Universal library for evaluating AI models
npm install autoevalsAutoevals is a tool to quickly and easily evaluate AI model outputs.
It bundles together a variety of automatic evaluation methods including:
- LLM-as-a-judge
- Heuristic (e.g. Levenshtein distance)
- Statistical (e.g. BLEU)
Autoevals is developed by the team at Braintrust.
Autoevals uses model-graded evaluation for a variety of subjective tasks including fact checking,
safety, and more. Many of these evaluations are adapted from OpenAI's excellent evals
project but are implemented so you can flexibly run them on individual examples, tweak the prompts, and debug
their outputs.
You can also create your own model-graded evaluations with Autoevals. It's easy to add custom prompts, parse outputs,
and manage exceptions.
- Python 3.9 or higher
- Compatible with both OpenAI Python SDK v0.x and v1.x
``bash`
npm install autoevals
Use Autoevals to model-grade an example LLM completion using the Factuality prompt.
By default, Autoevals uses your OPENAI_API_KEY environment variable to authenticate with OpenAI's API.
`typescript
import { Factuality } from "autoevals";
(async () => {
const input = "Which country has the highest population?";
const output = "People's Republic of China";
const expected = "China";
const result = await Factuality({ output, expected, input });
console.log(Factuality score: ${result.score});Factuality metadata: ${result.metadata?.rationale}
console.log();`
})();
When you use Autoevals, it will look for an OPENAI_BASE_URL environment variable to use as the base for requests to an OpenAI compatible API. If OPENAI_BASE_URL is not set, it will default to the AI proxy.
If you choose to use the proxy, you'll also get:
- Simplified access to many AI providers
- Reduced costs with automatic request caching
- Increased observability when you enable logging to Braintrust
The proxy is free to use, even if you don't have a Braintrust account.
If you have a Braintrust account, you can optionally set the BRAINTRUST_API_KEY environment variable instead of OPENAI_API_KEY to unlock additional features like logging and monitoring. You can also route requests to supported AI providers and models or custom models you have configured in Braintrust.
`typescript
// NOTE: ensure BRAINTRUST_API_KEY is set in your environment and OPENAI_API_KEY is not set
import { Factuality } from "autoevals";
(async () => {
const input = "Which country has the highest population?";
const output = "People's Republic of China";
const expected = "China";
// Run an LLM-based evaluator using the Claude 3.5 Sonnet model from Anthropic
const result = await Factuality({
model: "claude-3-5-sonnet-latest",
output,
expected,
input,
});
// The evaluator returns a score from [0,1] and includes the raw outputs from the evaluator
console.log(Factuality score: ${result.score});Factuality metadata: ${result.metadata?.rationale}
console.log();`
})();
There are two ways you can configure a custom client when you need to use a different OpenAI compatible API:
1. Global configuration: Initialize a client that will be used by all evaluators
2. Instance configuration: Configure a client for a specific evaluator
Set up a client that all your evaluators will use:
#### TypeScript
`typescript
import OpenAI from "openai";
import { init, Factuality } from "autoevals";
const client = new OpenAI({
baseURL: "https://api.openai.com/v1/",
});
init({ client });
(async () => {
const result = await Factuality({
input: "What is the speed of light in a vacuum?",
output: "The speed of light in a vacuum is 299,792,458 meters per second.",
expected:
"The speed of light in a vacuum is approximately 300,000 kilometers per second (or precisely 299,792,458 meters per second).",
});
console.log("Factuality Score:", result);
})();
`
Configure a client for a specific evaluator instance:
#### TypeScript
`typescript
import OpenAI from "openai";
import { Factuality } from "autoevals";
(async () => {
const customClient = new OpenAI({
baseURL: "https://custom-api.example.com/v1/",
});
const result = await Factuality({
client: customClient,
output: "Paris is the capital of France",
expected:
"Paris is the capital of France and has a population of over 2 million",
input: "Tell me about Paris",
});
console.log(result);
})();
`
Once you grade an output using Autoevals, you can optionally use Braintrust to log and compare your evaluation results. This integration is completely optional and not required for using Autoevals.
