Devendra Kumar Sahu

Senior Applied Scientist. Ex - Amazon & Microsoft. Curious, ML Enthusiast. Views are my own.

[Iris Hayes @irisneural]

🚨BREAKING: You can now run Claude Code for FREE.

No API costs. No rate limits. 100% local on your machine.

Here's how to run Claude Code locally (100% free & fully private):

[Devendra Kumar Sahu @_dksahuji]

@jonaslindeloev Discovered this today. I have been searching for this for a long time. Thank you for your contribution. :)

[Devendra Kumar Sahu @_dksahuji]

@karpathy Thank you for writing this. That is what I am trying to achieve at https://t.co/iXlQADwg15 . :D

[Devendra Kumar Sahu @_dksahuji]

@AmitAna1 @socialepfo @mygovindia @PMOIndia @LabourMinistry @mansukhmandviya @ShobhaBJP @MIB_India @PIB_India @narendramodi I have been trying for few years. Now I might have to go to epfo office.

[Devendra Kumar Sahu @_dksahuji]

@rlmcelreath Removing logistic will be better.

[Andrew Ng @AndrewYNg]

New short course on Fine-tuning LLMs! Many developers are moving beyond only prompting, to also fine-tuning LLMs - that is, taking a pre-trained model and training it further on your own data, which can deliver superior results inexpensively. In this course, @realSharonZhou, CEO of Lamini (disclosure: I’m a minor shareholder) shows you how to recognize when fine-tuning can be help, and how to train an open-source LLM on your own data. I hope you enjoy the course! https://t.co/3MDfIvmw6t

[Akshay 🚀 @akshay_pachaar]

Here are all the relevant links:

Paper:
https://t.co/Q4BRV45dsw

Code:
https://t.co/9yzwh45NXD

Blog:
https://t.co/iCxppxaj3w

[Akshay 🚀 @akshay_pachaar]

An open source LLM with 32k context length is here!

Introducing Giraffe🦒- The world's first commercially usable 32K long-context open-source LLM based on Llama-2.

In simple terms, context length is the amount of data you can send to a LLM in one API call.

32k here means, approximately 24,000 words! 🔥

AbacusAI have also released the research paper that provides a comprehensive overview of the innovations in context-length extrapolation.

The release also includes:

- training code
- evaluation scripts
- evaluation datasets

So, why you should care about longer context❓

Context length becomes very important when building a Custom ChatGPT - An AI brain for your organization based on your knowledge base.

Now, you are not constrained by the amount of proprietary data you can send in one API call.

Bigger the context length, better it is.

What makes Giraffe-32k special❓

Open source LLMs such as Llama 1 and 2 have a major limitation!

These LLMs have a very small context length of only 2K. This makes them unusable in any enterprise land.

This makes Giraffe-32k a force to recon in the open source LLM space.

Giraffe-32k can do more complex retrievals with fewer mistakes and is essential for maintaining long conversations, or for asking an LLM to help with coding on a large existing codebase.

It, can easily function like an AI brain for your business - boosting employee productivity, improving decision making and surfacing custom insights from your data

Furthermore, the research paper highlights the two novel approaches developed for context-length exploration.

It includes an extensively comparison and test of these approaches against other well-known methods.

Notably, one of our techniques, known as truncation, exhibits highly promising results.

You can find all the relevant links (paper + code + blog) in the next tweet!

Thanks for reading! :)

[Devendra Kumar Sahu @_dksahuji]

I am launching a 12-month online ML course in the first week of December 2023. And I am doing it full-time. Stay tuned.

[Anna Ivanova @neuranna]

Found a cutest demo of permutation testing
https://t.co/7pHzGKGay7

[ICML Conference @icmlconf]

Announcing the Outstanding Paper Awards at #ICML2023

https://t.co/3cN31sCmfY

Congratulations to all the winners!

[Devendra Kumar Sahu @_dksahuji]

@Akshat_World If someone had the clarity of path and was good enough to execute they would have already done it.

[Wittawat Jitkrittum @wittawatj]

A First Course in Causal Inference
https://t.co/8LhxoQ0ceD

[Devendra Kumar Sahu @_dksahuji]

@sa8ypr @Iamsamirarora Biggest chip making company is not a trillion dollar company.

[Devendra Kumar Sahu @_dksahuji]

@BhosalePratim @Ashish_feels If you earn 50 lacs why won't you spend INR 5k and hire help for taxes.

[Devendra Kumar Sahu @_dksahuji]

@ftlsid 400k USD with 10% return via equity can get you there. Let's say 500k to be safe and have asset allocation. In my country 10%+ is easily achievable because index gives 10+%.

[Devendra Kumar Sahu @_dksahuji]

@ravi_iitm Why only under age 27?

[Prof. Anima Anandkumar @AnimaAnandkumar]

How can we mix active and transfer learning in few-shot learning setting with pre-trained LLMs? We show that you need to label only few samples in-domain and leverage on transfer learning from out-domain along with ingrained knowledge in pretrained models
https://t.co/RPyN4lWXhb

[Raza Habib @RazRazcle]

Is LLM finetuning worth it?

If you know what you're doing, finetuned models can be 30x smaller❗️without losing performance. This can unlock applications that would otherwise be too expensive or slow.

STaR is a great example of doing it right.
(DIY instructions at the end)

[Kevin Patrick Murphy @sirbayes]

I am delighted to announce that the "real" camera-ready version of my new book, "Probabilistic Machine Learning: Advanced Topics", is now available. It will appear in print this summer, but it is already freely available online at https://t.co/XjR3no3D2Q.