I think I have succesfully DPO trained 😇. Time to finetune it some more

👀

🚀We release Kan-LLaMA [ಕನ್-LLama] — A 7B Llama-2 LoRA PreTrained and FineTuned on "Kannada" tokens🚀 We expand Llama-2's existing linguistic capabilities for Low Resource Indic languages and specifically Kannada!

WE HAVE A PLAYGROUND FOR Kan-LLaMA [ಕನ್-LLama]🚀🚀 tensoic.streamlit.app/#kan-llama 🥳 A big shout out to the folks E2E Networks for generously hooking us up with A100 GPUs and @xylem_ai for their amazing superfast inference stack! Do test it out!!

Leveraging Bard (Gemini Pro),we've embarked on the first steps of a promising journey in generating Hindi data for pretraining.While there's more to gather,this marks the beginning of an incredible adventure towards Indic language acceleration. #IndicNLP! 🇮🇳

💥 A small sample of what's possible in the Indic Space - Bhandara📜 A small yet high(textbook) quality pre training dataset for Hindi! huggingface.co/datasets/Tenso…

Generated a curriculum based philosophy q&a dataset to generate essays using bard. Someone found a temperature sweet spot 😅 Have a look for more -> huggingface.co/datasets/adi-k…

New Dataset Alert! FrontEnd CookBook dataset now available on >>> huggingface.co/datasets/Tenso… This dataset includes 45k rows of Q and A conversations prompted for frontend code generation for HTML, React, NextJS, Redux and more!!! Contributor : Adithya Kamath Adithya kamath 🇮🇳

Google's Brand new model - Gemma dropped yesterday! And today we have a full fine-tune of Gemma 2B (7B coming soon!) on Sarvam AI's Samvaad-hi-v1🚀🚀 huggingface.co/Tensoic/Gemma-…

🚨🚀We release "Guftagoo" a Hindi + Hinglish multi-turn conversational dataset comprised of 16k high quality examples⚡️ Guftagoo contains multi-turn conversations on multiple topics usually revolving around daily life experiences, along with COT, and coding. Author: Adithya kamath 🇮🇳

New dataset alert 🙌 P.s. big ups to @wh0ois for direction and looking through examples!!

Hey folks, wrote a blog on some of my experience on building datasets for LLMs. This is not exactly a thorough survey of all the methods, mostly about what worked for me. adikmt.hashnode.dev/create-synthet… Feedback and suggestions always welcome. #generative #LLM

Achieving Tokenizer Flexibility in Language Models through Heuristic Adaptation and Supertoken Learning. arxiv.org/abs/2505.09738

for those of you following pure rl shenanigans, here's the complete plan github.com/tokenbender/av…