WEBVTT

00:00.160 --> 00:01.200
Hey, they're eating here.

00:01.200 --> 00:02.240
Just a quick checking.

00:02.240 --> 00:07.520
And I made this video in order to address some feedback I got from students that try to integrate different

00:07.560 --> 00:13.400
lens, and got confused a bit with the stock arguments for those different LM providers.

00:13.680 --> 00:15.760
So I hope this video will clarify this.

00:16.560 --> 00:24.160
So in order to prevent parsing errors, we should always check our template isn't indented and to use

00:24.160 --> 00:26.960
the correct stop arguments for our LM.

00:27.200 --> 00:30.200
So this can change regarding which models you're using.

00:30.360 --> 00:35.440
When I made this video, just using backslash n observation worked for me.

00:35.840 --> 00:42.360
And you might want to add also observation without anything and you might want to try as well.

00:42.400 --> 00:47.200
Observation column and notice for some models like chat OpenAI.

00:47.840 --> 00:55.360
Then according to Link chains newest version, you can bind this argument with the bind method like

00:55.400 --> 00:56.640
you see over here.

00:57.400 --> 01:04.120
And because every vendor updates their SDK in different times, then if you're using Gemini, for example,

01:04.840 --> 01:08.020
currently you need to use the model query Quarrels.

01:08.060 --> 01:08.700
Argument.

01:08.940 --> 01:13.620
And if you want to be on the safe side, you can simply put all of those top tokens to cover all the

01:13.620 --> 01:17.540
cases because it's really sensitive, even for those in dense here.

01:18.500 --> 01:25.740
So if you get a parse error about both a final answer and a possible action, it's almost always because

01:25.740 --> 01:29.420
of an indentation in the prompt template itself.

01:29.620 --> 01:33.100
Or there's something wrong with the stop tokens here.

01:33.380 --> 01:35.460
So this is very, very important.

01:35.460 --> 01:42.500
And if you're having issues I recommend you checking in the featured question of this video here.

01:42.700 --> 01:44.340
Um, this might help for you.

01:44.460 --> 01:51.820
And also it's important to note that after this section, I created an exercise for you where we'll

01:51.820 --> 01:56.180
be migrating from the react pattern to function calling.

01:56.380 --> 02:02.980
And by doing function calling we're going to resolve this issue because we're going to shift the tool

02:03.020 --> 02:04.180
calling to the LM.

02:04.380 --> 02:09.300
But more on that when you finish this section and you'll do the exercise.

02:09.540 --> 02:12.220
So things have become more elegant.

02:12.460 --> 02:15.020
But this is the basis for everything.