@drisspg

bytes and nuggets @pytorch https://xitter.cf/t.co/gWVJmW6wbH

github.com/drisspg
Joined December 2023
  • Tweets690
  • Following302
  • Followers1.7K
99Media
Whatever happened with the midjourney scanner??
6
419
I said what I said
1
7
429
Gave a talk at PTC on flex_gemm which is a new gemm_epilogue API in pytorch built on Quack. I think its neat, not ground breaking but neat nonetheless :) github.com/drisspg/transform… planning to bootstrap this and continue to add fun examples
7
10
3
70
7,188
Gunna be up in the bay for PTC week.. where da parties at?
2
7
1,489
I find it kinda funny how all these kernel verifiers are goaling on bitwise to PyTorch Eager. This pre-supposes that PT eager is the gold standard of numerics. While in some sense this is true, but in many instances it isn't
8
98
5,889
more replies
So instead of wasting your precious gpus on exploring this space you can stand on the shoulder of giants. The cul de sacs tend to be found and marked with big red Xs. BUTTTTT there are in fact better ways for a lot of things, so take a beat do some math, and train some models
2
1
11
570
And then open issues and help pave the way :)
4
329
I love it when opus does cute lil tricks that surprise me
1
6
535
Need to sum up some numbers just leave em in tmem and mma with 1s
1
5
257
I have work to do but I have been nerd sniped so hard by locality domains
2
9
981
If only I could make a blog post on this
1
118
Github Notifications are broken. Whenever I want to build something custom - I thrash my own API rate limits. Like please let me pay 5 bucks a month to get higher rates to help let me manage my notifications for OSS projects @github
3
10
1,483
As someone who tried muse earlier on it is remarkable how much better it has gotten and actual value I'm getting out of it
5
453
I was lucky enough to have gotten the OSS pro plan from OAI and sadge it got yanked right before dev day with no notice :(. It was truly great though I built a lot of features for PyTorch using those tokens
16
906
driss guessous retweeted
Thien Tran
@gaunernstSep 27
Have been busy with work, but finally I find the time to finish the blogpost on @GPU_MODE @coreauto's QR competition that happened 3 months ago. I didn't win the 1st prize, but my panel QR kernel is definitely the fastest (by a large margin) gau-nernst.github.io/b200-qr…

World's fastest (panel) QR factorization on B200

GPU MODE recently hosted a series of leaderboards for linear algebra kernels. I participated in the first one, QR decomposition, and claimed the 2nd place. I think it’s a breath of fresh air, a...

gau-nernst.github.io
2
31
4
207
10,144
Somedays I miss the vscode debugger..
8
454
drisspg.github.io/nuggets/KD… wrote up a new one; lil bit product pitch, lil bit click bait

KDA Doesn't Care About the Future

Where Attention Gym is heading, a future-dependent rounding effect in the chunkwise KDA kernel, and TorchTitan runs asking whether models exploit it.

drisspg.github.io
2
18
2
134
17,597
Corpo appreciation post; I spend obscene amounts of money on tokens. And all of those tokens go towards helping make OSS software better/faster/stronger. I think that is neat
1
6
390
Silo s3e9 is weird as shit
1
6
491
Damn s3e10 coming in hotttt
1
176
Load more