OpenChat UPDATE: Best Opensource 7B Model EVER! Better Than ChatGPT & Mistral!

[Music] around a month ago we were amazed to see a 7 billion parameter model size outperformed GPT 3.5 as well as llama 70 billion parameter model size this was compared on various different benchmarks and we can see that it's able to outperform on multiple different types of categories this is a model called open chat it is possibly one of the best open- source 7 billion parameter model sizes that are out there it's able to surpass chat gbt as well as different typ of new models that are out there for example like gr now I mentioned a lot more detail in my previous video about open chat so you should 100% watch that video before you watch this because I don't cover the same details as I do in this video now you might be wondering what's the reason you're covering another video on open chat well this is because of a new upgrade that they have recently released and this is something that is badly adopted to open chat 3.5 this is where where the model is now focusing on increasing the performance in one of the most important areas of large language models and that is coding they managed to achieve a near 15o increase on human evaluation while maintaining and improving performance on other benchmarks this is what makes open chat 3.5 one of the most capable generalist models to date now for the people who do not know a 15o increase on human Evel is quite impressive while also maintaining the efficiency as well as improving the performance of the overall model this is a model that is now available throughout hugging face as well as with together. AIS compute optimize inference API these are some of the methods I will be showcasing as to how you can use this new model and we're just going to basically showcase this new upgrade further in detail so I highly recommend that you stay tuned throughout this video because we're going to uncover this new model by comparing it with other sorts of models like GPT as well as with llama so with that thought guys stay tuned and let's get straight to it if you would like to access our private Discord where you can access subscriptions to AI tools for free get networking opportunities collaboration daily AI news you have various different AI topics being discussed on the Discord as well as a lot of support definitely take a look at the patreon link in the description below if you would like to book a one-on-one with me where you can access my Consulting Services where I can help you grow your business or basically give you a lot of different types of solutions with AI definitely take a look at the calendar Link in the description below hey what is up guys welcome back to another YouTube video at the world of AI in today's video we're going to take a look at open chat once again now this is possibly the world's best open- source 7 billion large language model that is out there now this new upgrade of open chat is able to already surpass chat G 3.5 but various other models such as Gro as well as llama's 70 billion parameter model size this upgrade is something that has been widely adopted with this open chat 3.5 model which is centered on enhancing the performance in a critical area of large language models and that is coding now they were able to achieve an impressive 15o increase on human evaluation we can see that it's at 55.2 whereas with its previous release it's at a 44. 8 now this is something that is really really big for a 7 billion parameter model as this Benchmark it with its performance increase in this Benchmark it's able to show that it increased its ability to understand and generate more human-like responses this is a achievement that's coupled with sustained or improved performance on other sorts of benchmarks and this is something that we can see in the comparison with this new release Benchmark sheet in comparison to the old one now obviously for example Mt bench is is going to be comparable like it's still going to stay the same but in other cases you can see an increase or a decrease or not a decrease but mostly an increase on all different benchmarks in comparison to this new upgraded sheet now on my previous video I covered This research paper more in depth and it's basically something where you take smaller advancing open source language models with mixed quality data to improve the efficiency and the performance of it this is how they're able to take such smaller parameter sizes and have it comparable to larger par parameter sizes this is because of the approach as well as the development techniques that they use to basically have it more enhanced and more efficient in generation now I know I mentioned this at the start but the Open chat 3.5 model is one of the most capable generalist models to date we know that it's able to proficiently and effectively surpass various other models on most categories whether that's coding now which is is something that they really really drastically improved we know that it's able to do really good in logic and reasoning mathematics as well as in scientific reasoning we can see that they have done a great job by heightening the model's proficiency and handling all these different tyes of task so that it's able to be a superior language model in understanding and generation is something that is able to even outperform Gro which is a new model from Twitter or X you can say now and this is elon's model which is even under voluming compared to open chat open chat is able to do almost everything better upset MML U and this is something that is comparing grock's one model as well as Gro zero model which is a 33 billion parameter model size which is quite impressive to see smaller primer sizes outperforming larger ones and this is something that we saw with chat gbt 3.5 March model now throughout today's video I'm going to be showcasing as we go further down the video for how you can actually use this with together AI as well as how you can start playing around with it with their actual UI so the first way you can access this new upgraded model is through together. a they have basically made it an announcement that you can try it out with together ai's uh new optimized inference API and you can simply just click on this uh link which will take you to the playground chatbot area and this is where you can play around with different models but you can simply just click on open chat and have it generate various different types of things in this case you can ask it create me a story on a fox for example I don't know and it's able to do it fairly fast and you can start playing around with this chat bot have