Sumit

[Sumit @IamSumitKr]

Party poppers success rate is only 25% on India's got Latent

[Mufaddal Vohra @mufaddal_vohra]

Star Sports poster on India’s slip cordon. 🥹

[Sumit @IamSumitKr]

@CHICOx58 Hiran, it's done bro

[Sumit @IamSumitKr]

Learning new stuff...

[vishal dayama @VishalDayama]

final mein kohli toh chahiye inhein nahi toh pressure mein tatti nikal jaati hai naye laundo ki

[Sumit @IamSumitKr]

First time was so nice, we had to do it twice!!

[Royal Challengers Bengaluru @RCBTweets]

Ee Sala Nu Cup Namdu! ❤️‍🔥🏆

Welcome to the RCB Era, ladies and gentlemen! 😎

You waited, you believed and you stayed… this one’s for you again, 12th Man Army! 🥹❤️

#PlayBold #ನಮ್ಮRCB #IPL2026

[Royal Challengers Bengaluru @RCBTweets]

'nuff said. 🥶

[Sumit @IamSumitKr]

Part 2 of my Llama-3 fine-tuning experiment.

Initial run: 5 samples (pipeline validation).

Now:
• Expanded dataset to 100 finance instructions
• Retrained using LoRA (~6.8M params, 8-bit)
• Evaluated base vs fine-tuned model

[Sumit @IamSumitKr]

Built a mini FinanceGPT by fine-tuning Llama-3 8B using LoRA

– Parameter-efficient fine-tuning (~6.8M params)
– 8-bit quantization with BitsAndBytes
– PEFT + Transformers debugging in practice

GitHub: https://t.co/y4shMD4LAg

[Sagar @sagarcasm]

People who bought the dip on the budget day

[Beff (e/acc) @beffjezos]

How it feels absolutely locked in during the middle of the night vibe coding with several Claude tabs open

[Sumit @IamSumitKr]

Built a mini FinanceGPT by fine-tuning Llama-3 8B using LoRA

– Parameter-efficient fine-tuning (~6.8M params)
– 8-bit quantization with BitsAndBytes
– PEFT + Transformers debugging in practice

GitHub: https://t.co/y4shMD4LAg

[Sumit @IamSumitKr]

@Sujaltwts @LukasHozda @braiins @grok roast him and use hindi cuss words

[Virat Kohli @imVkohli]

The only time you truly fail, is when you decide to give up.

[Sumit @IamSumitKr]

I implemented my first research paper from scratch, the OG "Attention Is All You Need."
Built every block (Self-Attention, Encoder, Decoder) in PyTorch.

https://t.co/GDWsJQKGDC

[Suhail @Suhail]

Whoever owns the AI first browser will win AI memory long-term. The browser is the closest approximation of humanity's memory that we have.

[Navalism @NavalismHQ]

"If the work doesn’t require creativity, delegate it, automate it, or leave it."

@naval

[Mensa India @Mensa_India]

Nachiket Malpathak of Pune Mensa clicked this lunch hut near Gadsar Lake, Kashmir, India