@deploy_friday

building products & deploying on friday......

 Bengaluru, India
sourabhk.in
Joined November 2021
  • Tweets901
  • Following216
  • Followers92
264Media
forgot, this was the whole point
18
72
5
2,736
150,129
Need to test voice ai with a parrot
20
what kind of captcha is this
1
26
3rd patent 😭😭
2
2
58
when you go to dot.com, you will be redirected to x.ai/bot
29
Flipkart gave birth to @Flipkartminutes Swiggy gave birth to @instamart_it Zomato adopted @letsblinkit Tata adopted @bigbasket_com Zepto raised itself quick commerce state is too much fun to watch
2
43
Introducing Penny.py! It is a simple expense tracker build using the concepts of file handling, oops, functions and error handling. Here is a cool demonstration and working model explanation:
8
6
45
1,026
added 3 stops on my Uber....the 1st stop was to collect GPT-6 Luna, post that Grok 4.7 and Opus 5.5. It was me as tanos collecting Infinity gauntlet from the universe.......
3
50
J.G. Büttner, told his elementary class to add all the whole numbers from 1 to 100 to keep the students busy meanwhile Gauss had different plans.......
1
4
90
spent the last few weeks, working on top of @PrimeIntellect's Prime Flash MoE, trying to get close to the roofline. so here's Prime Super Flash MoE. ~1.2x faster on BF16 & ~1.6x on MXFP8, vs upstream across 4K–128K tokens on B200. do check it out: github.com/Itssshikhar/prime…

GitHub - Itssshikhar/prime-super-flash-moe: Fused MoE kernel

Fused MoE kernel. Contribute to Itssshikhar/prime-super-flash-moe development by creating an account on GitHub.

github.com
5
11
1
63
3,929
huggingface went from hosting inference providers to owning the everyone's cluster. kind of goal you should aim for.
1
1
8
268
i swear, just one more article to add in the to-read list.
@cdleary
@cdlearySep 1
"openai engineer explains /why/ he didn't need to understand the kernel line by line" 🙃 we're doing compilers 2.0

Compilers 2.0: AI as stochastic optimizer

Article · @cdleary
1
1
3
260
a new kind of user is emerging, they’re called agents.
3
61
when i was a kid (still am), i pronounced "Optimus Prime" as "Octopus Prime" at 13 someone told me the mistake i made , today our "Optimus Prime" is no more
5
119
If humming birds were of same scale as humans they would require 155,000 calories a day compared to 2000 calories a day for humans to survive.....
2
74
the most complete Taylor series and the expansion of Sin x
6
110
this is special bcoz this is 1st with @encapsulated007, he was the one who kept my spirits high that your research area has potential to be big, and together we published a simple research article on "Shape Memory Polymer (SMP)" 4 years later StarCloud powered by @PhilipJohnston are sending payloads to space and i do believe optimizing weight of the payload, will surely comes into play..... i am with team SMP-polymers when it comes up to optimizing the payloads compared to SMA (shape memory alloy).....lol
3
126
npm run poland
2
51
Sourabh retweeted
Ayush Garg
@ayushakrtaAug 11
We're fishing for new and hard problems to work on here at @AkrtaIndia! If you're a business owner (or even otherwise), what are some workflows in your business that you would love to see get automated but for which no AI product exists or works satisfactorily for you? We'd love to assess your particular issues and see if we can cook something that could help your business :) Comment or DM us directly! #AI #Automation #Business #B2B #Services
2
2
4
383
Day-6.1 of building LLM from scratch > I built an intuition of LLM’s build vocabulary > Just like humans have a dictionary, LLM has its own dictionary > With the help of mathematics LLM’s are able to understand (quite crazy how maths helps llm learn vocab) next up token id's
2
74
Day-6 of building LLM from scratch > got to know how to break a sentence into words with individual characters > bit it had a problem it had white spaces too > this white spaces and punctuation can cause extra tokens and might limit the processing power > Removing white spaces saves time and memory and computing requirements. > we need double dashes, double dashes as seperate tokens This also built an intuition that llm’s can be optimized so that we do not overload them with repeated information
2
51
Load more