From the uDemy course on LLM engineering.
https://www.udemy.com/course/llm-engineering-master-ai-and-large-language-models
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
298 lines
8.7 KiB
298 lines
8.7 KiB
WEBVTT |
|
|
|
00:00.050 --> 00:05.660 |
|
So we're now going to start week one of the course when we are going to be looking at exploring frontier |
|
|
|
00:05.690 --> 00:10.880 |
|
models, models at the forefront of what's possible in this first day, we're going to set up an environment |
|
|
|
00:10.880 --> 00:13.310 |
|
and then build your first LM solution. |
|
|
|
00:13.310 --> 00:17.000 |
|
We're then going to talk about how to be positioned for success in the next few weeks. |
|
|
|
00:17.000 --> 00:21.320 |
|
We're then going to compare models at the very frontier of what's possible. |
|
|
|
00:21.320 --> 00:23.900 |
|
Some that you know well and some I think will be new to you. |
|
|
|
00:23.930 --> 00:29.030 |
|
We're then going to talk about transformers, which is the architecture at the heart of the different |
|
|
|
00:29.030 --> 00:32.210 |
|
models that we use in this course, almost all of the models. |
|
|
|
00:32.270 --> 00:38.090 |
|
And then finally, we'll end with a juicy commercial project that you'll be able to use, and I'll have |
|
|
|
00:38.090 --> 00:40.310 |
|
an exercise for you to make it your own. |
|
|
|
00:40.310 --> 00:44.540 |
|
Apply it to different business areas, something that maybe you can use in your day job immediately. |
|
|
|
00:44.540 --> 00:49.580 |
|
Maybe you'll be able to use it as a project that you can have as an example project for yourself. |
|
|
|
00:50.240 --> 00:55.910 |
|
So we're now going to get hopefully some satisfying instant results for you. |
|
|
|
00:55.940 --> 01:00.080 |
|
We already used a llama to do some some fun things on your computer already, but now we're going to |
|
|
|
01:00.080 --> 01:03.890 |
|
take it a notch higher by calling out to APIs directly. |
|
|
|
01:03.980 --> 01:11.630 |
|
But first, the necessary first step is to set up your environment, create a full spec data science |
|
|
|
01:11.630 --> 01:16.700 |
|
environment that will have the the sufficient horsepower to take you through the entire eight week program |
|
|
|
01:16.700 --> 01:17.450 |
|
we have ahead. |
|
|
|
01:17.450 --> 01:22.190 |
|
And there's quite a lot of stuff that we're going to need to install to make sure we're at that point. |
|
|
|
01:22.400 --> 01:28.580 |
|
And I guess I should mention that sometimes, sometimes this experience can be a bit frustrating. |
|
|
|
01:28.580 --> 01:32.090 |
|
Most of the time it goes really smoothly and everything just fits in place. |
|
|
|
01:32.090 --> 01:33.560 |
|
And it's five minutes and you're done. |
|
|
|
01:33.560 --> 01:38.600 |
|
But sometimes some people hit speed bumps for whatever reason, because of something about the system |
|
|
|
01:38.630 --> 01:41.720 |
|
or or some other new effect. |
|
|
|
01:41.780 --> 01:47.720 |
|
And it can be a bit frustrating, but I am here to take you through it step by step and make sure it's |
|
|
|
01:47.720 --> 01:50.420 |
|
successful for you so that this will go smoothly. |
|
|
|
01:50.420 --> 01:55.280 |
|
It'll be behind you in no time, and we'll be on to actual projects, and we'll never have to worry |
|
|
|
01:55.280 --> 01:58.250 |
|
about environment setup again for the next eight weeks. |
|
