Crate aleph_alpha_client

Usage sample

use aleph_alpha_client::{Client, TaskCompletion, How};

#[tokio::main(flavor = "current_thread")]
async fn main() {
    // Authenticate against API. Fetches token.
    let client = Client::from_env().unwrap();

    // Name of the model we want to use. Large models give usually better answer, but are also
    // more costly.
    let model = "luminous-base";

    // The task we want to perform. Here we want to continue the sentence: "An apple a day ..."
    let task = TaskCompletion::from_text("An apple a day");

    // Retrieve the answer from the API
    let response = client.completion(&task, model, &How::default()).await.unwrap();

    // Print entire sentence with completion
    println!("An apple a day{}", response.completion);
}

Structs

ChatOutput

ChatSampling

Sampling controls how the tokens (“words”) are selected for the completion. This is different from crate::Sampling, because it does not supprot the top_k parameter.

Client

Execute Jobs against the Aleph Alpha API

CompletionOutput

Completion and metainformation returned by a completion task

DetokenizationOutput

Distribution

Logprob information for a single token

Explanation

The explanation for the target.

ExplanationOutput

The result of an explanation request.

Granularity

Granularity parameters for the TaskExplanation

How

Controls of how to execute a task

ImageScore

Score for a part of an image.

Logprob

Message

Prompt

A prompt which is passed to the model for inference. Usually it is one text item, but it could also be a combination of several modalities like images and text.

Sampling

Sampling controls how the tokens (“words”) are selected for the completion.

Stopping

Controls the conditions under which the language models stops generating text.

TaskBatchSemanticEmbedding

Create embeddings for multiple prompts

TaskChat

TaskCompletion

Completes a prompt. E.g. continues a text.

TaskDetokenization

Input for a crate::Client::detokenize request.

TaskExplanation

Input for a crate::Client::explanation request.

TaskSemanticEmbedding

Create embeddings for prompts which can be used for downstream tasks. E.g. search, classifiers

TaskSemanticEmbeddingWithInstruction

Allows you to choose a semantic representation fitting for your use case.

TaskTokenization

Input for a crate::Client::tokenize request.

TextScore

Score for the part of a text-modality

TokenizationOutput

Usage

Enums

ChatEvent

CompletionEvent

Error

Errors returned by the Aleph Alpha Client

ItemExplanation

Explanation scores for a crate::prompt::Modality or the target. There is one score for each part of a modality respectively the target with the parts being choosen according to the PromptGranularity

Logprobs

Modality

The prompt for models can be a combination of different modalities (Text and Image). The type of modalities which are supported depend on the Model in question.

PromptGranularity

At which granularity should the target be explained in terms of the prompt. If you choose, for example, PromptGranularity::Sentence then we report the importance score of each sentence in the prompt towards generating the target output. The default is PromptGranularity::Auto which means we will try to find the granularity that brings you closest to around 30 explanations. For large prompts, this would likely be sentences. For short prompts this might be individual words or even tokens.

SemanticRepresentation

Allows you to choose a semantic representation fitting for your use case.

Traits

Job

A job send to the Aleph Alpha Api using the http client. A job wraps all the knowledge required for the Aleph Alpha API to specify its result. Notably it includes the model(s) the job is executed on. This allows this trait to hold in the presence of services, which use more than one model and task type to achieve their result. On the other hand a bare crate::TaskCompletion can not implement this trait directly, since its result would depend on what model is chosen to execute it. You can remedy this by turning completion task into a job, calling Task::with_model.

StreamJob

A job send to the Aleph Alpha Api using the http client. A job wraps all the knowledge required for the Aleph Alpha API to specify its result. Notably it includes the model(s) the job is executed on. This allows this trait to hold in the presence of services, which use more than one model and task type to achieve their result. On the other hand a bare crate::TaskCompletion can not implement this trait directly, since its result would depend on what model is chosen to execute it. You can remedy this by turning completion task into a job, calling [Task::with_model].

StreamTask

A task send to the Aleph Alpha Api using the http client. Requires to specify a model before it can be executed. Will return a stream of results.

Task

A task send to the Aleph Alpha Api using the http client. Requires to specify a model before it can be executed.

Functions

cosine_similarity

Intended to compare embeddings.