Azure OpenAI in LangChain: Getting Started

00:00:00.000 | Today we're going to take a look at how we can use Lang chain with OpenAI in Azure.

00:00:06.680 | Now for those of you that don't know OpenAI can be accessed in one of two ways either

00:00:12.680 | through the OpenAI API directly or via the Azure version of OpenAI's API.

00:00:20.280 | Now I'm going to show you how to do all of this within Lang chain so we can have Lang

00:00:24.520 | chain, OpenAI and Azure.

00:00:27.140 | So the first thing that we're going to do is head on over to portal.azure.com.

00:00:33.880 | You will naturally need an Azure account if you don't already but once you have set that

00:00:39.680 | up all you're going to do is head on over to your subscriptions, make sure you have

00:00:43.920 | a subscription, you may need to create one so you add a subscription there.

00:00:48.200 | Mine is called development and testing so okay I know that is there.

00:00:53.840 | Then I'm going to head over to Azure OpenAI.

00:00:56.980 | If you can't find any of this stuff you can actually just go up here and type it in.

00:01:06.500 | You may also need to create that resource but once you have it, Azure OpenAI, you come

00:01:10.380 | over here and we already have a deployment down here but what I'm going to do is create

00:01:16.540 | a new one so we can go from start to finish.

00:01:20.060 | So we're going to create a new deployment.

00:01:24.940 | I'm going to use that subscription that I showed you earlier.

00:01:27.880 | I'm going to use a resource group so you may also need to create a resource group.

00:01:33.200 | You can click create new if you don't have one already and I'm going to go to instant

00:01:39.500 | details.

00:01:40.500 | The name will be call it YouTube demo.

00:01:45.740 | Pricing tier, a standard S0 and we can go on to next.

00:01:52.620 | Now here you just set up your security, network security items.

00:01:58.100 | For me I'm just going to go with all networks.

00:02:01.500 | Tags so this is naturally if you have many people and categories on your Azure instance

00:02:08.700 | I'm not going to do that.

00:02:12.540 | Okay and once you've done that you have the terms you just go down to create.

00:02:17.340 | You will have to wait a moment while that is deploying and once it has deployed you'll

00:02:23.460 | look at this.

00:02:24.460 | Your deployment is complete.

00:02:26.180 | From here we want to go to I'm not sure the best way but I'm going to go to home.

00:02:30.860 | I'm going to go back to Azure OpenAI and you'll be able to see your new resource or deployment

00:02:36.140 | in there.

00:02:37.140 | So I'm going to click on that deployment and there's now a few things that we need to grab

00:02:42.780 | from here.

00:02:43.780 | Okay so we have the endpoint.

00:02:45.620 | So let's copy this.

00:02:47.260 | I'm going to go to this notebook so you'll be able to find your own version of this notebook

00:02:52.160 | linked near the top of the video right now or if not in the video description.

00:02:57.700 | So I'm going to come down to here and you see that we have this OpenAI API base right

00:03:04.460 | that is that's the endpoint.

00:03:06.900 | So I'm going to replace that with what I just pasted from the endpoint.

00:03:12.020 | So YouTube demo openaiazure.com I also need the OpenAI API version.

00:03:18.580 | Right now the one of the most up-to-date versions or the version I'm using at least is this

00:03:25.980 | so there might be more recent versions but that's it works right now so I'm using that.

00:03:32.980 | We want to set the OpenAI API type which is Azure and we also want to set the OpenAI API

00:03:40.060 | key.

00:03:41.060 | Right so where do we get that we want to come over to keys and endpoint and we just copy

00:03:48.820 | this okay and we just paste it into here.

00:03:54.640 | So once we have our items there what we now need to do is we want to go to Azure OpenAI

00:04:01.300 | Studio.

00:04:02.300 | Open that and this is one of the main differences between the default version of OpenAI and

00:04:09.220 | the Azure version of OpenAI with Azure or actually with the default version we would

00:04:15.460 | just call the model endpoints we would say okay we want to use GPT-4 and we do that.

