@Creative_Math_

Deep Learning Research intern @blocks (Dessa). Did pure math in a past life. cashmere-y, grad student @UofT 🍁

 Toronto, Ontario
block.xyz/dessa
Joined September 2020
  • Tweets995
  • Following405
  • Followers3.2K
209Media
Include

Only show posts containing:

Exclude

Hide posts containing:

Time range
-
Minimum likes
CM retweeted
anpaure
@anpaure6h
Tinker’s Dimuon is Muon with delayed diagonal conditioning, Polar Express and a rank-dependent update norm. I recovered its update rule through the API, borrowing @mrtnm’s approach to StableReinforce. Here’s how 🧵
Tinker
@tinkerapiOct 15
Dimuon is a new optimizer we built for LoRA fine-tuning. In our SFT tests, it often reaches lower loss in fewer steps than AdamW. We're still experimenting with Dimuon; you can Tinker with it today and share your feedback! See our optimizers guide at tinker-docs.thinkingmachines…
13
16
2
227
10,636
Man, some people are so smart. It’s probably possible to reverse engineer the finetuning optimizer variant of Muon that Tinker released a few days ago as well
Martin Marek
@mrtnmOct 25
Tinker's new StableReinforce objective is Reinforce + Score Centering. Here's how I reverse-engineered it 🧵
1
27
2,479
Hard agree, especially with the last point. There is a lot more you can squeeze out from GPUs than people first imagine. Like a LOT more
Squawk Box
@SquawkCNBCOct 23
Startup @flappyairplanes is betting the future of AI is making models more efficient. cnb.cx/3UiruGm
3
410
Replying to @somewheresy @AlexStepherylh
yes we do, check the blog block.xyz/dessa for other samples with other voices

Dessa - the Conversation Model trained only on speech

block.xyz
1
14
Replying to @pfista
Check the linked blog, we even have 5 minute long convos
1
2
220
Replying to @mmaaz_98 @blocks
Ty !!!
2
104
Replying to @ashmrz10
One nvl72, so one rack (72 GPUs)
1
3
134
Replying to @dvirzg
we need to put you in an RL env with a sota optimizer
1
103
Key takeaway from the launch: We ended up sounding more like AI than our model did
1
13
853
Replying to @plotarmordev @blocks
it's because it's not text to speech, it's speech to speech
3
130
Replying to @MTSlive
speech-to-speech* It's not a TTS model, see block.xyz/dessa

Dessa - the Conversation Model trained only on speech

block.xyz
13
1,381
Replying to @tokifyi @KadriJibraan
Most of us are Waterloo + UofT
Ragavan Thurairatnam
@RagavanThurairOct 23
Meet Dessa, our AI model trained from scratch on nothing but speech. We believe speech is the natural interface for technology. So we built Dessa. And we’re inviting AI safety researchers to explore Dessa. Learn more: block.xyz/dessa
7
204
Replying to @yar_vol @RagavanThurair
@yar_vol @RagavanThurair
2
73
Replying to @nftsupply @jack
@nftsupply @jack
1
48
dessa by @jack's company Block incredibly impressive best speech to speech model ive seen bar none
19
15
3
212
34,076
Replying to @harleyfoote_ @jack
Check out the blog, it interrupts us many times in the samples there
1
44
Replying to @raw_avocado
😭😭😭
42
Replying to @nortonbreads @jack
😭😭😭😭😭
3
950
CM retweeted
jack
@jackOct 23
new from us at @blocks
Ragavan Thurairatnam
@RagavanThurairOct 23
Meet Dessa, our AI model trained from scratch on nothing but speech. We believe speech is the natural interface for technology. So we built Dessa. And we’re inviting AI safety researchers to explore Dessa. Learn more: block.xyz/dessa
70
96
11
1,488
183,715
Meet Dessa, our AI model trained from scratch on nothing but speech. We believe speech is the natural interface for technology. So we built Dessa. And we’re inviting AI safety researchers to explore Dessa. Learn more: block.xyz/dessa
130
127
101
1,334
937,203
Load more