[Feb 10] - Interesting things I learnt this week
This is second of my experiment series to compile a few interesting things which I read about during the week. I hope I can continue this for this year at least.
1. Eagle 7B - The race for better models is on. Gemini from Google is there but its another closed source one. But this one is open source. It claims to have transformers with 1 Trillion Tokens Across 100+ Languages.
My Take: Models are going to be get better in the open source world with each passing day. Very soon we will have GPT4 level models running on our local laptops and phones. The open source is going to be a big part of it. How people build on top of these models is going to become even more interesting. I have not tried this model yet, but I am guessing it will be faster than llama2
2. Text Generation WebUI - This is an amazing tool which starts a webservice on your machines and makes loading and configuring various LLMs a breeze. It will autodownload models once you have given hugginface project name.
My Take: I dont know what more can ask for. Probably one of the best tools out there to try all the customised model and prompt deployment that I have come across. Downloads and loads the models with ease. Not for average usage but more for prompt engineers and those looking to experiment with LLMs
3. Canon's new chip making stamp machine - Canon announced a new chip making machine that stamps chip designs onto silicon wafers rather than etching them using light. The process, says Canon, will be “one digit” cheaper and use up to 90 percent less power than Netherlands-based ASML’s market-dominating and light-based extreme ultraviolet (EUV) technology.
My Take : It's made as a big deal announcement as if something revolutionary has been created but as noted in the article the tech has been around for sometime. But I am still hopeful they will shake things up here with atleast the less complex chips moving to this process.
4. Gemini from Google - Bard is now Gemini and it has an Android app which replaces Assistant.
My Take : Having teeny tiny bit of contribution to this, I don't think it's right of me to talk much about it right now. But it does a whole lot of things and is moving in sort of right direction with miles and miles to go.