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.
 
 

325 lines
9.4 KiB

WEBVTT
00:00.050 --> 00:02.180
Congratulations are definitely in order.
00:02.210 --> 00:06.380
Yesterday was a mammoth first day on this course and you got through it.
00:06.380 --> 00:09.560
You set up your environment and we're off to the races.
00:09.560 --> 00:14.840
So as I recap, what you can now do is use a llama to run models locally.
00:14.840 --> 00:18.410
You can call OpenAI's API to run a frontier model.
00:18.410 --> 00:22.130
You understand, at least at a high level, the difference between a system and a user prompt.
00:22.130 --> 00:28.730
And you've built a summarization use case, which is an important commercial application.
00:29.120 --> 00:35.270
Today we're going to talk a bit about what are the steps to get to being an LLM engineer, set you up
00:35.270 --> 00:39.110
for success, and then talk more about frontier models.
00:39.110 --> 00:41.030
That is the plan.
00:41.060 --> 00:47.750
Now, let me start by just emphasizing that this course, this eight week course is a practical course.
00:47.750 --> 00:50.480
First and foremost we're going to cover theory.
00:50.510 --> 00:53.420
We're going to cover the foundational information that you need.
00:53.420 --> 00:56.600
But we're always going to do it in a practical context.
00:56.630 --> 00:59.810
I'm a big believer in the best way to learn is by doing.
00:59.810 --> 01:01.640
And that's what we'll be actually doing.
01:01.820 --> 01:03.620
And we're going to be putting it to good use.
01:03.620 --> 01:06.620
We're going to be building commercial projects as we go.
01:06.620 --> 01:11.390
And I'll often be asking you to try and think about how you could apply what you've learned to your
01:11.390 --> 01:15.710
day job and try and build little prototypes to bring that to life.
01:15.710 --> 01:18.470
So that is the flavour of the course.
01:18.950 --> 01:24.740
There are three different aspects to LM engineering that you'll be picking up as we go.
01:24.770 --> 01:29.960
The first of them is just understanding the models that are out there, the wide range of different
01:29.990 --> 01:34.610
types of LM and what they're capable of, whether we're talking about open source versions, the closed
01:34.610 --> 01:40.220
source paid versions, ones that are multi-modal, that can generate images or audio, the different
01:40.220 --> 01:45.920
architectures of LMS, and importantly, how you pick which LM you should be working with in the first
01:45.920 --> 01:46.610
place.
01:46.610 --> 01:53.270
We'll be looking at all of the tools of the trade, things like the ubiquitous hugging face, the super
01:53.270 --> 01:58.490
impressive glue code in Lange chain, the fabulous gradio that you will see.
01:58.520 --> 02:01.130
I'm a big fan of weights and biases.
02:01.130 --> 02:05.270
Super helpful and then modal for deploying it to production.
02:05.270 --> 02:10.790
And we'll be looking at techniques different ways that you can apply this.
02:10.880 --> 02:18.590
The field of AI to solve business problems from using API's Rag which is such a hot topic, fine tuning
02:18.590 --> 02:23.360
and then at the end a full Agentic AI solution.
02:24.830 --> 02:30.170
The idea of this course is that I've planned it so that it will apply almost no matter what your level
02:30.170 --> 02:31.400
of experience.
02:31.640 --> 02:36.350
You may find initially that some of the first few weeks are too simplistic for you, or the first few
02:36.350 --> 02:37.130
days at least.
02:37.130 --> 02:42.980
And there I would say hang in there, use it as a way to just sort of reinforce some of the of the foundational
02:42.980 --> 02:47.570
information that we'll talk a bit more detail about things like tokens than you've done before, perhaps.
02:47.570 --> 02:52.400
So there'll be stuff to pick up, but you can go through it quickly and then make the projects your
02:52.400 --> 02:52.730
own.
02:52.730 --> 02:59.600
So use it as a way to build deeper versions of what we do and then prepare for harder, fun projects
02:59.600 --> 03:00.980
coming up later.
03:01.190 --> 03:05.510
If it feels too challenging, then please, please don't worry.
03:05.540 --> 03:07.910
Take your time with the practicals.
03:07.940 --> 03:09.860
Take your time with the exercises.
03:09.860 --> 03:11.210
Work your way through them.
03:11.210 --> 03:15.650
There are some extra guides that I'll talk about in week one that would help give you a sort of firmer
03:15.650 --> 03:16.790
footing if you need it.
03:16.790 --> 03:19.550
And please, please, please ask for help.
03:19.550 --> 03:21.620
I am here, I respond quickly.
03:21.620 --> 03:27.290
You can always reach out to me, either as I say on the platform or through email or through LinkedIn.
