Llama 3, NEW AI Model by Meta just got Released! Is it Better?

this is amazing now we have the meta llama 3 Model the most capable openly available LM to date it will be soon available in AWS data bricks Google Cloud hugging face kagle and much more it is available in two different version one is 8 billion parameter model and 70 billion parameter model we can see the 8 billion parameter model outperforms Gemma 7B instruction fine tuned and myal 7B instruct similarly meta llama 370 billion parameter model is outperforming gerini pro 1.5 and clo 3 Sonet on many of the benchmarks based on the meta llama 3 instruct human evaluation we can see llama 3 is comparatively better than Cloe Sonet mistal medium GPD 3.5 and llama 270b the pre-trained model beats Gemini 1.0 and Mixel 8 22b on all the benchmarks coming to the model architecture a relatively standard decoder only Transformer architecture compared to llama 2 we made several key improvements llama 3 uses tokenizer with vocabulary size of 128,000 tokens this is huge because this means you can include many different languages with this model that's exactly what we're going to see today let's get started hi everyone I'm really excited to show you about llama 3 Model I'm going to break down every single thing which got released in regards to this model but before that I regularly create videos in regards to Artificial Intelligence on my YouTube channel so do subscribe and click the Bell icon to stay tuned make sure you click the like button so this video can be helpful for many others like you in regards to the training data meta heavily invested in pre-training data it is pre-trained on over 15 trillion tokens that we collected from publicly available sources training data is seven times larger than that used for llama 2 5% of llama 3 data set consist of high quality non-english data that covers over 30 languages that is a lot we can scale up the pre-training we can instruction finetune the model llama 3 is codeveloped with torch tune the new py toch native library for easily author fine tuning and experimenting with LMS torch tune is efficient hackable entirely in py torch we can see the safety features such as input Safeguard when we provide the input and output Safeguard when the model generates the output so what's next for llama 3 llama 3 8B and 70b models Mark the beginning of what we plan to release for llama 3 and there's a lot to come they also have a 400 billion parameter model which is getting trained as you can see it is beating many of the models on all the benchmarks you have the model card here these are the special tokens for the beginning of text end of message in a turn start header end header they also released llama 3 repo which I will put all the information in the description below they have provided information on how to download the model how to run using Transformers and much more I'm going to test this model and I'm going to show you how to run this model download this model in my next video so stay tuned I hope you like this video do like share and subscribe and thanks for watching

Channel: Mervin Praison

Views: 2,547

Keywords: llama3, meta llama 3, llama 3, meta ai, meta, meta llama3, llama3 model, llama 3 model, open source llama 3

Id: BHFaG4EMdaI

Length: 3min 37sec (217 seconds)

Published: Thu Apr 18 2024