Prompt validation for large language models (LLMs)

Run and document prompt validation tests for a large language model (LLM) specialized in sentiment analysis for financial news.

This interactive notebook shows you how to set up the ValidMind Library, initialize the library, and use a specific prompt template for analyzing the sentiment of given sentences. Prompt validation covers the initialization of a test dataset and the creation of a foundational model using the ValidMind Library, followed by the execution of a test suite specifically designed for prompt validation. The notebook also includes example data to test the model's ability to correctly identify sentiment as positive, negative, or neutral.

About ValidMind

ValidMind is a suite of tools for managing risk, including risk associated with AI and statistical models.

You use the ValidMind Library to automate documentation and validation tests, and then use the ValidMind Platform to collaborate on documentation. Together, these products simplify risk management, facilitate compliance with regulations and institutional standards, and enhance collaboration between yourself and validators.

Before you begin

This notebook assumes you have basic familiarity with Python, including an understanding of how functions work. If you are new to Python, you can still run the notebook but we recommend further familiarizing yourself with the language.

If you encounter errors due to missing modules in your Python environment, install the modules with pip install, and then re-run the notebook. For more help, refer to Installing Python Modules.

New to ValidMind?

If you haven't already seen our documentation on the ValidMind Library, we recommend you begin by exploring the available resources in this section. There, you can learn more about documenting records such as models and running tests, as well as find code samples and our Python Library API reference.

For access to all features available in this notebook, you'll need access to a ValidMind account.

Register with ValidMind

Key concepts

record: A tool tracked in the ValidMind inventory, such as a model. Records include traditional statistical models, legacy systems, artificial intelligence/machine learning models, large language models (LLMs), agentic AI systems, and other documentable items that benefit from oversight, testing, and lifecycle management.

model: SR 26-2 (which supersedes SR 11-7) defines a model as a "complex quantitative method, system, or approach that applies statistical, economic, or financial theories to process input data into quantitative estimates." Simple arithmetic, deterministic rule-based processes, or software without statistical, economic, or financial theories underpinning their design or use are generally outside SR 26-2’s definition of a model. Within ValidMind, a model is a type of record tracked in the inventory.

documentation, model documentation: A structured and detailed document pertaining to a record, encompassing key components such as its underlying assumptions, methodologies, data sources, inputs, performance metrics, evaluations, limitations, and intended uses. Within the realm of risk management, this documentation serves to ensure transparency, adherence to regulatory requirements, and a clear understanding of potential risks associated with the record's application.

document template: Lays out the structure of documents, segmented into various sections and sub-sections, and functions as a test suite specifying the tests that should be run, and how the results should be displayed. Document templates help automate your development, validation, monitoring, and other risk management processes. Document templates are available for default ValidMind document types as well as custom document types.

documentation template: A default ValidMind document type that serves as a standardized framework for developing and documenting records, including sections designated for record details, data descriptions, test results, and performance metrics. By outlining required documentation and recommended analyses, document templates ensure consistency and completeness across documentation and help guide developers through a systematic development process while promoting comparability and traceability of development outcomes.

test: A function contained in the ValidMind Library, designed to run a specific quantitative test on the dataset or record. Test results are logged to the ValidMind Platform, where they are attached to documents. Tests are the building blocks of ValidMind, used to evaluate and document records and datasets, and can be run individually or as part of a suite defined by your templates.

test suite: A collection of tests designed to run together to automate and generate documentation end-to-end for specific use cases. (Learn more: test_suites)

metric: A subset of tests that do not have thresholds. In the context of this notebook, metrics and tests can be thought of as interchangeable concepts.

custom test: Functions that you define to evaluate your record or dataset. These functions can be registered with the ValidMind Library to be used in the ValidMind Platform.

inputs: Objects to be evaluated and documented in the ValidMind Library. They can be any of the following:

model: A single record that has been initialized in ValidMind with init_model(). Despite the naming convention, model objects can be any type of record you want to test, document, validate, or monitor with ValidMind.
dataset: A single dataset that has been initialized in ValidMind with init_dataset().
models: A list of ValidMind records - usually this is used when you want to compare multiple records in your custom tests.
datasets: A list of ValidMind datasets - usually this is used when you want to compare multiple datasets in your custom tests. (Learn more: Run tests with multiple datasets)

parameters: Additional arguments that can be passed when running a ValidMind test, used to pass additional information to a test, customize its behavior, or provide additional context.

outputs: Custom tests can return elements like tables or plots. Tables may be a list of dictionaries (each representing a row) or a pandas DataFrame. Plots may be matplotlib or plotly figures.

Setting up

Install the ValidMind Library

To install the library:

%pip install -q validmind

Initialize the ValidMind Library

Register sample model

Let's first register a sample record (model) for use with this notebook:

In a browser, log in to ValidMind.
In the left sidebar, select Inventory.
Under the RECORD TYPE drop-down, select Model and click + Register Model. (Learn more: Register records in the inventory)
Enter the model details and click Next > to continue to assignment of inventory record stakeholders.
Select your own name under the RECORD OWNER drop-down.
Click Register Model to add the model to your inventory.

Apply documentation template

Once you've registered your model, let's select a documentation template. A template predefines sections for your documentation and provides a general outline to follow, making the documentation process much easier.

In the left sidebar that appears for your model, click Documents and select Development.

If you cannot locate your Development document, make sure Development type documents are enabled for model records and create a new document. (Learn more: Manage documents)
Under TEMPLATE, select LLM-based Text Classification.
Click Use Template to apply the template.

