Getting Started With New Google Gemini Flash MultiModal With Practical Implementation

Key Takeaways

hello all my name is Krish naak and welcome to my YouTube channel so guys I hope everybody has seen the announcement of the Google iio event uh that had just happened recently of 2024 and there Google had definitely announced many things one is the Astra project specifically for vision related things and you could see that it was much more similar to GPT 4 that was announced by openi along with that uh there was one more variant of gini model that was introduced that is called as Flash and uh this particular model we'll be talking about like this entire video we'll be talking about Gemini flash we'll also show you the implementation with respect to the demo and how you can specifically use it what the specific model is for where you can actually use it itself right uh so let's go ahead and discuss about this so Gemini flash is our lightweight model optimized for Speed and efficiency now as you all know there are three variants already that was introduced early with respect to gemin models one was Ultra gmany Ultra this is specifically your largest model for highly complex tasks then you had Pro right and uh I think I've created a lot of videos in my YouTube channel regarding various endtoend uh generative AI projects rag document Q&A and many more right so this uh is the best model for General performance across a wide range of task and one more model was something called as Nano and this Nano was efficient for on device task uh let's say mobile device U or any Edge devices iot devices as such you can actually specifically use this particular model which will be able to perform all the task very much in an amazing way but now the new model that has come is something called as Germany flash it is a lightweight model optimized for Speed and efficiency so this is one very important thing so here you can see lightweight fast and coste efficient while featuring multimodel reason so it is a multimodel model itself and a breakthrough long context window of up to 1 million tokens now this is definitely very huge with the help of 1 million tokens you can definitely do amazing task as you go ahead right so performance in the floss you can see it is having Built For Speed quality at lower cost long contexture understanding and it is amazing I have actually seen the demo and it looks very much uh you know all the task you'll be able to do it in an amazing way the speed will be amazing optimized apps you'll be able to create right so we will be talking more about it I'll be showing you the demo again with the help of code and each and everything so here you can see longer context now with the help of this 1 million context window by default which means you can process 1 hour of video 11 hours of audio code bases with more than 30,000 lines of code or our somewhere around 700k words right along with this if you probably go over here there are various benchmarks with respect to mlu Natural to code math where it is been compared with various variants of gini right and here you can probably see most of the performance metrics there you can see some minor differences when compared to the other variants but the best thing about this gini flash is that you will be able to use it for two important purposes one is optimization and one is performance right so you'll be able to use it but uh there are some more parameters like audio it is probably able to give 9.8 in video it is able to give 63.5 Which is far much better than all the three models that are available here right uh yes and then you can probably go ahead and try it out in Google AI studio and collab also now first of all what we will do we will go ahead and build with jini so as soon as you build it you will probably go into this now here Ive actually selected gini 1.5 so first of all 1.5 flash we will go ahead with a basic demo okay uh like just like in a playground so here let me go ahead and write provide me a python code for a tic tac toe game okay so I will just go ahead and execute it now let's see how fast it is so here you can see super fast super super super fast um when compared to other models I think uh this is pretty much fast itself and it is able to give you the result I've already tried it out you can go ahead and execute this particular code um there will be some dependency on the libraries that you can probably use it and you can install it okay uh and similarly you can just write provide me a 10K words essay on generative AI let's see whether it will be giving us 10K words or or I don't know let's see okay so I will just go ahead and execute it and quickly oh explanation of that particular code is also given so here you can PR Pro see that how quickly all the content is been getting generated and how it is getting generated and all right so you can definitely find out all these things in this particular Google AI Studio again just go to AI studio. google.com and you can probably check it out um to create the new API and because the next step that I'm going to probably do is that show you how you can use it in code all the links will be given in the description of this particular video now let's say that you have done this much of chat right around 10K words essay is also you are able to generate it uh if you really want to know the code so you can probably get the code not only it supports in Python it supports in JavaScript Android Swift so all this are there and it'll probably give you how to probably use all these things so in Python this is basically the code how you can actually do it this P history and all also you can basically add it and um by if you're using JavaScript automatically you'll be able to add it but just by writing some some rules and condition based on