meilisearch/milli/src/vector/openai.rs

use ordered_float::OrderedFloat;
use rayon::iter::{IntoParallelIterator, ParallelIterator as _};

use super::error::{EmbedError, NewEmbedderError};
use super::rest::{Embedder as RestEmbedder, EmbedderOptions as RestEmbedderOptions};
use super::{DistributionShift, Embeddings};
use crate::error::FaultSource;
use crate::vector::error::EmbedErrorKind;
use crate::ThreadPoolNoAbort;

#[derive(Debug, Clone, Hash, PartialEq, Eq, serde::Deserialize, serde::Serialize)]
pub struct EmbedderOptions {
    pub api_key: Option<String>,
    pub embedding_model: EmbeddingModel,
    pub dimensions: Option<usize>,
    pub distribution: Option<DistributionShift>,
}

impl EmbedderOptions {
    pub fn dimensions(&self) -> usize {
        if self.embedding_model.supports_overriding_dimensions() {
            self.dimensions.unwrap_or(self.embedding_model.default_dimensions())
        } else {
            self.embedding_model.default_dimensions()
        }
    }

    pub fn query(&self) -> serde_json::Value {
        let model = self.embedding_model.name();

        let mut query = serde_json::json!({
            "model": model,
        });

        if self.embedding_model.supports_overriding_dimensions() {
            if let Some(dimensions) = self.dimensions {
                query["dimensions"] = dimensions.into();
            }
        }

        query
    }

    pub fn distribution(&self) -> Option<DistributionShift> {
        self.distribution.or(self.embedding_model.distribution())
    }
}

#[derive(
    Debug,
    Clone,
    Copy,
    Default,
    Hash,
    PartialEq,
    Eq,
    serde::Serialize,
    serde::Deserialize,
    deserr::Deserr,
)]
#[serde(deny_unknown_fields, rename_all = "camelCase")]
#[deserr(rename_all = camelCase, deny_unknown_fields)]
pub enum EmbeddingModel {
    // # WARNING
    //
    // If ever adding a model, make sure to add it to the list of supported models below.
    #[default]
    #[serde(rename = "text-embedding-ada-002")]
    #[deserr(rename = "text-embedding-ada-002")]
    TextEmbeddingAda002,

    #[serde(rename = "text-embedding-3-small")]
    #[deserr(rename = "text-embedding-3-small")]
    TextEmbedding3Small,

    #[serde(rename = "text-embedding-3-large")]
    #[deserr(rename = "text-embedding-3-large")]
    TextEmbedding3Large,
}

impl EmbeddingModel {
    pub fn supported_models() -> &'static [&'static str] {
        &["text-embedding-ada-002", "text-embedding-3-small", "text-embedding-3-large"]
    }

    pub fn max_token(&self) -> usize {
        match self {
            EmbeddingModel::TextEmbeddingAda002 => 8191,
            EmbeddingModel::TextEmbedding3Large => 8191,
            EmbeddingModel::TextEmbedding3Small => 8191,
        }
    }

    pub fn default_dimensions(&self) -> usize {
        match self {
            EmbeddingModel::TextEmbeddingAda002 => 1536,
            EmbeddingModel::TextEmbedding3Large => 3072,
            EmbeddingModel::TextEmbedding3Small => 1536,
        }
    }

    pub fn name(&self) -> &'static str {
        match self {
            EmbeddingModel::TextEmbeddingAda002 => "text-embedding-ada-002",
            EmbeddingModel::TextEmbedding3Large => "text-embedding-3-large",
            EmbeddingModel::TextEmbedding3Small => "text-embedding-3-small",
        }
    }

    pub fn from_name(name: &str) -> Option<Self> {
        match name {
            "text-embedding-ada-002" => Some(EmbeddingModel::TextEmbeddingAda002),
            "text-embedding-3-large" => Some(EmbeddingModel::TextEmbedding3Large),
            "text-embedding-3-small" => Some(EmbeddingModel::TextEmbedding3Small),
            _ => None,
        }
    }

    fn distribution(&self) -> Option<DistributionShift> {
        match self {
            EmbeddingModel::TextEmbeddingAda002 => Some(DistributionShift {
                current_mean: OrderedFloat(0.90),
                current_sigma: OrderedFloat(0.08),
            }),
            EmbeddingModel::TextEmbedding3Large => Some(DistributionShift {
                current_mean: OrderedFloat(0.70),
                current_sigma: OrderedFloat(0.1),
            }),
            EmbeddingModel::TextEmbedding3Small => Some(DistributionShift {
                current_mean: OrderedFloat(0.75),
                current_sigma: OrderedFloat(0.1),
            }),
        }
    }

    pub fn supports_overriding_dimensions(&self) -> bool {
        match self {
            EmbeddingModel::TextEmbeddingAda002 => false,
            EmbeddingModel::TextEmbedding3Large => true,
            EmbeddingModel::TextEmbedding3Small => true,
        }
    }
}

pub const OPENAI_EMBEDDINGS_URL: &str = "https://api.openai.com/v1/embeddings";

impl EmbedderOptions {
    pub fn with_default_model(api_key: Option<String>) -> Self {
        Self { api_key, embedding_model: Default::default(), dimensions: None, distribution: None }
    }

    pub fn with_embedding_model(api_key: Option<String>, embedding_model: EmbeddingModel) -> Self {
        Self { api_key, embedding_model, dimensions: None, distribution: None }
    }
}

fn infer_api_key() -> String {
    std::env::var("MEILI_OPENAI_API_KEY")
        .or_else(|_| std::env::var("OPENAI_API_KEY"))
        .unwrap_or_default()
}

