HTML Form Extractor Model
In this walkthrough, you'll build an LLM that can extract HTML form elements. Feel free to also run the Colab notebook.
Import the LLM engine from the llama module
from llama import LLMEngine
llm = LLMEngine(id="extract_html", model_name="EleutherAI/pythia-410m-v0")
The Pythia 410M-parameter model is the largest on the free-tier to train for free :)
Define the LLM interface
Define the input and output types. Be sure to include the Context. This helps
the LLM understand your types in natural language.
from llama import Type, Context
class Input(Type):
html_dom : str = Context("html dom of a form page with first name, last name, and credit card number fields filled out")
class Output(Type):
first_name : str = Context("first name of the person")
last_name : str = Context("last name of the person")
credit_card_number : str = Context("credit card number of the person")
Add example data to the model for training
You can import data to this class by using the save_data method, which accepts a list of input-output pairs, formatted into your types, above.
Code for get_data()
def get_data():
data = [
[Input(html_dom="""<!DOCTYPE html>
<html>
<head>
<title>Form Example</title>
</head>
<body>
<form>
<div>
<label for="first_name">First Name:</label><br>
<input type="text" id="first_name" name="first_name" value="John" readonly>
</div>
<div>
<label for="last_name">Last Name:</label><br>
<input type="text" id="last_name" name="last_name" value="Doe" readonly>
</div>
<div>
<label for="cc_number">Credit Card Number:</label><br>
<input type="password" id="cc_number" name="cc_number" value="**** **** **** 1234" readonly>
</div>
<div>
<input type="submit" value="Submit">
</div>
</form>
</body>
</html>
"""), Output(first_name='John', last_name='Doe', credit_card_number='**** **** **** 1234')],
[Input(html_dom="""<!DOCTYPE html>
<html>
<head>
<title>Form Example</title>
</head>
<body>
<form>
<div>
<label for="first_name">First Name:</label><br>
<input type="text" id="first_name" name="first_name" value="Kelly" readonly>
</div>
<div>
<label for="last_name">Last Name:</label><br>
<input type="text" id="last_name" name="last_name" value="Seegal" readonly>
</div>
<div>
<label for="cc_number">Credit Card Number:</label><br>
<input type="password" id="cc_number" name="cc_number" value="493929193" readonly>
</div>
<div>
<input type="submit" value="Submit">
</div>
</form>
</body>
</html>
"""), Output(first_name='Kelly', last_name='Seegal', credit_card_number='493929193')],
[Input(html_dom="""<!DOCTYPE html>
<html>
<head>
<title>Form Example</title>
</head>
<body>
<form>
<table>
<tr>
<td>
<label for="full_name">Full Name:</label>
</td>
<td>
<input type="text" id="full_name" name="full_name" value="Ren Farley" readonly>
</td>
</tr>
<tr>
<td>
<label for="cc_number">Credit Card Number:</label>
</td>
<td>
<input type="password" id="cc_number" name="cc_number" value="90193990202" readonly>
</td>
</tr>
<tr>
<td colspan="2">
<input type="submit" value="Submit">
</td>
</tr>
</table>
</form>
</body>
</html>
"""), Output(first_name='Ren', last_name='Farley', credit_card_number='90193990202')],
[Input(html_dom="""<!DOCTYPE html>
<html>
<head>
<title>Form Example</title>
</head>
<body>
<form>
<div id="personal_details">
<h3>Personal Details:</h3>
<label for="full_name">Full Name:</label><br>
<input type="text" id="full_name" name="full_name" value="Ren Farley" readonly><br>
</div>
<div id="credit_card_details">
<h3>Credit Card Details:</h3>
<label for="cc_number">Credit Card Number:</label><br>
<input type="password" id="cc_number" name="cc_number" value="90193990202" readonly><br>
<label for="expiry_date">Expiry Date:</label><br>
<input type="text" id="expiry_date" name="expiry_date" placeholder="MM/YY" readonly><br>
</div>
<div id="submit_button">
<input type="submit" value="Submit">
</div>
</form>
</body>
</html>
"""), Output(first_name='Ren', last_name='Farley', credit_card_number='90193990202')],
]
return data
Train the LLM on this data
After you've added data, you can now train a model. Once the training is complete, you can view the eval results.
Training on free-tier is done here on Lamini servers (but if you're on enterprise tier, this can be done on your servers) and you can track the training job's progress at this training interface. Once it's done, you can view evaluation results and play with it on a playground, on the same page.
Once a model is trained you can check the eval results to see before and after comparisons of the base model and the trained model.
You can also query the new trained model like so:
html form to use as (test) input.
html = Input(html_dom="""<!DOCTYPE html>
<html>
<head>
<title>Form Example</title>
</head>
<body>
<form>
<div>
<label for="first_name">First Name:</label><br>
<input type="text" id="first_name" name="first_name" value="John" readonly>
</div>
<div>
<label for="last_name">Last Name:</label><br>
<input type="text" id="last_name" name="last_name" value="Doe" readonly>
</div>
<div>
<label for="cc_number">Credit Card Number:</label><br>
<input type="password" id="cc_number" name="cc_number" value="**** **** **** 1234" readonly>
</div>
<div>
<input type="submit" value="Submit">
</div>
</form>
"""
Now run the html form into the finetuned model.
See the extracted output!
Output: