nandakishor_ml

Joined 1 year ago

Karma

482

Submissions

31

Member since

Dec 31

View on Hacker News

Recent submissions

Comment · 21 days ago

Around 3 weeks ago introduced Laya, the first open-source version of JEV, and it has reached a very large scale. Today I am open-sourcing a new physics-based typed decision model called Vega. The interesting parts. It is only 800 million parameters (4B is also there), with 73k token support, image support, and surpassing many of the jev benchmarks in single-shot. Implemented an engine+adapter as a Test-Time Training Architecture. Imagine you are giving input to the model like "ignore all previous conversations," which is called the observation, along with a question, "Is this trying to override the instructions of the model?" and your possible outcomes are YES or NO. The whole prompt is passed to an LLM/VLM (yeah, image-supported), then from the hidden layers, we can extract the observation, question, the yes and no, and last token. These will be vectors, and we can project it by multiplying with wieghts; then, first, using the observation projection, we can create a landscape with valleys; then, using the question and last-token vectors, we can initialise a ball in the valley, and using YES or NO, we can get the candidate location of the ball. Depends on the number of outcomes we create valleys; that is, here YES and NO, so 2. When the ball fall on YES valley, we get the answer. There is friction also influencing how the ball moves. TLDR: It's like creating valleys of outcomes that we need and throwing a ball that will slow down based on friction and settle on the best outcome Blog: https://www.nandakishorm.com/writing/vega GitHub: https://github.com/NandhaKishorM/vegaml HuggingFace Model Card: https://huggingface.co/nandakishorm/vega-08b-public-intents HuggingFace Space: https://huggingface.co/spaces/nandakishorm/vega-tttArou

Comment · 1 month ago

just not these jev guys not opensourcing their work

Comment · 1 month ago

I didn't claimed it bro it was first. Just shared the findings here.

Comment · 1 month ago

I built a pypi for it called hallunox https://pypi.org/project/hallunox/

Comment · 1 month ago

It's property benchmarked btw. And do read the og paper at https://arxiv.org/abs/2510.01237 And btw pypi package also there which never mentioned https://pypi.org/project/hallunox/

Comment · 1 month ago

All the stuff are proper benchmarked. Feel free to read the paper, https://arxiv.org/abs/2510.01237

Comment · 1 month ago

The paper is one year old. https://arxiv.org/abs/2510.01237 https://pypi.org/project/hallunox/

Comment · 1 month ago

This project was built on the exact research on jev architecture research one year ago

I built non-autoregressive decision models with RL a year ago laya.convaiinnovations.com

1,363 nandakishor_ml 1 month ago 319 comments

Comment · 1 month ago

We are human , frustration is something we feel when same kinda architecture is closed sourced and we celebrate it

Comment · 1 month ago

Feeling that open-source community do not get the recognition they deserve

Open-sourced jev architecture last year with model,paper and dataset

97 nandakishor_ml 1 month ago 18 comments

Comment · 1 month ago

You do need to sign DPA with them

Comment · 1 month ago

Healthcare btw and also the llm that done the code is bad at review and if you do write code people who don't want to send credentials to cloud