
Jyotinder Singh
@Jyotinder_Singh
Followers
4K
Following
6K
Media
138
Statuses
1K
Keras and Core ML @Google | Built @thedicedb | I travel and take photos πΈ
Bengaluru, India
Joined December 2015
Spent the Saturday digging into a fascinating paper from Facebook that explains why PEFT methods like LoRA are so effective β¨ It builds the mathematical intuition behind why adjusting only a few hundred parameters can allow you to fine-tune multi-billion parameter LLMs on new
3
10
97
My view is that every language evolves in response to the needs of its users. Over time, it becomes less malleable, and that's when new alternatives begin to emerge. Our needs even after having a unified stack will continue to evolve, and so will the languages that serve them.
The world will eventually converge to one programming language. LLMs will handle all coding in it, fast, precise, and context-aware. We humans will only need to know this one language to read, review, and debug what they create. Everything will be hyper-optimized for this
0
0
31
Been working on this for several weeks and it's finally done - Keras now supports GPTQ!
1
0
35
Didn't know @Samdish was also an engineer
2
0
32
sorry guys didn't know asking people to learn new things and not stay stuck at surface level knowledge was controversial my bad
My last role was basically 2.5 years of writing graph sorting, traversal, and mutation algorithms for a query compiler. Graphs and trees became second nature for me by the time I left. You need to actively reject crud slop in your career if you want to grow as an engineer.
8
1
108
My last role was basically 2.5 years of writing graph sorting, traversal, and mutation algorithms for a query compiler. Graphs and trees became second nature for me by the time I left. You need to actively reject crud slop in your career if you want to grow as an engineer.
1 Year ago I switched from Backend to Compilers. Since then, my day to day consists of designing and implementing Leetcode extra-hard graph algorithms. I escaped the CRUD matrix.
53
213
4K
Love this series. I hope he gets to the one about just having one json column per table. There's even a whole company built around this concept I hear.
PRO TIP: Always keep your database and application on different continents. Database in EU-North, app servers in US-West. Shared regional infrastructure creates a single point of failure.
4
2
33
An overfitting model forgets the beauty of having once captured the training manifold perfectly
1
1
22
The 3rd edition of my book Deep Learning with Python is being printed right now, and will be in bookstores within 2 weeks. You can order it now from Amazon or from Manning. This time, we're also releasing the whole thing as a 100% free website. I don't care if it reduces book
205
771
5K
Big day today thank you so much to @tinytapeout and @matthewvenn ! Submitted our first tapeout of actual silicon romless cordic core with @always_ff_rohan @AksharVastarpar
@Vicharak_In
@Jyotinder_Singh man wish you could have been here, this was a lot of fun! check out the
7
10
58
Pros of living in Bengaluru: no electricity bill Cons of living in Bengaluru: no electricity
1
0
29
My research paper just got published π, and we were able to improve the cache-hit ratio for nginx workloads by 146%. Last week, Aayush (not on X) and I were brainstorming an idea to improve the cache hit ratio, and that's when we applied Deep Q-Networks to see if we could get
59
81
2K
Loving how the beautiful illustrations in the udlbook are!
1
0
17
Guess what π ? I will be speaking at the Grace Hopper Celebration (GHC) for the second time π₯ π ! This year, my colleagues @smn_sdt and @penstrokes75 and I are teaming up to lead a workshop on π€ building agents with Google ADK and KerasHub models π€ . @ghc @AnitaB_org
0
3
10
Y'all are sleeping on Gemini's Guided Learning mode. Game changer.
3
0
33
born too low IQ to build AGI born too high IQ to complain about AI water usage born with the perfect IQ to say "you have two years to escape the permanent underclass" any time a new model is released
88
477
10K