it tested and you can do various different things throughout like through within together. you're also able to play around with this new model through open chats live demo this is a really nice UI and you can actually play around with it to have it generate various different types of things in my last video I go into more detailed with the test that I did but we're also going to take a look at some of these demos at the end of the video the next method to access open chat locally under desktop is through LM Studio which is an easy way for you to run any sort of Open Source large language model now if you do not have LM Studio which is an application that lets you host the model locally you should definitely watch this video which I made and showcase how you can install it it's super easy and you can possibly run any sort of Open Source model that is out there but if you're going to be running this locally you simply need to go onto the hugging face model card and once you are over here you just need to copy whatever model card that you want to work with in this case this is the new upgraded model so you want to click on this once this like finishes loading up you simply copy the model card go onto hugging face and search for the model card p paste it in and once you have found it you can then download it over here you need to have the GG UF model which is already available and you can simply just click download Once you have downloaded it it will take about 10 depending on your download speed it'll take around like 10 minutes approximately but once you have it downloaded you can just select the model that you want to play around with and you can then select the Open chat upgraded model and you can start chatting around with it locally on your desktop and that's easy as that now let's go test open chat versus chat GPD 3.5 in terms of code generation now we know that it's able to surpass GPT 3.5 on most benchmarks but I want to now assess it on code generation now I'm going to ask it this prompt develop a python function named filter positive numbers that accepts a list of numbers as inputs and returns a new list containing only the positive numbers from the original list if the input list is empty or contains no positive numbers the function should remain or return an empty list ensure that the function handles a variety of scenarios including different numeric types and edge cases additionally consider the efficiency of the solution for a larger list now I'm going to send this prompt to both of these large language models to see how the output is but basically this is a prompt that I'm basically giving to the large language model so that it can generate a python function that's able to filter posi numbers from a given list and this is where the prompt is more detailed which covers various different typ scenarios and edge cases this is a way that I'm trying to assess the large language model on correctness versatility as well as in efficiency of generating code so I'm going to now send it out on both of these large language models and see how it's able to generate this and once this has fully generated I'll be right back now before we go on to comp paare the generation for the code that we requested to generate I just want to talk about the inference speed for open chats demo it's a like bit slower compared to chat gbt obviously but this is because this is not hosted on an optimized INF API whereas if you are to try it with together AI it's able to generate the code as well as the responses fairly fast but from both implementations that are provided for these two different models we're able to see that it provided functional code and actually tested both of them out and they work fully functionally now in terms of the clarity we're able to see that the Open chat response is concise and straightforward whereas the chat GT response is a little bit more detailed and this is by giving it more comprehension and more of a concise report and generation in terms of different types of explanation we're able to see that you get more like I would say it's the same sort of explanation but this one's a little bit more detailed in terms of the response or the generation formula that they use for filtering the different numbers for the positive numbers but in general we're able to see that both of them did a good job in generation and I would say it's a 50/50 in terms of generating code now we took a look at different types of generation in comparison to the GPT 3.5 in my previous video so if you want to take a look at the different prompts we use to test it out I highly recommend that you take a look at that but we can drastically see a huge Improvement in terms of coding uh in the previous videos we were able to see that the coding code generation that was done from open chat wasn't as proficient as other models for example like gbt or llama's 70 billion parameter model this is where they done a great job in improving the efficiency as well as the generation in code and that is basically it for today's video on open chat now this new upgrade is very promising as to the development as well as the performance techniques that they are actually using for upgrading their models I definitely recommend you check this out as this is a really efficient model that anyone can run as its parameter size is quite small but that's basically it for today's video huge props to the open chat team I'll leave all these links in the description below so that you can access it fair easily definitely check out the patreon link if you guys haven't already this is a way for you to access our private Discord if you guys haven't followed world of AI on Twitter I definitely recommend that you do so and lastly make sure you guys subscribe turn off the notification Bell like this video and check out our previous videos so you can stay up to date with the latest AI news but with that thought guys thank you guys so much for watching have an amazing day spread positivity and I'll see you guys fa shortly peace out fellas

Channel: WorldofAI

Views: 4,707

Keywords: artificial intelligence, mistral 7b, machine learning, Mistral uncensored, openchat, open chat, openchat 3.5, better than chatgpt, better than chat gpt 4, gpt 4, gpt 5, ai, gpt 4 demo, gpt 4 free, gpt 4 plugins, chatgpt killer, how to use gpt 4 for free, chat gpt 4, bing gpt 4, gpt 4 bing, new bing, new bing gpt 4, bing chat gpt 4, bing ai gpt 4, chatgpt 4, chatgpt4, gpt4, gpt 4 for free, how to use gpt 4, how to use chat gpt 4, poe chatgpt, gpt 4 how to use, mixtral

Id: Vlr7Xz6bNWQ

Length: 12min 51sec (771 seconds)

Published: Tue Dec 19 2023