Brent M. Spell

[Brent M. Spell @bmspell]

Finally got around to catching up on 2015-2017 #AdventOfCode. https://t.co/QfhYJdA22T

[Brent M. Spell @bmspell]

@rasbt Hi Sebastian, thank you for this post! In your readme, you listed the RoPE θ as increasing from 10,000 to 50,000. I believe the paper says they increased it to 500,000. Is that correct?

[Brent M. Spell @bmspell]

@TKyaman Hi there, I'm glad you found this code useful. It was such a small snippet that I didn't bother licensing it :). Feel free to modify as you like.

[Brent M. Spell @bmspell]

A vectorized np.interp for Tensorflow: https://t.co/Y8IueaQTdJ

[Brent M. Spell @bmspell]

not sure I'm doing this right...

[Brent M. Spell @bmspell]

@erogol Thanks so much for the shout-out! You guys have been doing some really interesting work at Coqui.

[Brent M. Spell @bmspell]

Kudos to @benfrederickson for py-spy. Just tracked down a nasty problem with a CPU tensor that should have been on GPU in PyTorch using the sampling profiler, which brought my GPU utilization back up to 100%.

[Brent M. Spell @bmspell]

@GoodGood014 Nice work on the library! When I went looking I was surprised something like this didn't already exist, so it's good to see a library for calculating FAD in PyTorch.

[Hao Hao Tan @GoodGood014]

@bmspell saw you working on sth similar on the Torch VGGish repo (thanks for sharing your trick), love to have your feedback too!

[Brent M. Spell @bmspell]

New 5K PB #CamRun

[Brent M. Spell @bmspell]

Trimming audio with Gaussian Mixtures: https://t.co/TwZs5FpRrL

[Brent M. Spell @bmspell]

Circling the Supreme Court and the Lincoln Memorial, the perfect 10K.

[Brent M. Spell @bmspell]

@neonbjb At least for speech, bandwidth extension has been a popular technique lately for getting higher sample rates out of audio models. I have had some success working with the HiFi-GAN+ model https://t.co/lM7rFvQRUa. BTW, Tortoise is an amazing achievement.

[Brent M. Spell @bmspell]

@mjsMLP @jmtomczak I think I enjoyed chapter 5 on hybrid modeling the most, but chapter 8 on neural compression was a great introduction.

[Brent M. Spell @bmspell]

I just finished Deep Generative Modeling by @jmtomczak. Covering topics from VAEs to GANs to EBMs, it has changed the way I think about ML. This book also demonstrates the reasoning behind the pesky sqrt(2) voodoo that shows up in so many neural networks.

[Brent M. Spell @bmspell]

@PrivateNLP I really enjoyed working through the annotated transformer: https://t.co/ojcvMWuYsQ

[Kelly Spell @KellySpell]

I call this “being stranger adjacent”, and places like estate sales and thrift stores are generally where I like to get my fix.

[Brent M. Spell @bmspell]

Featuring experiments logged in @wandb and demos in @huggingface spaces and @GoogleColab.

[Brent M. Spell @bmspell]

I published an open source implementation and pretrained models for the HiFi-GAN+ architecture, from the paper "Bandwidth Extension is All You Need" https://t.co/AlXBCsp35d

[Brent M. Spell @bmspell]

@CEStephens I'll bet you can do it. I have given myself a year to try to do a full marathon in 2023. I'm just hoping the weather is cooler.

[Sasha Rush @srush_nlp]

Do you (really) know PyTorch 🔥?

Try out my Tensor Puzzles 🧩. 16 mini-puzzles for those ready to take off the stackoverflow training wheels.

https://t.co/B5L2bja1DN