The Birth of Baby Llama

Can't read or see images? View this email in a browser

The excitement is real. The quest for running LLMs on a single computer nudged OpenAI’s Andrej Karpathy, known for his contributions to the field of deep learning, to embark on a weekend passion project to create a simplified version of the Llama 2 model – aka Baby Llama, or llama 2.c. “Yay, Llama 2 can now load and inference the Meta released models! 🙂,” shared an excited Karpathy, citing the smallest 7B model at ~3 tokens/second on 96 OMP threads on a cloud Linux box. “Still just CPU, fp32, expecting ~300 tokens/second tomorrow 🙂,” he shared, on X (formerly Twitter).  Further, he said, if we can get the 7B model to run at nice and interactive rates then you can go from “scratch-trained micromodels” to LoRA finetuned 7B base model’, all within the code of the minimal llama2.c repo (both training and inference). “Can reach more capability and with less training data,” he added.  So, a person who can easily build GPT-5 over the weekend is surprisingly spending time testing out the capabilities of open-source Llama 2. To this, Karpathy said that all of this is quite generic to just transformer language models. “if/when OpenAI was to release models as weights (which I can neither confirm nor deny!) then most of the code here would be very relevant,” he added.  Read: OpenAI Karpathy Creates Baby Llama Instead of GPT-5 OpenAI turns to open source: All of this hints at the company releasing the weights of its GPT models in the coming months. The credit definitely goes to Meta that OpenAI is now (at least) seriously thinking about open-sourcing its models, bringing back the good-old OpenAI, which was actually an open-source, non-profit company.  Turnitin Turns Off Timnit  In an unfortunate event, a student with a GPA of 4.0, and a member of the President’s Honor Roll at the University, was recently declared “failed” by an AI system developed by Turnitin claiming that 67% of his paper was written by AI, which the student denied.  Feeling helpless, the student reached out to former Google ethicist and the director of the Distributed Artificial Intelligence Research Institute (DAIR), Timnit Gebru. Taking this to Twitter (now X), she said, “How many people’s lives are being ruined like this? This should be unacceptable.” She also questioned the workings of Turitin in evaluating AI-generated content and more.  Read the full story here.  The AGI Race Begins  OpenAI’s Sam Altman believes that LLMs could pave the way for building an AGI. He also believes that this entity will not have a body. However, OpenAI is not alone in this race. Other tech labs like Meta, Google DeepMind and Tesla have different perspectives when it comes to achieving AGI.  Here’s a quick glimpse:  But, who will get there first? Read to find out. Reducing LLM Hallucinations LLMs have been haunted by hallucinations, raising concerns about their ability to generate credible information. Despite numerous efforts by top AI think-tanks to mitigate these hallucinations, they remain an inevitable aspect of language models due to their architecture. But, there is still a glimmer of hope. Many experts believe that vector databases might possibly be the key to quelling LLM hallucinations. A new technique called VectorSQL, developed by MyScale, enables users to query vector databases instead of trying to generate the answers to queries by themselves. This is said to reduce hallucinations and make LLMs suitable for widespread use. Read more here. TAUSIF ALAM & AMIT RAJA NAIK Tuesday, Jul 25, 2023 | Was this email forwarded to you? Sign up here DOWNLOAD OUR MOBILE APP Stay Connected info@analyticsindiamag.com © 2023 Analytics India Magazine

This email was sent by support@analyticsindiamag.com to alexvarboffin.abbb@blogger.com Not interested? Unsubscribe | Manage Preference | Update profile Analytics India Magazine | 280, 2nd floor, 5th Main, 15 A cross, Sector 6, HSR layout Bengaluru, Karnataka 560102