LlamaIndex: How to Get Structured Data from LLMs

Key Takeaways

hey everyone how's it going today's video very quick about how to get structured output out of your language model in Lama index okay now the idea behind this is that usually when you have your language model and you query it you're going to get just text from it okay now what happens if instead of getting text you want to get something like a Json file with very specific keys and value um formats that you want from that Json file okay your average language model is not going to be able to accurately get that just by prompting it okay so the idea that we have right here is that we're going to be initializing a schema using pantic adding it to our language model using Lama index and this is the language model that we're going to query and every time we're going to get a response with the schema that we have determined okay so let's take a look at how to do that [Music] okay so the first thing to do as I mentioned before is to create the schema of the data that we want to get in our case we're going to be using pantic in order to uh initialize our schema in case you're not familiar with it pantic is essentially just a library uh for python for high for data validation and essentially what it allows you to do is to create classes which are going to become the schemas of your data and you're going to be able to validate your data using these schemas okay um in this case I'm going to be asking my language model to create an album and an album is going to contain a name the name is always going to be a string an artist which is always going to be a string and a list of songs and each song is also going to have its own schema and it it's going to have a title and a length in seconds and the length in seconds is in integers okay now for the record this example comes straight from the documentation of L index um but let's actually take a look at this so I'm going to execute this and right here I'm going to go from L index core import a chat message and here as you can see I had previously initialized my language model let me just show you before because I was doing another tutorial before this one um so here I have initialized my language model from open AI okay so in this case I'm going to be using open ey for this I am using GPT 40 mini and this language model is the same one that I'm using right here however this is not the instance that I'm going to be quering like that I am actually going to going to be running as structured llm method on it and I'm going to Define this parameter right here output CLS as the uh output schema that I want my language model to return on every single location okay and in this case I am passing it the album schema that I created right here which contains a reference to the song schema as well okay and this I am going to be assigning it to a variable called s llm for structured llm and now anytime I am going to query this this language model right here I am going to get the response in the exact schema that I specialized right here so let's see I'm going to initialize a chat model um sorry a chat message from this string right here generate an example album for the film uh for the film uh who frame Roger Rabbit and let's see what it returns to us I am going to call the slm using the chat method and let's see now here we have the response and as you can see here the chat message is from the assistant and the content is not just the string of text it is actually the object that I want to get so here you have the name of the album is Who Framed Roger Rabbit the soundtrack um Roger Rabbit soundtrack sorry the artist is going to be various artists and then we have a list of songs okay so let's take a look at it under the microscope and here I'm for the record using pretty print I imported It Up Above in the dictionary in the notebook but essentially I'm using pretty print to print these objects more neatly but as you can see we have the artist right here various artists the name is who frame Roger Rabbit and the songs which are all of them I mean each one of them an element an instance of song which each one has a title and a length in seconds so there you go I mean as you can see this follows exactly the same um the same schema that we defined above right here so there we go that is how to use on how to use structured outputs in language models using L index let me know if you have any questions and let's continue with this course [Music]