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.
181 lines
5.5 KiB
181 lines
5.5 KiB
WEBVTT |
|
|
|
00:00.110 --> 00:05.630 |
|
Welcome back to Jupyter Lab, where I want to show you the assignment, the homework exercise for you |
|
|
|
00:05.630 --> 00:09.170 |
|
for the end of week one, coming into the week one folder. |
|
|
|
00:09.170 --> 00:14.330 |
|
And you'll see here week one exercise so you can see there's rather little here. |
|
|
|
00:14.330 --> 00:16.550 |
|
There's a little framework to tell you what we're doing. |
|
|
|
00:16.550 --> 00:23.150 |
|
So the idea is for you to build a little tool that will be used by yourself as something that can be |
|
|
|
00:23.150 --> 00:25.130 |
|
your guide during this course. |
|
|
|
00:25.160 --> 00:33.170 |
|
It's going to be a tool which will be able to answer questions on code, on llms, on things that you |
|
|
|
00:33.170 --> 00:39.290 |
|
might run into along the way and be your sidekick, your your customized copilot for this experience. |
|
|
|
00:39.290 --> 00:41.120 |
|
And this is what it's going to do. |
|
|
|
00:41.240 --> 00:44.300 |
|
Uh, it's going to you're going to fill in some inputs. |
|
|
|
00:44.300 --> 00:47.840 |
|
You're going to you can just use these two constants we've got here. |
|
|
|
00:47.840 --> 00:55.790 |
|
You're going to be using both GPT and open source llama running in llama on your computer, something |
|
|
|
00:55.790 --> 01:00.140 |
|
which you're familiar with from our week two assignment or day two assignment. |
|
|
|
01:00.170 --> 01:03.300 |
|
You'll put some code in here to set up your environment. |
|
|
|
01:03.360 --> 01:09.780 |
|
And then this is where you will type the question that you want to ask your tutor, your personal private |
|
|
|
01:09.780 --> 01:10.500 |
|
tutor. |
|
|
|
01:10.500 --> 01:15.330 |
|
And in this case, I am asking it to explain what this code does and why. |
|
|
|
01:15.360 --> 01:20.340 |
|
And this again, is that line of code that we looked at a couple of times, which is has a few different |
|
|
|
01:20.340 --> 01:24.270 |
|
things going on there, and this is the better version of the code than the one I originally wrote. |
|
|
|
01:24.270 --> 01:26.370 |
|
This is the version that I think it was. |
|
|
|
01:26.400 --> 01:31.230 |
|
Yeah, it was GPT four with canvas that wrote this version of it. |
|
|
|
01:31.500 --> 01:39.450 |
|
So we will provide this, and then you will fill in this piece of code here, which will get GPT four |
|
|
|
01:39.450 --> 01:40.530 |
|
mini to answer. |
|
|
|
01:40.530 --> 01:47.310 |
|
And you should try and answer and have that response stream back and use markdown, uh, so that it |
|
|
|
01:47.310 --> 01:49.560 |
|
appears in full glory. |
|
|
|
01:49.620 --> 01:55.620 |
|
Uh, and then do the same thing again for llama 3.2 and see the answer there. |
|
|
|
01:55.620 --> 02:00.090 |
|
So this will need you to to write the code from scratch and sort of type it out. |
|
|
|
02:00.090 --> 02:06.460 |
|
And of course you should look back at the fragments of code that we came up with on the prior days and |
|
|
|
02:06.460 --> 02:12.010 |
|
used that to build this, and hopefully it will be something that will come reasonably naturally and |
|
|
|
02:12.010 --> 02:16.360 |
|
you'll be used to the familiar .0., blah, blah, blah, blah, blah. |
|
|
|
02:16.360 --> 02:22.390 |
|
It's all going to be something that will be in your memory and you'll also be very comfortable, I hope, |
|
|
|
02:22.390 --> 02:28.120 |
|
with creating the messages list of dictionaries, something that you're used to at this point. |
|
|
|
02:28.240 --> 02:31.600 |
|
Um, and when you do that, you'll see what answer that you get for this. |
|
|
|
02:31.600 --> 02:34.750 |
|
And it's actually a remarkably good answer. |
|
|
|
02:34.750 --> 02:36.850 |
|
Again, I think much better than I could do. |
|
|
|
02:37.000 --> 02:42.010 |
|
Uh, and it would be interesting to see whether you're also happy with the answer that you get from |
|
|
|
02:42.010 --> 02:46.000 |
|
a minuscule model that is much, much smaller by comparison. |
|
|
|
02:46.000 --> 02:47.950 |
|
And see how you compare those two. |
|
|
|
02:47.980 --> 02:52.480 |
|
And then in the future, you'll be able to use this as a bit of a resource. |
|
|
|
02:52.630 --> 02:55.450 |
|
And so I hope this this is useful for you. |
|
|
|
02:55.480 --> 02:59.740 |
|
And in fact, after week two, you might want to come back to this and add a user interface to it. |
|
|
|
02:59.740 --> 03:04.450 |
|
So you could have like a little tool running on the side that would be uh, could be super handy. |
|
|
|
03:04.630 --> 03:09.650 |
|
Uh, one also thing that you might do is that what I've done here is I've just put down the question |
|
|
|
03:09.650 --> 03:11.060 |
|
as a global variable. |
|
|
|
03:11.060 --> 03:15.350 |
|
You might instead want to see if you can think about a way that you could take that in as an input and |
|
|
|
03:15.350 --> 03:16.880 |
|
be a bit more dynamic about this. |
|
|
|
03:16.880 --> 03:21.890 |
|
So it could just be a cell that you run, but maybe do that as a as a later improvement. |
|
|
|
03:22.430 --> 03:28.910 |
|
Now I've done this and the solution is sitting in this solutions folder, but I urge you not to look |
|
|
|
03:28.940 --> 03:29.840 |
|
at the solution. |
|
|
|
03:29.840 --> 03:33.350 |
|
Uh, not not unless you you find yourself feeling stuck. |
|
|
|
03:33.380 --> 03:34.730 |
|
There's nothing worse than feeling stuck. |
|
|
|
03:34.760 --> 03:41.870 |
|
If you do feel stuck, you could always try first asking ChatGPT or Claude for a nudge, but otherwise, |
|
|
|
03:41.870 --> 03:43.790 |
|
please, please, please reach out to me. |
|
|
|
03:43.790 --> 03:46.190 |
|
There is no reason whatsoever to be stuck with this. |
|
|
|
03:46.220 --> 03:49.160 |
|
It should be something that you that you're able to to get through. |
|
|
|
03:49.160 --> 03:50.750 |
|
And I would love to help you with that. |
|
|
|
03:50.750 --> 03:57.710 |
|
Get you to a point when you have your own personal, technical and data science tutor to help you along |
|
|
|
03:57.710 --> 03:58.190 |
|
the way. |
|
|
|
03:58.190 --> 04:06.560 |
|
So, um, above all, be sure to have fun with this and I will see you back for the wrap up video for |
|
|
|
04:06.560 --> 04:07.220 |
|
week one.
|
|
|