Gergely Papp

QLoRA: 4-bit finetuning of LLMs is here! With it comes Guanaco, a chatbot on a single GPU, achieving 99% ChatGPT performance on the Vicuna benchmark: Paper: arxiv.org/abs/2305.14314 Code+Demo: github.com/artidoro/qlora Samples: colab.research.google.com/drive/1kK6xasH… Colab: colab.research.google.com/drive/17XEqL1J…

Make a ML Engineer cry with just one screenshot. I'll go first

@nucc @ELLISforEurope In some countries, yes. But some countries do allow pursuing a PhD with a BSc. Or at least this is what they say, it looks they don't actually mean it.

@gergopool @ELLISforEurope I’ve thought master is a requirement for doing phd. When did it change?

Is a master degree really a must for a #PhD ? @ELLISforEurope advertised they accepted applicants with BSc only. BSc + 4 years of ML experience + NeurIPS publication was not enough to pass the very first round. Am I really THAT far from being a competition of MSc students?

@mido_assran This paper to semi-supervised learning is a treasure. Not using labels directly for supervised training, but using them to provide a nice uniform distribution over the predictions instead. Very nice work! I wonder how far representation learning will go in the next few years.

#ICCV2021 Hello world! Super excited to announce that our recent paper on semi-supervised visual representation learning has been accepted as an oral presentation at ICCV! arxiv.org/pdf/2104.13963… github.com/facebookresear…

Visualizing and understanding InceptionV1's learning process on Celeb A dataset with lucid. I've used a StyleGan2 and gradient ascend to find out what a neural network thinks who's the most attractive generated human being. Here it is. #DeepLearning

🔥 Latest Deep Learning OCR with Keras and Supervisely in 15 minutes hackernoon.com/latest-deep-le… #machinelearning #ai #deeplearning Is it only me getting confused around shapes? From the code it looks like the image is transposed, switching its width and height. Why is it necessary?

First day @ #MIPS2019 for Asura is a hit! Time and time again it is our live demos of vehicle recognition using make and model recognition and #LPR all integrated into #Milestone XProtect drawing lots of interest and opportunities for new acquaintances. Getting ready for day 2!

Sometimes classification is not as easy as the dog/cat dataset introduces it. Very nice thoughts on classification problems where even the supervising humans are unsure about their decisions. #DeepLearning #classification towardsdatascience.com/what-is-empiri…

Such a multifunctional course from deeplearning.ai!!! If you ever wanted to be the master of deep learning and the rockstar of drawing, @AndrewYNg is your guy! 😉 #DeepLearning #NeuralNetworks

Been experimenting with PPO+Curiosity on a more sparse-rewarding environment. Agent has to learn to press the button to spawn a pyramid, then knock the pyramid over to get the (+1 reward) gold brick. Vanilla PPO completely fails. PPO+Curiosity solves it. 🎉

Releasing Evolved Policy Gradients, an experimental metalearning technique to let agents rapidly learn to solve novel tasks: blog.openai.com/evolved-policy…

Kickstarting Deep Reinforcement Learning proposes a paradigm where 'teacher' agents help train 'student' agents. Benefits include faster research cycles and students that can surpass their teachers: arxiv.org/abs/1803.03835

A Generalised Method for Empirical Game Theoretic Analysis: arxiv.org/pdf/1803.06376…

@awjuliani @mikegeig Could you please tell us the timezone?

Myself and @mikegeig are giving a #GDC18 talk on ML-Agents tomorrow afternoon! Feel free to stop by if you are around. schedule.gdconf.com/session/unity-…

My thoughts about possible improvements on #AlphaZero. #ArtificialIntelligence @DeepMindAI @gergopool/looking-beyond-alphazero-ee501dd3aebc" target="_blank" rel="nofollow noopener">medium.com/@gergopool/loo…

@togelius @awjuliani @DeepMindAI @unity3d @Ubisoft @YOKOZUNAdata @theSpiritAI I wish I could afford :’) #studentlife

Our summer school on AI and Games now partners with @DeepMindAI @unity3d @Ubisoft @YOKOZUNAdata @theSpiritAI and more. We just published the full program, full of exciting lectures. At least, I'm excited to see them! Tickets still available. school.gameaibook.org

goodbye stephen hawking now that the stars can finally meet you i think that they will be so proud that their atoms created someone quite as special and as brilliant as you xox

My very first blog post. Yay ^^ Here I share my personal experience on #AlphaZero. @gergopool/my-experience-on-alphazero-2c2b146567d4" target="_blank" rel="nofollow noopener">medium.com/@gergopool/my-…

Do you want to feel as dumb as an RL agent? Try playing this game: high-level-4.herokuapp.com/experiment where we tried to remove many of the priors that humans typically use. Checkout our paper alongwith @pathak2206 at rach0012.github.io/humanRL_websit…