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.
 
 

79 lines
2.2 KiB

WEBVTT
00:00.650 --> 00:08.030
I'm so happy to welcome you to week four, day four on our journey to LLM Engineering and Mastery.
00:08.180 --> 00:12.620
Here we are looking again at Code Generation.
00:12.620 --> 00:19.190
We had tons of fun last time generating code using frontier models, using what ended up being a very
00:19.190 --> 00:26.150
sharp UI to allow us to try out different kinds of puzzles and see the startling performance improvement
00:26.150 --> 00:31.040
that Claude was able to generate for us winning the day in our competition.
00:31.040 --> 00:39.380
What was it more than 60,000 times faster when it reinterpreted the problem and applied Shannon's algorithm?
00:39.770 --> 00:42.740
So today we're moving to open source.
00:42.740 --> 00:45.860
You're going to be able to use open source models for coding.
00:45.860 --> 00:50.690
We're going to use something called hugging face endpoints, which is another feature of hugging Face
00:50.690 --> 00:56.900
that lets you deploy models and run them in the cloud privately for you for inference purposes.
00:57.170 --> 01:02.960
So it's another great feature of the fabulous resource that is hugging face.
01:03.230 --> 01:10.490
And we're going to build a solution that uses open source LMS along with frontier LMS to generate code.
01:10.520 --> 01:11.720
Let's see.
01:12.800 --> 01:15.260
First, quick reminder of the challenge.
01:15.290 --> 01:16.250
I'm sure you remember it.
01:16.280 --> 01:21.980
We're building a product that is able to convert C plus, Plus or Python code into high performance
01:22.010 --> 01:23.330
C plus plus code.
01:23.600 --> 01:32.450
And we used GPT four and Claude last time, uh, the Pi program ran 100 times faster.
01:32.750 --> 01:39.950
Um, GPT four optimized the code at least, uh, when it when it ran, it was able to do it with a speed
01:39.980 --> 01:40.490
up.
01:40.490 --> 01:45.110
Uh, Claude rewrote the algorithm for more than 60,000 times.
01:45.110 --> 01:46.550
Performance improvement.
01:46.550 --> 01:48.350
Fantastic.
01:48.380 --> 01:55.640
Now the time has come to see how open source measures up against the beasts that are the frontier models.
01:55.670 --> 01:57.050
Let's have a try.
01:57.080 --> 01:58.820
See you in a second.