llm_chain/

traits.rs

//! # Traits Module
//!
//! Welcome to the `traits` module! This is where llm-chain houses its public traits, which define the essential behavior of steps and executors. These traits are the backbone of our library, and they provide the foundation for creating and working with different models in llm-chain.
//!
//! Here's a brief overview of the key concepts:
//! - **Steps**: These are the building blocks that make up the chains. Steps define the parameters, including the prompt that is sent to the LLM (Large Language Model).
//! - **Executors**: These are responsible for performing the steps. They take the output of a step, invoke the model with that input, and return the resulting output.
//!
//! By implementing these traits, you can set up a new model and use it in your application. Your step defines the input to the model, and your executor invokes the model and returns the output. The output of the executor is then passed to the next step in the chain, and so on.
//!

use std::{error::Error, fmt::Debug};

use crate::{
    options::Options,
    output::Output,
    prompt::Prompt,
    schema::{Document, EmptyMetadata},
    tokens::{PromptTokensError, TokenCount, Tokenizer, TokenizerError},
};
use async_trait::async_trait;

#[derive(thiserror::Error, Debug)]
#[error("unable to create executor")]
pub enum ExecutorCreationError {
    #[error("unable to create executor: {0}")]
    InnerError(#[from] Box<dyn Error + Send + Sync>),
    #[error("Field must be set: {0}")]
    FieldRequiredError(String),
}

#[derive(thiserror::Error, Debug)]
/// An error indicating that the model was not succesfully run.
pub enum ExecutorError {
    #[error("Unable to run model: {0}")]
    /// An error occuring in the underlying executor code that doesn't fit any other category.
    InnerError(#[from] Box<dyn Error + Send + Sync>),
    #[error("Invalid options when calling the executor")]
    /// An error indicating that the model was invoked with invalid options
    InvalidOptions,
    #[error(transparent)]
    /// An error tokenizing the prompt.
    PromptTokens(PromptTokensError),
    #[error("the context was to small to fit your input")]
    ContextTooSmall,
}

#[async_trait]
/// The `Executor` trait represents an executor that performs a single step in a chain. It takes a
/// step, executes it, and returns the output.
pub trait Executor: Sized {
    type StepTokenizer<'a>: Tokenizer
    where
        Self: 'a;

    /// Create a new executor with the given options. If you don't need to set any options, you can use the `new` method instead.
    /// # Parameters
    /// * `options`: The options to set.
    fn new_with_options(options: Options) -> Result<Self, ExecutorCreationError>;

    fn new() -> Result<Self, ExecutorCreationError> {
        Self::new_with_options(Options::empty().clone())
    }

    async fn execute(&self, options: &Options, prompt: &Prompt) -> Result<Output, ExecutorError>;

    /// Calculates the number of tokens used by the step given a set of parameters.
    ///
    /// The step and the parameters together are used to form full prompt, which is then tokenized
    /// and the token count is returned.
    ///
    /// # Parameters
    ///
    /// * `options`: The per-invocation options that affect the token allowance.
    /// * `prompt`: The prompt passed into step
    ///
    /// # Returns
    ///
    /// A `Result` containing the token count, or an error if there was a problem.
    fn tokens_used(
        &self,
        options: &Options,
        prompt: &Prompt,
    ) -> Result<TokenCount, PromptTokensError>;

    /// Returns the maximum number of input tokens allowed by the model used.
    ///
    /// # Parameters
    ///
    /// * `options`: The per-invocation options that affect the token allowance.
    ///
    /// # Returns
    /// The max token count for the step
    fn max_tokens_allowed(&self, options: &Options) -> i32;

    /// Returns a possible answer prefix inserted by the model, during a certain prompt mode
    ///
    /// # Parameters
    ///
    /// * `prompt`: The prompt passed into step
    ///
    /// # Returns
    ///
    /// A `Option` containing a String if  prefix exists, or none if there is no prefix
    fn answer_prefix(&self, prompt: &Prompt) -> Option<String>;

    /// Creates a tokenizer, depending on the model used by `step`.
    ///
    /// # Parameters
    ///
    /// * `step`: The step to get an associated tokenizer for.
    ///
    /// # Returns
    ///
    /// A `Result` containing a tokenizer, or an error if there was a problem.
    fn get_tokenizer(&self, options: &Options) -> Result<Self::StepTokenizer<'_>, TokenizerError>;
}

/// This marker trait is needed so the concrete VectorStore::Error can have a derived From<Embeddings::Error>
pub trait EmbeddingsError {}

#[async_trait]
pub trait Embeddings {
    type Error: Send + Debug + Error + EmbeddingsError;
    async fn embed_texts(&self, texts: Vec<String>) -> Result<Vec<Vec<f32>>, Self::Error>;
    async fn embed_query(&self, query: String) -> Result<Vec<f32>, Self::Error>;
}

/// This marker trait is needed so users of VectorStore can derive From<VectorStore::Error>
pub trait VectorStoreError {}

#[async_trait]
pub trait VectorStore<E, M = EmptyMetadata>
where
    E: Embeddings,
    M: serde::Serialize + serde::de::DeserializeOwned,
{
    type Error: Debug + Error + VectorStoreError;
    async fn add_texts(&self, texts: Vec<String>) -> Result<Vec<String>, Self::Error>;
    async fn add_documents(&self, documents: Vec<Document<M>>) -> Result<Vec<String>, Self::Error>;
    async fn similarity_search(
        &self,
        query: String,
        limit: u32,
    ) -> Result<Vec<Document<M>>, Self::Error>;
}