Ashvanth.S
@ashvanth_s1
Followers
254
Following
10K
Media
124
Statuses
2K
Data Scientist @ Linarc | Community Researcher @CohereForAI | Deep Learning Practitioner | Interested in Continual learning and Generative Models.
Joined September 2021
An open-source effort at its fullest: open weights, open data, open code. 🌐 Explore the code: https://t.co/WtpE5c1ZyB 📄 Read the paper: https://t.co/t4sroUNH5B 📂 Access the dataset: https://t.co/vem4hOyhB6 🤖 Try the model:
huggingface.co
Introducing Maya – A New Multimodal Multilingual Vision-Language Model. Maya is completely open source, open weight and open dataset, designed to handle 8 languages, cultural diversity, and nuanced real-world contexts in vision-language models.
1
1
10
The only shortcut that is the solution to the problem will be the long route to take.
0
0
1
Probably my final book of 2025 , the first sci-fi book of Sujatha "Sorga Theevu" Quite reminiscent of 1984 and classic tropes of dystopian and surveillance filled world. Do read and take a peek at one of the early works of sci-fi in Tamil.
0
0
1
In 2020, we did something very odd. Well, K (Kailash, CTO) did. He helped open-source Alar, a Kannada–English dictionary. It's a little absurd, considering we're a stockbroking company, but the project itself is one of monumental importance. The story of how Alar came to be is
85
374
3K
This effort results in a highly nuanced cultural benchmark . Glad we got to contribute to Tamil. This is just the beginning. Seeing @prajdabre contribute with his mom made me realize i could too and yea went for it. While it is not a standalone major contribution, means a lot.
0
0
1
Finally the GlobalPIQA is out , got to contribute to Tamil alongside my parents ! When i started my journey ( it has just began ) dreamt of writing a paper with my parents . We got to be a small part of this huge effort taken by more than 300+ researchers worldwide.
Introducing Global PIQA, a new multilingual benchmark for 100+ languages. This benchmark is the outcome of this year’s MRL shared task, in collaboration with 300+ researchers from 65 countries. This dataset evaluates physical commonsense reasoning in culturally relevant contexts.
3
2
11
Introducing Global PIQA, a new multilingual benchmark for 100+ languages. This benchmark is the outcome of this year’s MRL shared task, in collaboration with 300+ researchers from 65 countries. This dataset evaluates physical commonsense reasoning in culturally relevant contexts.
2
57
112
For people thinking that DeepSeek-OCR is the first model to render text as images, the University of Copenhagen already did this in 2023 Paper is called "Language Modelling with Pixels". They trained a Masked AutoEncoder (MAE) by rendering text as images and masking patches
24
56
538
Interesting paper and even more impressive to note how the increased size of the model (as reported in the ablation study) only leads to poor generalization. Probably trying to solve all tasks by incorporating language (like CoT) might not be the best approach.
New paper 📜: Tiny Recursion Model (TRM) is a recursive reasoning approach with a tiny 7M parameters neural network that obtains 45% on ARC-AGI-1 and 8% on ARC-AGI-2, beating most LLMs. Blog: https://t.co/w5ZDsHDDPE Code: https://t.co/7UgKuD9Yll Paper:
0
0
0
Have you ever wondered about software engineering in a project with 1M LOC, 150K stars, hundreds of contributors, thousands of users? We don't claim to have the answer, but we share what we do in the transformers team. Spoiler: not your usual SWEng principles. ML is different.
1
2
15
To seek or to find , to live to the fullest or to restrain ? Will read this again probably few years down the line and reflect back.
0
0
2
Love @Gradio for existing, helps me so much in inspecting data and build tiny demos . Best part is how you can get to build a simple interface to inspect your data . The minutes of effort put in delivers outsized value in return.
0
0
3
New hobby is to engage with Claude in learning mode on a RPG. Where I'm the student who is trying to learn concepts I'm curious about. As @justinskycak puts it, curiosity gets the wheels greased and not move it. While it might not move it immediately, down the line, it might
0
0
1
Fuck it. Today, we open source FineVision: the finest curation of datasets for VLMs, over 200 sources! > 20% improvement across 10 benchmarks > 17M unique images > 10B answer tokens > New capabilities: GUI navigation, pointing, counting FineVision 10x’s open-source VLMs.
23
112
934
"One of the great beauties of the scientific occupation is the pride of being private in the great army of differentiators" will etch this in my mind forever.
1
0
1
after leading a few projects, i've found that once you've set up the evals + experiment harness and make it easy to tweak config and prompts with 1-click run + eval, teams enjoy running experiments and hill climbing those numbers, and progress comes quickly. but setting up that
26
52
702
Its taking a while to get used to this agentic development workflow but highly promising in terms of the direction its headed towards.
0
0
3
Vertex ai docs feels like a convoluted mess. Half the time is spent in meandering through asking why is this sdk used here , when should i get use which method . Phew
1
0
4
Overheard someone saying "kafka is a light read and you need to probably try chetan bhagat" while waiting at the counter for billing. Thanks guys didnt know this all this while
0
0
1