Introducing LQ-LoRA Decomposing pretrained matrices into (fixed) quantized + (trainable) low-rank components enables more aggressive quantization. We can quantize LLaMA-2 70B to 2.5 bits with minimal degradation in instruction-tuning performance. arxiv.org/abs/2311.12023 🧵1/n

Been a while since I've posted, but this calls for one. Excited to be going back to my alma mater (IIT Madras) to share what we've been upto at @yurtsai . Thank you for hosting me Prof Karthik Raman! If y'all are in the area, i.e. around Chennai/IITM, please come by!

"The Genius of Srinivasa Ramanujan" youtube.com/watch?v=SYBOOj… is a documentary film made by a mathematician named Raghuram and his colleagues at IISER Pune. Later today I have the pleasure of meeting Raghuram. Hope you enjoy his film!

Writing provides insane clarity. Don't offload your cognitive skills. Why do you think most therapists and life coaches recommend writing as a means for clarity and action?

Curious to try this out! -- this is a hard problem especially when the type/template of structured information isn't fixed. And when using llms for the extraction - the big challenge of hallucination always looms unless we use encoder llms or older neural network solutions!

Everyone needs to hear this…

will frame this!

life gets so much more exciting and fulfilling when we begin acknowledging the countless "miracles" we're surrounded by. hint: Think about a single breath - and the "symphony" that is orchestrated for a SINGLE breath!

I was re-watching Bhaag Milka Bhaag (bollywood film that popularized the story of the Flying Sikh - Milkha Singh) and only noticed now that the ending had a gem - It's in hindi, signed by Milkha singh, but there's an english subtitle too.