|
|
01:58.250 --> 02:04.590 |
|
So creating the full spec data science environment, it ain't as easy as I wish it were. |
|
|
|
02:04.950 --> 02:07.860 |
|
There is some stuff to do, but we'll get through it fast. |
|
|
|
02:07.860 --> 02:11.910 |
|
First of all, we'll clone the repo for people who are familiar with git. |
|
|
|
02:11.910 --> 02:16.500 |
|
That's a that's easy, but but I will make sure I walk you through it in case this is a new thing. |
|
|
|
02:16.620 --> 02:21.930 |
|
Then we will go through the Readme instructions to set up the environment. |
|
|
|
02:21.960 --> 02:27.180 |
|
Now I recommend that we use something called Anaconda that some of you may have already used Anaconda. |
|
|
|
02:27.180 --> 02:30.960 |
|
It's a high powered thing and it's quite heavyweight. |
|
|
|
02:30.960 --> 02:37.140 |
|
It creates an entire environment for you, a dedicated, isolated environment on your computer for this |
|
|
|
02:37.140 --> 02:43.200 |
|
course, and it does so in a way that guarantees high compatibility between what you're doing on your |
|
|
|
02:43.200 --> 02:46.590 |
|
computer and what I do on mine and everyone else on this course. |
|
|
|
02:46.680 --> 02:48.900 |
|
So it's very good at that. |
|
|
|
02:49.170 --> 02:50.550 |
|
But it is quite heavyweight. |
|
|
|
02:50.550 --> 02:55.110 |
|
It involves downloading a lot from the internet and installs a lot because it builds everything from |
|
|
|
02:55.110 --> 02:56.130 |
|
the ground up. |
|
|
|
02:56.520 --> 02:59.100 |
|
If it causes you problems, there's an alternative. |
|
|
|
02:59.100 --> 03:01.080 |
|
We can use a simpler approach. |
|
|
|
03:01.080 --> 03:06.440 |
|
The standard Python virtual environment approach with Pip, and I've got instructions for that too. |
|
|
|
03:06.470 --> 03:13.580 |
|
It's quicker, it's simpler, and it's just not as guaranteed to be compatible as Anaconda, so I'd |
|
|
|
03:13.580 --> 03:15.740 |
|
prefer we try Anaconda if possible. |
|
|
|
03:15.770 --> 03:17.150 |
|
It does work most of the time. |
|
|
|
03:17.150 --> 03:18.380 |
|
Any problems at all? |
|
|
|
03:18.380 --> 03:20.900 |
|
The Python virtual environment is your fallback. |
|
|
|
03:20.930 --> 03:24.620 |
|
We then need to set up an OpenAI key and I'll talk more about that. |
|
|
|
03:24.620 --> 03:28.310 |
|
I'll talk about API costs and what to do if that's a that's a problem for you. |
|
|
|
03:28.400 --> 03:29.930 |
|
We'll go through that at the time. |
|
|
|
03:29.930 --> 03:35.960 |
|
We have to create something called a EMV file, which has your private keys in it. |
|
|
|
03:35.960 --> 03:40.460 |
|
And it's going to be important that that file is called exactly dot EMV, as we'll see. |
|
|
|
03:40.460 --> 03:45.410 |
|
I will explain, but it can't be called keys dot EMV or dot EMV dot text. |
|
|
|
03:45.440 --> 03:47.600 |
|
It has to be called exactly dot EMV. |
|
|
|
03:48.050 --> 03:53.840 |
|
And then finally we activate your environment and we run JupyterLab to start the program. |
|
|
|
03:54.560 --> 03:59.720 |
|
So as I say, most of the time all of this will go great. |
|
|
|
03:59.930 --> 04:06.190 |
|
Most people report that everything goes through fine, but if it doesn't, then then there are some |
|
|
|
04:06.190 --> 04:09.460 |
|
secret weapons at your disposal to make sure that it does. |
|
|
|
04:09.490 --> 04:15.490 |
|
