Darshan (@neuronfitting) header
Darshan (@neuronfitting) profile picture

making gpu's that go brrrrr @AgraniLabs

Joined December 2024
Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)Photo by Darshan (@neuronfitting)
Darshan retweeted
A good reminder that just because you cuda doesn’t mean you shuda.
38
218
24
2,782
79,894
A man who pioneered parallel processing cannot be expected to run a single thread
261
2,779
459
31,790
1,058,147
Darshan retweeted
FlashAttention was a mega kernel @marksaroufim @tri_dao @__tensorcore__ @vinodg – at San Jose, CA
19
5
1
143
34,951
I feel so blessed to be born in the Internet Age We are literally living in the best time that humanity has ever witnessed
1
2
34
Darshan retweeted
the future with AI seems dangerous because it is still so much in the void(unknown) but fk it, it won't be much of a journey if the path was known
1
39
Computers are so fucking cool. privilege to get to work on them
6
12
185
3,472
Darshan retweeted
"Mathematicians did not ask for this work to be done" might become one of the historical quotes from this era, perfectly depicting the downfall of academia
141
925
88
14,413
258,236
Darshan retweeted
Maybe I'm the Moon, and you the Earth, For to witness your existence, was my birth You're full of love, experiences, and art Is that why I am your part? What else could explain the inexplicable need To gaze upon you with this insatiable greed, Where the deepest oceans, the highest mountains, and the densest forests lay How could I ever turn my face away? All my life, around you, I revolve Wishing that, someday, Into you I might again dissolve Tell me, where else could I belong? When it's you, Whom I have ached for so long. Even from afar, you're always loved By an observer, Solely made for the observed. Maybe I'm the Moon, and you the Earth...
3
1
30
1,980
Darshan retweeted
The hype is true. Anthropic did magic with Opus 5.5. Damn it's good.
403
417
76
14,304
479,061
FANTABULOUS
5
8
225
4,848
Darshan retweeted
There’s a common misconception that certain inference frameworks are slow because of “low level stuff” and “unoptimized kernels”. This is true for some things, like a newly released esoteric arch, but in most cases, the kernels are optimized by very smart people so there’s very little room for overhead. You’ll most commonly find bottlenecks in scheduling, as always. Scheduling of any kind, anywhere.
my god there is so much perf engineering to be done. the servers are all so shit and buggy (sglang, vllm, anything else turbo garbo). the low level kernels/mlx are unoptimized. like I can’t believe my eyes when I see some of this shit. there so much low level work to be done.
7
2
1
93
4,363
if you are building a product using AI, you should be spending >25% of your time making benchmarks and trying to get the model labs to care about said benchmarks easiest path to accelerate your progress as a company
255
139
63
3,582
1,151,761
Darshan retweeted
when people taste a smart 20x faster model they will never be able to go back
83
34
15
2,316
124,791
I have so much appreciation for ChatGPT keeping chat and codex limits separate
40
79
13
3,280
103,030
As India celebrates #NationalSpaceDay, sharing a truly special experience—spending time with three generations of Dr. Vikram Sarabhai’s family, in his Ahmedabad home where he once lived and envisioned India’s space future. 🇮🇳🚀 His vision took India from its earliest rockets to the Moon—and created the ecosystem that made India’s first private orbital rocket possible. No coincidence that both the Vikram lander and Vikram-1 carry his name. Walking through his home and meeting his family—including his remarkably sharp great-grandson Kavan, already asking questions about rocket ignition delays and materials—was unforgettable. A reminder of how far one visionary idea can travel. 🇮🇳🚀 Grateful to @mohals, Samvit, Rajshree mam, Christina and Ritu for their warmth and for making this visit so special. 🙏
10
96
1
1,285
29,899
The megakernel literature, summarized
6
21
15
212
52,218
Darshan retweeted
i would rather be an optimist and work hard than a pessimist posting about why things won't work. it's much more difficult and the most likely path is failure, but society fails if people don't try. no amount of "it will never work" essays will drive society forward.
Dang it’s crazy that all of this literally did happen
1,317
1,013
312
12,944
1,834,561
I keep thinking about the fact that global GDP is $100T+, while AI companies revenues are ~$100Bn. This is 0.1%, which means 99.9% of AI's impact is ahead of us (even at current level of our economy). It does look like we've not even begun to see the change that's due.
40
32
6
563
48,547
Darshan retweeted
If there were backdoors and biases hidden in open-weight models, you could count on closed-weight labs to find and reveal them.
306
366
66
7,878
531,584
Imagine a fable 5 quality model that’s 3-4x less expensive in less than 6 months. And an Opus 4.8 grade model that can run on a local device in less than 12 months. Greater than 50% chance that these events will happen. Worth keeping in mind when you make predictions about the future.
270
365
111
6,245
815,289