Create a file named example.eval.js (it must take the form *.eval.[ts|tsx|js|jsx]):
`typescript
import { Eval } from "braintrust";
import { Factuality } from "autoevals";
Eval("Autoevals", {
data: () => [
{
input: "Which country has the highest population?",
expected: "China",
},
],
task: () => "People's Republic of China",
scores: [Factuality],
});
`
Then, run
`bash`
npx braintrust run example.eval.js
- Battle
- Closed QA
- Humor
- Factuality
- Moderation
- Security
- Summarization
- SQL
- Translation
- Fine-tuned binary classifiers
- Context precision
- Context relevancy
- Context recall
- Context entity recall
- Faithfulness
- Answer relevancy
- Answer similarity
- Answer correctness
- Semantic list contains
- JSON validity
- Embedding similarity
- Levenshtein distance
- Exact match
- Numeric difference
- JSON diff
Autoevals supports custom evaluation prompts for model-graded evaluation. To use them, simply pass in a prompt and scoring mechanism:
`typescript
import { LLMClassifierFromTemplate } from "autoevals";
(async () => {
const promptTemplate = You are a technical project manager who helps software engineers generate better titles for their GitHub issues.
You will look at the issue description, and pick which of two titles better describes it.
I'm going to provide you with the issue description, and two possible titles.
Issue Description: {{input}}
1: {{output}}
2: {{expected}};
const choiceScores = { 1: 1, 2: 0 };
const evaluator = LLMClassifierFromTemplate<{ input: string }>({
name: "TitleQuality",
promptTemplate,
choiceScores,
useCoT: true,
});
const input = As suggested by Nicolo, we should standardize the error responses coming from GoTrue, postgres, and realtime (and any other/future APIs) so that it's better DX when writing a client,
We can make this change on the servers themselves, but since postgrest and gotrue are fully/partially external may be harder to change, it might be an option to transform the errors within the client libraries/supabase-js, could be messy?
Nicolo also dropped this as a reference: http://spec.openapis.org/oas/v3.0.3#openapi-specification;Standardize error responses from GoTrue, Postgres, and Realtime APIs for better DX
const output = ;Standardize Error Responses across APIs
const expected = ;
const response = await evaluator({ input, output, expected });
console.log("Score", response.score);
console.log("Metadata", response.metadata);
})();
`
You can also create your own scoring functions that do not use LLMs. For example, to test whether the word 'banana'
is in the output, you can use the following:
`typescript
import { Score } from "autoevals";
const bananaScorer = ({
output,
expected,
input,
}: {
output: string;
expected: string;
input: string;
}): Score => {
return { name: "banana_scorer", score: output.includes("banana") ? 1 : 0 };
};
(async () => {
const input = "What is 1 banana + 2 bananas?";
const output = "3";
const expected = "3 bananas";
const result = bananaScorer({ output, expected, input });
console.log(Banana score: ${result.score});`
})();
There is nothing particularly novel about the evaluation methods in this library. They are all well-known and well-documented. However, there are a few things that are particularly difficult when evaluating in practice:
- Normalizing metrics between 0 and 1 is tough. For example, check out the calculation in number.py to see how it's done for numeric differences.
- Parsing the outputs on model-graded evaluations is also challenging. There are frameworks that do this, but it's hard to
debug one output at a time, propagate errors, and tweak the prompts. Autoevals makes these tasks easy.
- Collecting metrics behind a uniform interface makes it easy to swap out evaluation methods and compare them. Prior to Autoevals, we couldn't find an open source library where you can simply pass in input, output, and expected values through a bunch of different evaluation methods.
The full docs are available for your reference.
We welcome contributions!
To install the development dependencies, run make develop, and run source env.sh to activate the environment. Make a .env file from the .env.example file and set the environment variables. Run direnv allow to load the environment variables.
To run the tests, run pytest` from the root directory.
Send a PR and we'll review it! We'll take care of versioning and releasing.