Soham Gupta

engineer @SarvamAI | GSoC'23 @VideoLAN | Full Stack AI & Web3 Developer 🌐

[Soham Gupta @sohamgpt]

Very pleased to announce that I successfully completed my GSoC'23 journey with @videolan

I would like to thank the community for this opportunity and my mentor Alaric Sénat for his awesome guidance throughout the project.

#GSoC #opensource #development #shareinpublic #vlc

[Soham Gupta @sohamgpt]

man first nextjs and now a multichain npm attack, things in sde needs to change, best practices and sanity checks are really non-negotiable for any dev atp (the bar has really risen to ship secure software)

[Mehul Mohan @mehulmpt]

The recent tanstack supply chain attack shows us once again how important it is to:

1. Version lock your packages
2. Enforce the minimum release age parameter on your package manager.

Diving deep into today's video on how this attack was carried out, from pull_request_target to cache poisoning

Video out now!

[Sarvam @SarvamAI]

We are excited to announce that Sarvam is partnering with @PixxelSpace to power the AI backbone of India's first orbital data centre satellite.

This is a first for the country, with India-built AI models running on an India-built satellite and both training and inference happening directly in orbit, without any dependence on foreign cloud or ground infrastructure.

[Soham Gupta @sohamgpt]

maan the BLR heat is killng me, hmu if you're here

[Soham Gupta @sohamgpt]

Most teams are still copy-pasting prompts between tabs and calling it “AI strategy.”

I replaced that with one system that actually understands your repo.

@emergentlabs @mukundjha @ycombinator

https://t.co/qg89LagqLL

[Soham Gupta @sohamgpt]

lore begins ig

hope procrastination doesn't get the best of me 🤞

[Sebastian Raschka @rasbt]

While waiting for DeepSeek V4 we got two very strong open-weight LLMs from India yesterday.

There are two size flavors, Sarvam 30B and Sarvam 105B model (both reasoning models).

Interestingly, the smaller 30B model uses “classic” Grouped Query Attention (GQA), whereas the larger 105B variant switched to DeepSeek-style Multi-Head Latent Attention (MLA).

As I wrote about in my analyses before, both are popular attention variants to reduce KV cache size (the longer the context, the more you save compared to regular attention).

MLA is more complicated to implement, but it can give you better modeling performance if we go by the ablation studies in the 2024 DeepSeek V2 paper (as far as I know, this is still the most recent apples-to-apples comparison).

Speaking of modeling performance, the 105B model is on par with LLMs of similar size: gpt-oss 120B and Qwen3-Next (80B). Sarvam is better on some tasks and worse on others, but roughly the same on average.

It’s not the strongest coder in SWE-Bench Verified terms, but it is surprisingly good at agentic reasoning and task completion (Tau2). It’s even better than Deepseek R1 0528.

Considering the smaller Sarvam 30B, the perhaps most comparable model to the 30B model is Nemotron 3 Nano 30B, which is slightly ahead in coding per SWE-Bench Verified and agentic reasoning (Tau2) but slightly worse in some other aspects (Live Code Bench v6, BrowseComp).

Unfortunately, Qwen3-30B-A3B is missing in the benchmarks, which is, as far as I know, is the most popular model of that size class. Interestingly, though, the Sarvam team compared their 30B model to Qwen3-30B-A3B on a computational performance analysis, where they found that Sarvam gets 20-40% more tokens/sec throughput compared to Qwen3 due to code and kernel optimizations.

Anyways, one thing that is not captured by the benchmarks above is Sarvam’s good performance on Indian languages. According to a judge model, the Sarvam team found that their model is preferred 90% of the time compared to others when it comes to Indian texts. (Since they built and trained the tokenizer from scratch as well, Sarvam also comes with a 4 times higher token efficiency on Indian languages.

[Rahul @selfawareatom]

30: https://t.co/db9fJtxOy4
105: https://t.co/xc8oCfnNtk
blog: https://t.co/WfuwrGL38S

[Fernando @Franc0Fernand0]

If you're a software engineer who wants to upskill in system design, read these 14 articles (links below):

[Soham Gupta @sohamgpt]

@itsOmSarraf_ @JainSwajan It's intresting that I found this thread

[Soham Gupta @sohamgpt]

@KimNoel399 How?

[neural nets. @cneuralnetwork]

best thing about time is, it changes

[Soham Gupta @sohamgpt]

very helpful reminder

[Om Sarraf ( ॐ ) @itsOmSarraf_]

Mom Dad finally know where I work thank you modiji 😎🫡

[Soham Gupta @sohamgpt]

@itsOmSarraf_ People are completely missing the context here 😭

[Soham Gupta @sohamgpt]

Join us

[Sarvam @SarvamAI]

The stage is set for India's AI.

Join us for the big drop at Bharat Mandapam, Plenary Hall B in 10 minutes.

[Soham Gupta @sohamgpt]

Warikoo's voice dubbed in english 👀

try out @ https://t.co/Q8v5txatPL

join the waitlist to be among the first to get access to higher limits! 🎟️📈

[Pratyush Kumar @pratykumar]

Drop 9/14: Today we are introducing Sarvam Studio, our product to help creators go multilingual. One piece of content, every corner of India.

With AI video dubbing, Studio generates high-fidelity dubs in 11 Indian languages. In an expert study, participants preferred Sarvam Studio for overall quality and production readiness.

With agentic document translation, Studio excels in contextually translating long-form content across genres. Our evaluations demonstrate that readers strongly preferred the output from Studio across different genres.

[Pratyush Kumar @pratykumar]

Drop 9/14: Today we are introducing Sarvam Studio, our product to help creators go multilingual. One piece of content, every corner of India.

With AI video dubbing, Studio generates high-fidelity dubs in 11 Indian languages. In an expert study, participants preferred Sarvam Studio for overall quality and production readiness.

With agentic document translation, Studio excels in contextually translating long-form content across genres. Our evaluations demonstrate that readers strongly preferred the output from Studio across different genres.

[Soham Gupta @sohamgpt]

🏃🚀📈

[Sarvam for Developers @SarvamForDevs]

Mics on.
Keyboards off.
Something new lands tomorrow.

[Soham Gupta @sohamgpt]

Proud to be part of @SarvamAI and to see it organically connect Indian diaspora tech Twitter across the globe!

[Sarvam @SarvamAI]

Access Bulbul V3 at no cost for the rest of the month.

The mic is yours! Start building 🚀

https://t.co/B6UYd0NEWU

[Soham Gupta @sohamgpt]

Day 2 at @SarvamAI