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.
115 lines
3.3 KiB
115 lines
3.3 KiB
WEBVTT |
|
|
|
00:00.710 --> 00:03.590 |
|
And welcome to week four, day three. |
|
|
|
00:03.590 --> 00:09.650 |
|
As we are about to embark upon another business project which will add real commercial value. |
|
|
|
00:09.680 --> 00:12.260 |
|
It's going to be an exciting project. |
|
|
|
00:12.260 --> 00:17.450 |
|
It's going to be a difficult challenge, and we're going to have, as always, a blast doing it. |
|
|
|
00:17.480 --> 00:19.250 |
|
Let's get started. |
|
|
|
00:19.280 --> 00:25.400 |
|
So first of all, just a recap of all the things you can do this time I'm going to let you read it, |
|
|
|
00:25.400 --> 00:31.190 |
|
but after today you're going to be able to assess models for coding ability and use frontier models |
|
|
|
00:31.190 --> 00:36.710 |
|
to generate code building a solution that uses Llms for code generation. |
|
|
|
00:36.710 --> 00:39.710 |
|
And of course, it's going to include a user interface. |
|
|
|
00:39.710 --> 00:45.470 |
|
It's going to include a lot of the bits and pieces to package it up nicely to be an app. |
|
|
|
00:45.920 --> 00:47.810 |
|
So a quick reminder of the challenge. |
|
|
|
00:47.810 --> 00:53.660 |
|
We want to build a tool that can convert Python to C plus plus for performance. |
|
|
|
00:53.660 --> 00:57.320 |
|
And this time today we're going to do a frontier model solution. |
|
|
|
00:57.320 --> 01:04.820 |
|
Next time we're going to add open source models as we battle the the frontier closed source versus open |
|
|
|
01:04.820 --> 01:07.520 |
|
source and see what wins out. |
|
|
|
01:08.030 --> 01:12.380 |
|
Uh, first, just a quick a quick explanation of what we're going to do. |
|
|
|
01:12.380 --> 01:18.710 |
|
So we're going to prompt our models with a prompt that explains what we want them to re-implement Python |
|
|
|
01:18.710 --> 01:19.550 |
|
code and C plus. |
|
|
|
01:19.550 --> 01:20.390 |
|
Plus. |
|
|
|
01:20.420 --> 01:24.320 |
|
Um, it's maybe going to be a little bit more wordy than this one, but it'll be along these lines, |
|
|
|
01:24.350 --> 01:30.680 |
|
uh, trying to convince the model not to do any explanation, just simply to provide the C plus plus |
|
|
|
01:30.680 --> 01:31.430 |
|
code. |
|
|
|
01:31.430 --> 01:37.250 |
|
And its objective should be to do whatever it takes to run faster and give the same output. |
|
|
|
01:37.250 --> 01:40.040 |
|
But otherwise, the world is its oyster. |
|
|
|
01:40.070 --> 01:44.120 |
|
It can do what it wants, it just has to give the same output. |
|
|
|
01:44.150 --> 01:49.190 |
|
And this is a bit of code that we'll start with to try out in Python. |
|
|
|
01:49.190 --> 01:56.120 |
|
Uh, you may know already that there is a series which very slowly converges to a nice number, and |
|
|
|
01:56.120 --> 02:05.270 |
|
that series is one minus a third plus a fifth, minus a seventh plus a ninth, minus 11th, and so on. |
|
|
|
02:05.300 --> 02:06.350 |
|
You get the idea. |
|
|
|
02:06.380 --> 02:12.950 |
|
Uh, if you keep that series going and you multiply the result by four, you gradually converge on pi. |
|
|
|
02:12.980 --> 02:14.840 |
|
But it takes a very long time. |
|
|
|
02:14.960 --> 02:21.890 |
|
And so we can run it for the first 100 million iterations, and we will actually get quite close to |
|
|
|
02:21.920 --> 02:23.030 |
|
pi at that point. |
|
|
|
02:23.150 --> 02:28.670 |
|
But you'd have to do an awful lot of typing on your calculator to to do 100 million iterations. |
|
|
|
02:28.730 --> 02:30.830 |
|
Uh, but Python will do it reasonably fast. |
|
|
|
02:30.830 --> 02:35.690 |
|
But we'll be hoping that C plus plus can do it a whole lot faster, and that Frontier models will help |
|
|
|
02:35.690 --> 02:36.410 |
|
us get there. |
|
|
|
02:36.410 --> 02:37.400 |
|
Let's find out.
|
|
|