The first of them is that I've put a troubleshooting notebook that I will show you in, in the in, |
|
|
|
04:15.520 --> 04:20.650 |
|
in the GitHub repo, which will take you step by step through diagnosing problems, running some code |
|
|
|
04:20.650 --> 04:24.340 |
|
that will figure out what's wrong, and then fixing each problem in turn. |
|
|
|
04:24.340 --> 04:25.900 |
|
So that is there. |
|
|
|
04:25.930 --> 04:31.840 |
|
The second resource you have is a kind of surprising one at ChatGPT and and Claude. |
|
|
|
04:31.870 --> 04:34.840 |
|
They are spooky in how good they are. |
|
|
|
04:34.870 --> 04:42.670 |
|
If you say a problem and paste like a stack trace in there, they are very, uh, impressive at giving |
|
|
|
04:42.670 --> 04:45.310 |
|
quite nuanced answers about what you need to do. |
|
|
|
04:45.340 --> 04:48.970 |
|
I've been caught off guard by the times I've run into an infrastructure problem, and I thought, this |
|
|
|
04:48.970 --> 04:49.870 |
|
is way too much. |
|
|
|
04:49.870 --> 04:52.300 |
|
I can't even get the answer on Stack Overflow. |
|
|
|
04:52.480 --> 04:55.570 |
|
Uh, but I've been rescued by Claude. |
|
|
|
04:55.660 --> 04:57.400 |
|
I think Claude is the stronger of the two. |
|
|
|
04:57.430 --> 04:57.640 |
|
But. |
|
|
|
04:57.640 --> 04:59.440 |
|
But both of them give good answers. |
|
|
|
04:59.440 --> 05:04.990 |
|
So it's worth realizing that you can often just fix the problem by pasting it in there. |
|
|
|
05:05.020 --> 05:06.670 |
|
It's worth giving it a shot. |
|
|
|
05:06.730 --> 05:12.140 |
|
But if if if those two fail, or even if you'd just rather go straight to this third option. |
|
|
|
05:12.140 --> 05:15.770 |
|
The third option is there, and the third option is to contact me. |
|
|
|
05:15.890 --> 05:17.780 |
|
I am standing by to help you. |
|
|
|
05:17.780 --> 05:19.070 |
|
That is my job. |
|
|
|
05:19.130 --> 05:20.990 |
|
So please, please reach out. |
|
|
|
05:20.990 --> 05:23.900 |
|
You can reach out to me by messaging me on the platform. |
|
|
|
05:23.900 --> 05:25.130 |
|
You can email me. |
|
|
|
05:25.160 --> 05:28.700 |
|
My email address is in the is in the readme and you can also linked in with me. |
|
|
|
05:28.700 --> 05:30.500 |
|
I love getting LinkedIn connections by the way. |
|
|
|
05:30.560 --> 05:36.650 |
|
So LinkedIn with me for sure and message me there too and you'll find that I'm very responsive. |
|
|
|
05:36.740 --> 05:41.900 |
|
I typically if I'm not sleeping or traveling, then you'll see from my other answers that I get back |
|
|
|
05:41.990 --> 05:47.210 |
|
very quickly, and I see it as my job to fix it and to get you up and running. |
|
|
|
05:47.210 --> 05:48.320 |
|
And I have never failed. |
|
|
|
05:48.320 --> 05:55.340 |
|
Yet I have always, I think, 5000 people through and everyone has a running environment, as will you. |
|
|
|
05:55.340 --> 05:59.570 |
|
So if you do hit any speed bumps at all, you can go with these other two options. |
|
|
|
05:59.570 --> 06:00.860 |
|
But you can come straight to me. |
|
|
|
06:00.860 --> 06:01.670 |
|
I will fix it. |
|
|
|
06:01.670 --> 06:03.200 |
|
I will find what's going on. |
|
|
|
06:03.200 --> 06:05.690 |
|
We will have you up and running quickly. |
|
|
|
06:05.840 --> 06:06.680 |
|
All right. |
|
|
|
06:06.680 --> 06:11.630 |
|
With that, let me now take you on a walkthrough of what it takes to set up your environment.
|
|
|