Aman Goyal

Sharing a bit more of our progress. Try it out now on chat.maruthlabs.com Read more on - maruthlabs.com

@sumanta35679701 Hi, sorry for the late response. We are a research-first company focused on building high-performance Small Language Models (SLMs) like Madhuram, which utilizes our proprietary Perseus architecture. You can read more about our work and us at maruthlabs.com

@goyalaman03 Would love to know more about your work !!

Hiring: Research Intern @ MaruthLabs We are looking for a Research Intern to join us for a 3-month internship focused on pushing the boundaries of high-performance Small Language Models (SLMs). The Role: • Research & Experimentation: You will be given access to 0.5x H100 GPU compute to test and iterate on your own research ideas. • Scaling Up: Upon reaching your research milestones, you will be granted access to an 8x H100 node for a full-scale training run. • Integration: Successful experiments and optimizations will be integrated directly into our core model training pipelines. Requirements: • Strong proficiency in Python and a deep understanding of Transformer architectures. • A research-oriented mindset with an interest in SLMs, efficiency, and context-length expansion. • Degree is not a barrier: We value proof of work, GitHub contributions, and technical curiosity over formal credentials. Details: • Stipend: ₹15,000 per month. • Duration: 3 Months (Extendable). • Location: Remote. How to Apply: Interested candidates should send their CV and a brief outline of a research idea they would like to explore on an H100 to contact@maruthlabs.com. #MaruthLabs #LLM #Research #Hiring #MachineLearning #SLM

@VazeKshitij This generational run needs to be studied.

Been a while, but we back at it again lads🫡

@ChinmayKak @AnshumanAI Hi Chinmay, while spending those 15K on a coding agent subscription seems like a good option, this is more to build a team. Plus an agent can't replace human thinking and novelty "yet". We will be increasing the stipend though once we get the green light to do so. :)

@goyalaman03 @AnshumanAI Genuine question wouldn’t those 15k be better spent then on a coding agent subscription? Might help you guys more. In any case, all the best! Dm if you need help pls:)

@amresh_war Please e-mail us with your idea and CV on contact@maruthlabs.com if you haven't already. :D

@goyalaman03 Hey, currently I am a student and want to pursue research in Automatic Modulation Classification for wireless communication in defence equipments. Iam applying for this role hoping I would get the opportunity to perform my research

@sumanta35679701 We could but we are also looking at this as a way to build a team 😀

@goyalaman03 No offense, but honestly look over in angellist or get some one to work voluntarily in exchange of equity. However You can hire claude as well in that money.

@IamRamratan Will give it a read

@goyalaman03 Try reading The Godfather

Recommend some good books to read to pass time while the GPUs free up 🫠

Planning to do a space in about 30 minutes. See you there.

@silver__tsuki Hi Rahul, I think 0.5x H100 is good enough for running ablation studies for a 150M params model. Besides, it will be available 24x7 for use without any restrictions and we will provide 8xH100 (or even more and better GPUs) for the complete training run. :D

@goyalaman03 > 15000 stipend for ML research intern > 0.5x H100 😭

@AnshumanAI Hi Anshuman, we understand that 15,000 might feel less to you but it is the best we can do currently.

@goyalaman03 15,000 me safai karmi jhadu na maare. "Research"

@Naive_enough Hi Aditya, thank you for your interest. Sadly we don't have any full-time roles currently.

@goyalaman03 Any full-time roles? ~2 yrs exp!

@notanikdey Received. Will get back to you soon. 😄

@goyalaman03 just applied via email, interested in working on SLM reasoning through structured adapters

@eliebakouch Would you suggest having a smaller model (so that it could be served) with a great performance with a very long context like 1M? The model wouldn't obviously be the best at very complex tasks but would be good enough for stuff like legal-tech where long context is a challenge.

separating infra and science for long context doesn't make sense, most long context science is about making computation and memory (capacity and bandwidth) feasible at scale. today's infra wouldn't support MHA on a 1T model at 1M context

@funirudh Me

how many of yall still buy books from offline bookstores im trynna see sumn

Please be for Araujo. Please be for Araujo.

Yes, those cloud bills aren't gonna pay themself.

@FioraStarlight Yes. It's been a common practice for a long time.

Do people ever do, like, reverse SFT? Like, constructing an example of an unwanted behavior, and having the loss signal say "you should put zero probability on every token in this sequence"?

What's up with X showing different sports related posts? All I want to see is new papers. This is frying my brain.