00:04:21.660 | You can't do that with the Azure version you actually need to create a deployment for every

00:04:27.320 | model that you'd like to use.

00:04:29.620 | So you'd come over to create new deployment and you see I currently have no deployments

00:04:35.740 | okay so I'm going to create a new one I'm going to choose a model so we have a few here

00:04:43.300 | let's go with a GPT-4 model version we can let's use the default and deployment name

00:04:52.140 | so this is the deployment name that you will use within line chain to connect to this deployment

00:05:00.460 | right so it needs to be unique I don't currently have any other deployments so I can call it

00:05:05.420 | whatever I want essentially so I'm going to call it chat endpoint okay I'm going to create

00:05:12.300 | that okay so I now have chat endpoint so I'm going to come over to the code again I'm just

00:05:19.300 | going to run the pip install up at the top here and then I'm going to come down to here

00:05:25.480 | so I have deployment name I need to change this because I called it chat endpoint so

00:05:33.140 | we have the chat endpoint as our deployment name we have our model name which aligns to

00:05:39.000 | our deployment over here now if we try to run this straight away we'll get an error

00:05:48.420 | it won't work so after waiting a little while it will work as we can see here so for this

00:05:55.580 | it took sort of three to four minutes for that to actually deploy now what we've done

00:06:01.660 | here is we're using a chat model so when we're using chat model we use line chains system

00:06:08.500 | messages human message and AI message to create that sort of chat log so we initialize it

00:06:14.360 | with the system message then we basically system message just tells the model how to

00:06:20.180 | behave okay so I'm saying that it's basically it's an AGI but it cannot answer any questions

00:06:26.700 | about cheese because it just doesn't understand cheese as a concept okay so we can ask it

00:06:32.820 | like hey how are you doing today what's the meaning of life and it's going to respond

00:06:37.660 | you know it's really typical like as an AI I don't have emotions or personal experiences

00:06:42.220 | so on and so on but I'm here to help and then it gives you kind of generic meaning of life

00:06:48.180 | answer okay cool so after that we want to take this message the AI message and append

00:06:56.860 | it to our messages okay and then we can feed that back into the model so I'm gonna take

00:07:03.620 | all those interactions so far and I'm gonna ask another question can you give me one concrete

00:07:07.100 | example one of these interpretations so interpretations that's a very I'm not specifying what exactly

00:07:14.100 | I want to know about in that query in order for that to work it has to rely on the previous

00:07:19.500 | messages so it's going to need to take a look at this but because it's a chat model and

00:07:23.900 | because we appended the response to those messages it will understand that we're asking

00:07:29.480 | about the interpretation of something around here so we're saying what are the interpretations

00:07:36.100 | of the sort of meaning of life that people have been debating for centuries okay so in

00:07:44.060 | order to answer that it needs to have some of that conversational history which it does

00:07:50.420 | so we ask that question okay and we get this so one concrete example of an interpretation

00:07:57.180 | meaning life is found within the philosophy of existentialism and yeah it's like it's

00:08:04.140 | a pretty good kind of like meaningful answer I think so that's great it's working as an

00:08:11.000 | expert AGI now just to confirm that it is still considering its system message that

00:08:17.140 | we initially set it up with we can say okay something really simple what is the color

00:08:21.540 | of cheese so we can ask that and it says as an AGI system I am unable to provide any information

00:08:28.140 | related to cheese I apologize for the inconvenience and so on and so on right so it's considering

00:08:34.340 | all those previous messages so everything seems to be working and we have rather than

00:08:39.420 | use the typical OpenAI API endpoints we're doing this through Azure's like OpenAI offering

00:08:47.940 | so everything seems to be working that looks good and that is exactly what I wanted to

00:08:53.820 | show you in this video so that is it for now I hope this has been useful and interesting

00:09:01.820 | so thank you very much for watching and I will see you again in the next one. Bye.

00:09:06.300 | [Music]

00:09:16.300 | [Music]

Azure OpenAI in LangChain: Getting Started

Chapters