Get your code snippet

Initialize the ValidMind Library with the code snippet unique to each record per document, ensuring your test results are uploaded to the correct record and automatically populated in the right document in the ValidMind Platform when you run the Library.

On the left sidebar that appears for your model, select Getting Started and select Development from the DOCUMENT drop-down menu.
Click Copy snippet to clipboard.
Next, load your model identifier credentials from an .env file or replace the placeholder with your own code snippet:

# Load your model identifier credentials from an `.env` file

%load_ext dotenv
%dotenv .env

# Or replace with your code snippet

import validmind as vm

vm.init(
    # api_host="...",
    # api_key="...",
    # api_secret="...",
    # model="...",
    document="documentation",
)

Preview the documentation template

Let's verify that you have connected the ValidMind Library to the ValidMind Platform and that the appropriate template is selected for your model.

You will upload documentation and test results unique to your model based on this template later on. For now, take a look at the default structure that the template provides with the vm.preview_template() function from the ValidMind library and note the empty sections:

vm.preview_template()

Get ready to run the analysis

Import the ValidMind FoundationModel and Prompt classes needed for the sentiment analysis later on:

from validmind.models import FoundationModel, Prompt

Check your access to the OpenAI API:

import os

import dotenv

dotenv.load_dotenv()

if os.getenv("OPENAI_API_KEY") is None:
    raise Exception("OPENAI_API_KEY not found")

from openai import OpenAI

model = OpenAI()


def call_model(prompt):
    return (
        model.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[
                {"role": "user", "content": prompt},
            ],
        )
        .choices[0]
        .message.content
    )

Set the prompt guidelines for the sentiment analysis:

prompt_template = """
You are an AI with expertise in sentiment analysis, particularly in the context of financial news.
Your task is to analyze the sentiment of a specific sentence provided below.
Before proceeding, take a moment to understand the context and nuances of the financial terminology used in the sentence.

Sentence to Analyze:
```
{Sentence}
```

Please respond with the sentiment of the sentence denoted by one of either 'positive', 'negative', or 'neutral'.
Please respond only with the sentiment enum value. Do not include any other text in your response.

Note: Ensure that your analysis is based on the content of the sentence and not on external information or assumptions.
""".strip()

prompt_variables = ["Sentence"]

Get your sample dataset ready for analysis

To perform the sentiment analysis for financial news we're going to load a local copy of this dataset: https://www.kaggle.com/datasets/ankurzing/sentiment-analysis-for-financial-news.

This dataset contains two columns, Sentiment and Sentence. The sentiment can be negative, neutral or positive.

import pandas as pd

df = pd.read_csv("./datasets/sentiments.csv")

df_test = df[:10].reset_index(drop=True)
df_test

Perform the prompt validation

First, use the ValidMind Library to initialize the dataset and model objects necessary for documentation. The ValidMind predict_fn function allows the model to be tested and evaluated in a standardized manner:

vm_test_ds = vm.init_dataset(
    dataset=df_test,
    input_id="test_dataset",
    text_column="Sentence",
    target_column="Sentiment",
)

vm_model = vm.init_model(
    model=FoundationModel(
        predict_fn=call_model,
        prompt=Prompt(
            template=prompt_template,
            variables=prompt_variables,
        ),
    ),
    input_id="gpt_35_model",
)

# Assign model predictions to the test dataset
vm_test_ds.assign_predictions(vm_model)

Next, use the ValidMind Library to run validation tests on the model. These tests evaluate various aspects of the prompts, including bias, clarity, conciseness, delimitation, negative instruction, and specificity.

Each test is explained in detail, highlighting its purpose, test mechanism, and the importance of the specific aspect being evaluated. The tests are graded on a scale from 1 to 10, with a predetermined threshold, and the explanations for each test include a score, threshold, and a pass/fail determination.

test_suite_results = vm.run_test_suite(
    "prompt_validation",
    inputs={
        "dataset": vm_test_ds,
        "model": vm_model,
    },
)

Here, most of the tests pass but the test for conciseness needs further attention, as it fails the threshold. This test is designed to evaluate the brevity and succinctness of prompts provided to a large language model (LLM).

The test matters, because a concise prompt strikes a balance between offering clear instructions and eliminating redundant or unnecessary information, ensuring that the LLM receives relevant input without being overwhelmed.

Next steps

You can look at the results of this test suite right in the notebook where you ran the code, as you would expect. But there is a better way — use the ValidMind Platform to work with your model documentation.

Work with your documentation

From the Inventory in the ValidMind Platform, go to the model you registered earlier. (Learn more: Working with the inventory)
Click and expand the Model Development section.

What you see is the full draft of your documentation in a more easily consumable version. From here, you can make qualitative edits to documentation, view guidelines, collaborate with validators, and submit your documentation for approval when it's ready. (Learn more: Working with documentation)

Discover more learning resources

We also offer many interactive notebooks to help you use the ValidMind Library to streamline your work:

Or, visit our documentation to learn more about ValidMind.

Upgrade ValidMind

After installing ValidMind, you’ll want to periodically make sure you are on the latest version to access any new features and other enhancements.

Retrieve the information for the currently installed version of ValidMind:

%pip show validmind

If the version returned is lower than the version indicated in our production open-source code, restart your notebook and run:

%pip install --upgrade validmind

You may need to restart your kernel after running the upgrade package for changes to be applied.