. Learn more and grab the code at https://t.co/LxdA6bpxcy
— Google AI (@GoogleAI) May 26, 2020
Interested in visualizing text-based data? Here are a bunch of examples of using tidytext, ggplot, and #rstats to visualize text and do fun basic computational linguistics things like bigram counting, tf-idf, sentiment analysis, and part-of-speech tagging https://t.co/wKpOemTpZ0 pic.twitter.com/B7zmj7ZMqW
— Andrew Heiss (@andrewheiss) May 26, 2020
Thrilled to share new work! “Retrieval-Augmented Generation for Knowledge-Intensive NLP tasks”.
— Patrick Lewis (@PSH_Lewis) May 26, 2020
Big gains on Open-Domain QA, with new State-of-the-Art results on NaturalQuestions, CuratedTrec and WebQuestions.
check out here: https://t.co/SVZ6K4tDn5.
1/N pic.twitter.com/w4CwLxiWxr
ahhhhhh visualizing text in #rstats is so fun (code here: https://t.co/encbbeOZec) pic.twitter.com/BjPUUFjlaL
— Andrew Heiss (@andrewheiss) May 26, 2020
Long-range sequence modeling meets 🤗 transformers! We are happy to officially release Reformer, a transformer that can process sequences as long as 500.000 tokens from @GoogleAI. Thanks a million, Nikita Kitaev and @lukaszkaiser! Try it out here: https://t.co/GwvMrt9lYk pic.twitter.com/q2HymBrSed
— Hugging Face (@huggingface) May 22, 2020
FashionBERT: Text and Image Matching with Adaptive Loss
— roadrunner01 (@ak92501) May 21, 2020
for Cross-modal Retrieval
pdf: https://t.co/jW3Rcmta56
abs: https://t.co/Ck6XtdM0AH pic.twitter.com/AiTKxPD59L
My new AI text generation package, aitextgen, is here! Based on @huggingface Transformers and @PyTorchLightnin, it finetunes GPT-2 based AIs faster and generates faster with even more control over the resulting text! https://t.co/ZlrFdyzAb0
— Max Woolf (@minimaxir) May 18, 2020