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.
139 lines
3.8 KiB
139 lines
3.8 KiB
WEBVTT |
|
|
|
00:00.770 --> 00:07.670 |
|
Now, look, I know that I went through that very fast, but maybe, uh, you're still, uh, blinking |
|
|
|
00:07.700 --> 00:08.450 |
|
at the end of that. |
|
|
|
00:08.450 --> 00:13.490 |
|
But that's because the point is that you should go back now and do this yourself and see it. |
|
|
|
00:13.490 --> 00:17.540 |
|
And as you run that code and you see what's going on in the modal screens, I think it's going to make |
|
|
|
00:17.540 --> 00:18.770 |
|
complete sense. |
|
|
|
00:18.950 --> 00:24.470 |
|
Um, in case you have any problems with that, uh, hugging face token, I'm going to put better instructions |
|
|
|
00:24.470 --> 00:25.310 |
|
in the Jupyter lab. |
|
|
|
00:25.310 --> 00:26.570 |
|
So that's very clear for you. |
|
|
|
00:26.570 --> 00:32.540 |
|
And yeah, if I think you'll find that it will, that it will be fairly straightforward and you'll see |
|
|
|
00:32.540 --> 00:34.820 |
|
how that works and why it's working so fast. |
|
|
|
00:34.850 --> 00:37.010 |
|
There's the first time you run it for a while. |
|
|
|
00:37.010 --> 00:38.660 |
|
There's several minutes for it to warm up. |
|
|
|
00:38.660 --> 00:44.120 |
|
But then subsequently, because we've cached the model weights and we've loaded it into memory, it |
|
|
|
00:44.120 --> 00:47.390 |
|
should run quickly as it did just then. |
|
|
|
00:47.450 --> 00:54.650 |
|
So with that, you have now learned how to take a model and how to deploy it to production so that people |
|
|
|
00:54.650 --> 01:02.450 |
|
could call it just with Python code for production purposes within applications outside something like |
|
|
|
01:02.480 --> 01:03.590 |
|
a Jupyter Lab. |
|
|
|
01:04.040 --> 01:07.490 |
|
And hopefully you're now beginning to appreciate that. |
|
|
|
01:07.490 --> 01:12.350 |
|
We do have a big week, and it is an epic project, and there's a lot to be done. |
|
|
|
01:12.350 --> 01:16.790 |
|
In fact, uh, the next day's worth of activities is the biggest of the lot. |
|
|
|
01:16.820 --> 01:18.740 |
|
There's an awful lot happening. |
|
|
|
01:18.860 --> 01:26.700 |
|
Um, but just to remind yourself for today that that this was about deploying models in production using |
|
|
|
01:26.700 --> 01:30.390 |
|
modal, the serverless platform. |
|
|
|
01:30.600 --> 01:37.350 |
|
In some ways, it's similar to when we deployed a model or I deployed a model to the hugging face using |
|
|
|
01:37.380 --> 01:38.940 |
|
hugging face endpoints. |
|
|
|
01:39.120 --> 01:44.850 |
|
Um, but you can see the the extra functionality that you get with this, the ability to configure infrastructure |
|
|
|
01:44.850 --> 01:49.650 |
|
with code and the way that the pricing works, it's a very, very powerful platform. |
|
|
|
01:49.920 --> 01:57.330 |
|
Next time you'll be able to build an advanced Wragg solution, you're saying I already you already got |
|
|
|
01:57.330 --> 01:57.690 |
|
Wragg. |
|
|
|
01:57.690 --> 01:58.620 |
|
We've done Wragg. |
|
|
|
01:58.620 --> 01:59.670 |
|
We know Wragg well. |
|
|
|
01:59.670 --> 02:01.140 |
|
You're going to know it even more. |
|
|
|
02:01.140 --> 02:01.650 |
|
Next time. |
|
|
|
02:01.650 --> 02:03.030 |
|
We're going to use Wragg. |
|
|
|
02:03.030 --> 02:05.010 |
|
We're going to do it directly without chain. |
|
|
|
02:05.010 --> 02:05.940 |
|
We're pros now. |
|
|
|
02:05.940 --> 02:06.930 |
|
We don't need long chain. |
|
|
|
02:06.930 --> 02:08.070 |
|
We can do it ourselves. |
|
|
|
02:08.070 --> 02:14.820 |
|
We're going to look things up in a Chrome data store and use it to give context to a model, but it's |
|
|
|
02:14.820 --> 02:18.180 |
|
going to be an enormous great data store. |
|
|
|
02:18.180 --> 02:23.790 |
|
And we're going to build something called an ensemble model, which is a kind of model that combines |
|
|
|
02:23.790 --> 02:25.590 |
|
the best of multiple models. |
|
|
|
02:25.590 --> 02:32.400 |
|
And we're going to be able to deliver production ready code that will span multiple models. |
|
|
|
02:32.400 --> 02:39.360 |
|
So it's going to be about, uh, really strengthening your skill set, building expertise as you make |
|
|
|
02:39.360 --> 02:46.140 |
|
the transition from being a knowledgeable in LM engineering to being a master of LM engineering. |
|
|
|
02:46.350 --> 02:48.330 |
|
And with that, I'll see you next time.
|
|
|