Super nice Transformer breakdown from hacker lax . I'll be reading this with my cup of tea later 🫖

🔥Preprint out: `Toward a realistic model of speech processing in the brain with self-supervised learning’: arxiv.org/abs/2206.01685 by J. Millet*, Charlotte Caucheteux @ICML24* and our wonderful team: The 3 main results summarized below 👇

📢 If you are interested in how to scale 🚀 multilabel classsification to thousands or millions of labels 🏷 I am giving a talk on Sunday in PyData London titled “Extreme multilabel classification in the biomedical NLP domain” 🗣 Also super excited to be back in London 🇬🇧

Nick Sorros shares his approach to the challenges of extreme multilabel classifier prediction spaces with custom #neuralnetwork #architecture & inspiration from #XLinear & #BertMesh PyData London #biomedical #NLP #bigdata #DataScience #pydatalondon2022 Wellcome @mantisnlp

Great talk by Nick Sorros on Multilabel Classification at scale, for biomedical publications. The trained model BertMESH is available on Hugging Face PyData London #pydatalondon Wellcome

Great talk by Nick Sorros on Extreme Multilabel Classification: classification problems with thousands or millions of classes. - svm.sparsify() is useful - TFIDF vocabs can be made smaller with BPE/WordPiece - "Multilabel attention" is used in these models #PyDataLDN

the number of times I have thought "surely there's no way this is made in excel" only to discover that yes, in fact, it is made in excel

Made a playlist for your weekend hike

How did I miss this? rostrum.blog/2022/06/07/ass… Matt Dray is the arch troll of the #RStats world. I feel like there should be an award for this.

I've got a ~surprise~ video for you today! I had a great conversation with Nick Sorros and Matt Upson from @mantisnlp about reproducibility, open source, data ethics & more. I hope you enjoy it as much as I did. 😁▶️ youtube.com/watch?v=uztK9g…

After more than five years of work: our new publication on how to almost double woodland carbon overnight; or, when models work a bit too well. Full paper: bit.ly/3uXiG8D [1/10]

✨Here is quick example of a retrieval based #ChatGPT equivalent you can easily build using your own data and the excellent library from deepset, makers of Haystack

The Adam optimizer is at the heart of modern AI. Researchers have been trying to dethrone Adam for years. How about we ask a machine to do a better job? Google AI uses evolution to discover a simpler & efficient algorithm with remarkable features. It’s just 8 lines of code: 🧵

We recently switched to using Hugging Face inference endpoints as it makes it much easier to deploy and manage the models for our clients. We also created a nice CLI tool called hugie 🐻 to make inference endpoints available from the command line 🚀

My #ChatGPT fears are over.

Here is a nice CLI tool we created on top of the Hugging Face inference endpoints which allow you to deploy any transformer model from the command line or a CI/CD pipeline. Kudos mainly to Matt Upson 👏

Vicious Self-Degradation > you Google > Quora spots query and id’s as frequent > Quora uses ChatGPT to generate answer > ChatGPT hallucinates > Google picks up Quora answer as highest probability correct answer > ChatGPT hallucination is now canonical Google answer

🔎 What if only a base model were needed for preference alignment? Welcome to ORPO, introduced in our latest blog with @MantisNLP. argilla.io/blog/mantisnlp…

🎉 Sieves v0.11.0 is out! Now featuring model distillation for classification tasks using SetFit/Model2Vec! Plus smart document caching to reduce compute waste.🔗 sieves.ai

> Self-styled prophets are claiming they have "awakened" chatbots and accessed the secrets of the universe through ChatGPT Sobering reading. Who saw this coming?rollingstone.com/culture/cultur…