that whatever history you specifically want okay so this is one of the thing now let us go and see one demo over here I'll just show you that how quick how good it is and how fast and efficient it is okay so first of all we will go ahead and install the Gen Google generative AI package okay so once this is probably generated and there are a lot of models that are probably coming up right and it is always good right and since I've created so many ENT projects if you are following my playlist with respect to Google generative a it is very much simple just to use the specific model and understand this is a multimodel that basically means it can also understand text it can also understand uh images videos and all right so here you have Google generative doai so here I'm going to probably execute this also markdown just to display some information um I've already created the API in order to create the API just click on get API and you will be able to create over here so I will just cancel it I've already created it because I don't want to create it again and all all my previous video I've also shown you how to probably create this right so I will just go ahead and execute this and here if you probably go and see in my secret key I have my Google API key okay you can create yours so that you can actually use this and just by using this user data you'll be able to get your API key like just you write user data. Google API key which is present over here on the same name Google API key so I will just go ahead and execute it and then we'll configure this particular API key once it is done now see in the generic uh uh notebook that was given by Google right you had two different models one is gin Pro gin Pro Vision now let's go ahead and list down all the models that are probably present over here so for that you'll just need to write for m in gen ai. listor model gen AI is nothing but what we have actually configured and here you'll be able to see all the St uh generated models so all the all the supported models that it has and right now it has giny 1.0 Pro 1.0 Pro 001 1.0 Pro latest 1.0 Pro Vision latest so this is basically for text this is for vision then what happened they combined both of them and they came up with gini 1.4 Pro latest and then you can also see gin 1.5 latest which is again a multimodel which usually is used for both of functionalities that is nothing but uh your text and images okay so all these things are there now what we are going to do is that we are specifically going to use this particular model I'm just going to execute it and here you can see you just need to paste it over here right so now generative gen. generative model I'm going to call the specific model that is gin 1.5 flash latest now let's see the response time and since I'm using a T4 GPU in the Google collab I'm just going to generate a Content model. generate what is the meaning of life or I can just say write me write an essay of th000 words on generative Ai and all the specific tasks you can actually do it like what we checked in the playground okay so if I execute this and here you can probably see with respect to the response time and all how much time it is going to take because we are generating a thousand words essay okay and we should be able to see the response over here right somewhere around 87.6 milliseconds so let us go ahead and see the text so here is the text the entire text that is probably even generated over here and it is getting displayed right so uh this was the major thing and uh you you can do any kind of task specifically chat and all already I'm planning to create a lot of end to end projects using gin uh 1.5 flash that also I'll be showing you but I also want to give you one important thing with respect to the images Okay so let's say this is my image that I'm probably downloading okay and let's say this image looks like this okay this is the image that we are specifically using okay now what I can do is that I can ask some information regarding this specific image what this image is all about so if you probably see over here I'm just going to write gen. generative model I'm calling this specific model over here and model. generate content image and then we going to display the text from that particular image like this model will be able to understand what this image is all about and it'll be able to give you the answer the image shows two class container filled with a mail of chicken rice broccoli carrots pepper and CM seeds these containers are are on Gray textured background with two chopsticks and all the information are basically there now what I'm going to do I'm just is going to write model. generate content I'm going to give a uh I can also give my prompt so here you can see write a short engaging blog post on this picture it should included a description of the meal in the photo and talk about Journey mail papering and uh here you can see I've given the image and I've kept that stream is equal to true so that I will be able to get the response so if I just execute this now it'll be able to give us the entire uh blog that is probably shown over here but again I'm just executing it in front of you to show that yes this is the entire blog that you can probably see now similarly chat conversation here you have you you just need to write model. start chat and keep on adding in the history okay so this in short uh was an brief overview about this specific model um the best thing is that now instead of directly using gin Pro 1.5 you can actually use gin flash because it is optimized for Speed and efficiency that basically means you'll be able to get the response very much quickly so I hope you like this particular video this was it from my side I'll see you in the next video have a great day thank you wonder take care bye-bye