03:27.320 --> 03:29.180
Details are in the GitHub repo.
03:29.180 --> 03:30.290
Reach out to me.
03:30.320 --> 03:31.130
Get help.
03:31.130 --> 03:32.600
That is what I'm here for.
03:32.900 --> 03:35.810
And then if it feels just right, then excellent.
03:35.810 --> 03:36.770
Keep going.
03:37.580 --> 03:43.880
So the prerequisite is beginner to intermediate level Python.
03:43.880 --> 03:49.370
And if you have intermediate level Python you're going to find it easiest and you'll get the most out
03:49.370 --> 03:50.060
of it.
03:50.180 --> 03:55.280
And if you look for example at this line of code right here, which is just a random line from within
03:55.280 --> 03:55.910
a function.
03:55.910 --> 04:01.280
If you basically know what that's probably doing, then you're in great shape if you know exactly what
04:01.280 --> 04:07.250
it's doing and if you know in fact why, it's not perhaps the most optimal way of doing it, then you're
04:07.250 --> 04:12.260
more than more than, well set that you're advanced, and that's great if you don't know what this does
04:12.290 --> 04:19.910
and you're not familiar with a world with a word like yield or set or the dot get, then there is a
04:19.910 --> 04:27.890
special notebook, a special Jupyter Lab in week one for you, which is a guide to to Python at this
04:27.890 --> 04:28.490
level.
04:28.490 --> 04:33.890
And as you go through that notebook, I will take you through each of the stepping stones until we get
04:33.890 --> 04:36.620
to a point where a line like this should make sense.
04:36.620 --> 04:40.550
And of course, you can also use ChatGPT and Claude.
04:40.580 --> 04:45.320
The the the genies are really good at explaining code.
04:45.320 --> 04:49.310
And indeed, if you put something like this in there, they would tell you exactly what it does and
04:49.310 --> 04:50.780
why and step you through it too.
04:50.780 --> 04:54.200
And they're probably just as good as, as my, my notebook.
04:54.200 --> 04:57.050
So either way, that should give you what you need to do.
04:57.050 --> 05:01.640
And at any point, if you don't understand some code, you can always use ChatGPT.
05:03.230 --> 05:08.630
So to get the most out of this course, there are a few things that I would ask you.
05:08.660 --> 05:12.410
First of all, follow along as I do my coding.
05:12.410 --> 05:16.640
So when I'm when I'm going through in the labs and I'm executing cells, then either at the same time
05:16.670 --> 05:18.890
or afterwards go through and do it yourself.
05:18.890 --> 05:25.100
And if you hit snafus, which you might do for various reasons, then have a crack at trying to to sort
05:25.130 --> 05:25.640
out why.
05:25.670 --> 05:26.540
Do some debugging.
05:26.540 --> 05:33.380
That's a great way to learn and complete the exercises, and then put your code examples up on GitHub.
05:33.380 --> 05:34.670
It's actually a great way as well.
05:34.670 --> 05:39.770
If you're new to this space, and you're trying to build up something of a kind of resume to show that
05:39.770 --> 05:45.290
you've built some of this experience, the best kind of resume you can have is a GitHub repo, because
05:45.290 --> 05:48.380
people will look at it and will see the sorts of things you've worked on.
05:48.380 --> 05:51.170
And of course, you don't want to put exactly the projects we do.
05:51.200 --> 05:52.910
You want to make them your own.
05:52.910 --> 05:57.800
You want to figure out, okay, how can I apply this to my business area or to a personal project that
05:57.800 --> 06:01.160
I'm working on to make it something that's similar, slightly different?
06:01.190 --> 06:03.950
Take on it for new business value.
06:04.010 --> 06:10.760
That is a great way to get the most out of this course, and then look to share your code.
06:10.760 --> 06:13.130
If you're happy with it and you and you're okay with that.
06:13.130 --> 06:18.140
I've got instructions about how you can submit a pull request, which means that I can see your code.
06:18.170 --> 06:23.660
I can give you feedback on it if you'd like, and I can also then republish it so that other students
06:23.660 --> 06:27.590
taking the course will see your examples and we can all share in it together.
06:27.590 --> 06:29.540
And of course stick at it.
06:29.570 --> 06:30.950
Hang on in there.
06:30.980 --> 06:34.550
This course gets better and better and better, I assure you.
06:34.550 --> 06:39.440
And I you know, there's there's going to be so many projects, so much commercial application.
06:39.440 --> 06:43.550
And I definitely encourage you to stay the course.
06:43.580 --> 06:48.410
And the thing I didn't put down here one more time is that getting the most out of the course.
06:48.410 --> 06:52.220
Also, please reach out to me if I can help at any point.