🐢 mini-LLM (github):
A 19M bilingual LLM, which is built and trained using pytorch. It is trained on a small bilingual dataset of hindi and english which is tokenized using a custom tokenizer and follows the gpt decoder achitecture. this is capable of generating stories in Hindi and English.
🐢 Kodo-cli(github):
An advanced AI-powered coding assistant that understands your entire codebase through intelligent context management and AST analysis with the multimodel support.
🐢 Varnika (github):
Varnika, a simple Hindi tokenizer built using the Byte Pair Encoding (BPE) algorithm. It provides functionality for tokenizing Hindi text into subwords and decoding tokenized outputs back into text. Vocab size: 121K
🐢 paper-implemented (github)
A collective repo of papers implemented by me. So far, the list is really small(only 7 papers) but more papers coming soon!! :)
🐢 100M_LLM (huggingface)
A 100M mini gpt model, trained on a small subset of wikipedia dataset. Later instruct fine-tuned for conversations. It was a failure project but learnt a lot.