"what i can't create, i don't understand" - Richard Feynman
Hello, I'm Shivendra. I like to code and make videos. Check out some of my previous work at Vakya
1- SmallLanguageModel [completed, closed]: Making a LLM from scratch all the way from generating raw training data to tokenizing it, creating a model & then training it.
2- Enigma [completed, closed]: Transformer model trained on raw DNA data to predict the next letter of the DNA.
3- Axon [completed, updating]: Lightweight multi-dimensional array manipulation library powered by GPU.
4- Axgrad [completed, updating]: Lightweight Tensor
manipulation library like PyTorch written in c/c++, cuda & python.
5- Axon.drop [completed, closed]: A small & lightweight, experimental Tensor
manipulation library wrapped on top of Scalar level autograd, written in c/c++, cuda & python.
6- WebGraze [completed, closed]: A Python-based library for scraping data from various sources on the internet. For ML usecases.
7- Synapse [completed, closed]: A free platform for streaming music & audio/podcasts, based on Youtube V3 API.
8- Micrograd.c [completed, closed]: Micrograd by Karpathy written in C & C++.
9- Tqdm.c [completed, closed]: Tqdm library of python for c/c++ code usage.
10- AVA [in progress, backlogged]: A multimodal ai system inspired by AVA from Ex-Machina, but currently more like OpenAi's 4o, (I started working on it prior to the model launch).
11- Shred [completed, updating]: Fast & efficient BPE tokenizer written in C & python for LLM tranining.
12- Enigma2 [in progres, backlogged]: Second version of Enigma to predict & classify DNA & proteins more accurately using Transformers.
13- Biosaic [completed, updating]: Tokenizer for DNA & Protein specific ML applications, for Enigma2 specifically and other applicable programs if any.
Instagram | Twitter | LinkedIn | Youtube