Valentin

@apo1ina Курсы достойные, было что полезное?

малыш бросай работу + выходи за меня я прошла все курсы по машинному обучению в karpov courses и теперь буду тебя содержать пока ты создаешь дома уют и играешь в компьютер

@katyadrinks5 @splashyoo Это из магаза sorry I am not (sorryiamnot.com)

@splashyoo бляяяяяяяяяяяяяя мне надо

приходя в библиотеку, соблюдайте режим тишины

@ukushuukushu Привет! А коуч это для чего? Интересно стало не найти ли и себе

Вот я и стала человеком, у которого психолог, коуч, занятия английским, массаж, баня, тренер по бегу, курс по арт фотографии и еще куча какой-то херни. Стала ли я от этого счастливее? Однозначно да

@ostrisai Well t=0 corresponds to data and when sampling you go from noise (t=T) to data (t=0). So timesteps (for discrete sampler) go in reverse

Noise scheduler timesteps being the inverse of the scheduler timestep indices has stolen countless days of my life away from me. Am I the only one who struggles with this? timesteps[1] = 999 timesteps[999] = 1 I've been fighting a bug due to this for weeks. FINALLY found it!

сделал мем оцените

@MokadyRon This is because ddim (or any deterministic solver) is approximately optimal transport from the noise distribution to the image space. We study exactly this phenomenon (similiarity of images generated from the same noise by different ddpma) in this paper: openreview.net/pdf?id=6PIrhAx…

🔬Exploring Alignment in Diffusion Models - a 🧵 TL;DR: Diffusion models trained on *different datasets* can surprisingly generate similar images when fed with the same noise 🤯 [1/N]

@Daria_CMC Первая тема в курсе по алгоритмам на курсере! Клевая штука. (coursera.org/learn/algorith…)

> How does Union Find work and how to optimize (talk about path compression in find operation) Ай-ой, а меня в школе этому не учили И красно-чёрным деревьям тоже (но про них я хотя бы слышала)

At #ICML2023, Yandex Research will share our latest research on topics from diffusion models for tabular data to model-parallel training on unreliable nodes. We have also co-organized a tutorial on Reinforcement Learning with Human Feedback. Read more at research.yandex.com/blog/yandex-at…

@Nitin_wysiwyg @yoavgo That someone also made lists of wrongthinkers on twitter 😂

@yoavgo You forget the Twitter mob mobilised by a certain someone from a company well known for their GPUs, who didn't realize her Google targeted ads were based on her browsing habits.

just as a reminder: nips took a community wide poll about a name change. results were indecisive. they announced no name change. then magically two weeks later the named was changed.

@marksaroufim Hey! I have tried it for training a diffusion model but have been hitting a lot of dynamo cache limits and broken graphs while training (also using flash attention) - do you know if there is anything torch.compile unfriendly with diffusion models?

If you've been benchmarking PyTorch 2.0 over the weekend would highly recommend you get yourself an 1. A10G which is a cost-effective A100 so you can use tensor cores 2. Add to your code torch.backends.cuda.matmul.allow_tf32 = True otherwise you won't be using your tensor cores

@sabina90408574 Кабзда. А я помню

Вообще забыла этот день.

Вчера был день, который буду помнить, как маркер. Встали на аварийку, чтобы посмотреть запуск.

@peter_richtarik This reviewer is a complete idiot, so sorry that it happened

1. What would you do if an #iclr2023 reviewer ignored the science of your "Markov Chain Monte Carlo" paper, but would give you the lowest score (1) solely because "chains" may refer to slavery, and because linking slavery to the fine people of Monte Carlo is offensive to them?

Introducing Petals, a system for running and fine-tuning 100B+ LMs (e.g. BLOOM by @BigscienceW) over many volunteer devices. No cluster required — just take your own hardware and join others! (1/8) 🌐 petals.ml 📜 arxiv.org/abs/2209.01188 🖥️ github.com/bigscience-wor…

@apsarathchandar @OpenAI @StableDiffusion Classic

An @OpenAI scientist feeling conflicted about the release of @StableDiffusion that could kill their controlled closedAI business. Of course!

@julien_c @huggingface 🌭

Which emoji reaction should we add to the @huggingface Hub? Wrong answers only 👀

Тред персонажи «Гарри Поттера» по описанию из книг, созданные нейросетью Авторка: lettuceown #гаррипоттер #harrypotter Золотое трио:

I like how paperswithcode.com/paper/hyperbol… looks for our paper 🏆 🙂

Oh man, what an effect. Who made this?

Last year we ran the Shifts Challenge at #NeurIPS2021. We had amazing keynotes and a panel discussion at our break-out workshop and it's high time I posted those recordings🤪! Full Playlist: tinyurl.com/4vwyjydt Paper: tinyurl.com/2cs6z925 GitHub: tinyurl.com/yhrn2ns2

@12345thing @boblsturm @natashajaques As a person from that list I am curious who is using it?

Pedro’s comments are so damaging. I’m applying for faculty jobs now and it feels like even if I get a good position, whether I *deserve* to hold that position will always be in question. I will continually have to re-prove that I am as qualified as my male colleagues.