@akashmjn Profile picture

Akash Mahajan

@akashmjn

MTS @ContextualAI | prev PNW 🏔️& @Azure Speech; @Stanford @atherenergy @iitmadras

Similar User
BUT Speech photo

@ButSpeech

INTERSPEECH 2025 photo

@ISCAInterspeech

Shinji Watanabe photo

@shinjiw_at_cmu

AlphaCephei photo

@alphacep

Saulnier Lucile photo

@LucileSaulnier

Desh Raj photo

@rdesh26

Somshubra Majumdar photo

@HaseoX94

Karthik A Sankararaman 🇮🇳🇺🇸 photo

@karthikabinav

Loren Lugosch photo

@lorenlugosch

Vaishnavh Nagarajan photo

@_vaishnavh

Shreyas Nair photo

@nair_shreyas

Rudolf A. Braun photo

@fasttosmile

Dawei Zhu @ EMNLP2024 photo

@dwzhu128

Digvijay S Mahra photo

@agesBack

Pinned

Like the shift in speech transcription a few years ago, end-to-end optimized systems are the way to go. Here's an update from the team at Contextual AI on what makes truly production-grade RAG systems.

Today, we’re excited to announce RAG 2.0, our end-to-end system for developing production-grade AI. Using RAG 2.0, we’ve created Contextual Language Models (CLMs), which achieve state-of-the-art performance on a variety of industry benchmarks. CLMs outperform strong RAG…

Tweet Image 1


Akash Mahajan Reposted

🚨 Introducing "ColPali: Efficient Document Retrieval with Vision Language Models" ! We use Vision LLMs + late interaction to improve document retrieval (RAG, search engines, etc.), solely using the image representation of document pages ! arxiv.org/abs/2407.01449 🧵(1/N)


Akash Mahajan Reposted

AI4Bharat discord will go live soon! Time to involve the community at scale :)


Akash Mahajan Reposted

RAG 2.0 is turning LLMs from being an awesome toy to a tool that one can safely rely on - so businesses can actually start using AI in their workflows. We at Contextual AI have done an awesome groundbreaking work to make it work. Please see the break down of how and why it works…

Today, we’re excited to announce RAG 2.0, our end-to-end system for developing production-grade AI. Using RAG 2.0, we’ve created Contextual Language Models (CLMs), which achieve state-of-the-art performance on a variety of industry benchmarks. CLMs outperform strong RAG…

Tweet Image 1


Taking a moment to celebrate a life update: 🎉 I joined @ContextualAI and moved to the Bay Area last month. Life has begun an exciting new chapter and I'm looking forward to it! I’m grateful for the opportunity at Microsoft, to work with leading researchers and ship models used…

Tweet Image 1
Tweet Image 2
Tweet Image 3
Tweet Image 4

Akash Mahajan Reposted

Excited to announce that pplx-api is coming out of beta and moving to usage based pricing, along with the first-ever live LLM APIs that are grounded with web search data and have no knowledge cutoff! pplx.ai/online-llms


Akash Mahajan Reposted

A big trap in your 20's and early 30's is "vanity knowledge". Learning things that you think will impress people or get you to the next tier of status in your career. This is rampant in tech/software. A lot of the things "experts on the stage" talk about are fairly irrelevant to…


Akash Mahajan Reposted

The multi task gradient balancing operator we introduced for training EnCodec is picking up steam 🚂⚖️ Think of it as having 1 Adam per loss term, except with no runtime or memory extra cost. No more lambda_1=0.001 and lambda_2=250 🤨🧘

I released today my mini-torch toolkit for multitask learning. github.com/guillaumeBelle… The most useful code-bit is minimal re-implementation of @honualx solution to auto-scale losses with very different scaling. Happy to chat if someone's interested.

Tweet Image 1


Akash Mahajan Reposted

The 100 billion neurons in a human brain are each connected to ~1000 others. It’s a very sparse connection graph, very parallel and efficient. We’ve only scratched the surface on AI architectures. The current deep learning approaches rely on dense tensors and are good for some…

We've barely scratched the surface of the space of deep learning architectures. It's a high dimensional space, so the volume is almost entirely contained in the surface. But we've scratched a tiny subset of the surface.



Akash Mahajan Reposted

voice to music we just launched a feature that allows you to sing and turn your notes into any instrument you want pretty cool to see how AI is giving humans the ability to do things they never could before


Akash Mahajan Reposted

Microsoft CEO Satya Nadella watched the semifinal match before a keynote address. (📸- @devajainn)


Akash Mahajan Reposted

Training code release! Distil your own Whisper model in 3️⃣ steps: 1. Pseudo-label the audio data 2. Shrink the teacher into a student model 3. Train the student on the knowledge distillation objective Training code and examples at: github.com/huggingface/di…

Tweet Image 1

Akash Mahajan Reposted

Just released version 3.1 of #pyannote speaker diarization toolkit 🥇Same accuracy. 🏝️Less dependency hell. ⚡️Probably faster. Try it here and please RT 🙏 huggingface.co/spaces/pyannot…


Akash Mahajan Reposted

Imagine a couple months from now, OpenAI has amassed a massive dataset of these configs because everyone is creating “their GPTs”. The learning on that dataset is going to be wild.

Tweet Image 1

Meanwhile, outside of the tech bubble we can at times forget we inhabit.

BREAKING: US nuclear submarine has arrived at the Middle East, per The Pentagon



Akash Mahajan Reposted

Who’s building the Bharath GPT? 🇮🇳


Akash Mahajan Reposted

Distil-Whisper weights are now available as part of 🤗 Transformers 4.35 Complete with chunking, flash attention 2 and speculative decoding ⚡️ Code examples: github.com/huggingface/di… 6x faster than Whisper on short and long-form audio, within 1% WER performance 🚀

Tweet Image 1

Loading...

Something went wrong.


Something went wrong.