#[derive(Debug)]
pub struct Embedder {
    tokenizer: tiktoken_rs::CoreBPE,
    rest_embedder: RestEmbedder,
    options: EmbedderOptions,
}

impl Embedder {
    pub fn new(options: EmbedderOptions) -> Result<Self, NewEmbedderError> {
        let mut inferred_api_key = Default::default();
        let api_key = options.api_key.as_ref().unwrap_or_else(|| {
            inferred_api_key = infer_api_key();
            &inferred_api_key
        });

        let rest_embedder = RestEmbedder::new(RestEmbedderOptions {
            api_key: Some(api_key.clone()),
            distribution: None,
            dimensions: Some(options.dimensions()),
            url: OPENAI_EMBEDDINGS_URL.to_owned(),
            query: options.query(),
            input_field: vec!["input".to_owned()],
            input_type: crate::vector::rest::InputType::TextArray,
            path_to_embeddings: vec!["data".to_owned()],
            embedding_object: vec!["embedding".to_owned()],
        })?;

        // looking at the code it is very unclear that this can actually fail.
        let tokenizer = tiktoken_rs::cl100k_base().unwrap();

        Ok(Self { options, rest_embedder, tokenizer })
    }

    pub fn embed(&self, texts: Vec<String>) -> Result<Vec<Embeddings<f32>>, EmbedError> {
        match self.rest_embedder.embed_ref(&texts) {
            Ok(embeddings) => Ok(embeddings),
            Err(EmbedError { kind: EmbedErrorKind::RestBadRequest(error), fault: _ }) => {
                tracing::warn!(error=?error, "OpenAI: received `BAD_REQUEST`. Input was maybe too long, retrying on tokenized version. For best performance, limit the size of your document template.");
                self.try_embed_tokenized(&texts)
            }
            Err(error) => Err(error),
        }
    }

    fn try_embed_tokenized(&self, text: &[String]) -> Result<Vec<Embeddings<f32>>, EmbedError> {
        pub const OVERLAP_SIZE: usize = 200;
        let mut all_embeddings = Vec::with_capacity(text.len());
        for text in text {
            let max_token_count = self.options.embedding_model.max_token();
            let encoded = self.tokenizer.encode_ordinary(text.as_str());
            let len = encoded.len();
            if len < max_token_count {
                all_embeddings.append(&mut self.rest_embedder.embed_ref(&[text])?);
                continue;
            }

            let mut tokens = encoded.as_slice();
            let mut embeddings_for_prompt = Embeddings::new(self.dimensions());
            while tokens.len() > max_token_count {
                let window = &tokens[..max_token_count];
                let embedding = self.rest_embedder.embed_tokens(window)?;
                embeddings_for_prompt.append(embedding.into_inner()).map_err(|got| {
                    EmbedError::openai_unexpected_dimension(self.dimensions(), got.len())
                })?;

                tokens = &tokens[max_token_count - OVERLAP_SIZE..];
            }

            // end of text
            let embedding = self.rest_embedder.embed_tokens(tokens)?;

            embeddings_for_prompt.append(embedding.into_inner()).map_err(|got| {
                EmbedError::openai_unexpected_dimension(self.dimensions(), got.len())
            })?;

            all_embeddings.push(embeddings_for_prompt);
        }
        Ok(all_embeddings)
    }

    pub fn embed_chunks(
        &self,
        text_chunks: Vec<Vec<String>>,
        threads: &ThreadPoolNoAbort,
    ) -> Result<Vec<Vec<Embeddings<f32>>>, EmbedError> {
        threads
            .install(move || {
                text_chunks.into_par_iter().map(move |chunk| self.embed(chunk)).collect()
            })
            .map_err(|error| EmbedError {
                kind: EmbedErrorKind::PanicInThreadPool(error),
                fault: FaultSource::Bug,
            })?
    }

    pub fn chunk_count_hint(&self) -> usize {
        self.rest_embedder.chunk_count_hint()
    }

    pub fn prompt_count_in_chunk_hint(&self) -> usize {
        self.rest_embedder.prompt_count_in_chunk_hint()
    }

    pub fn dimensions(&self) -> usize {
        self.options.dimensions()
    }

    pub fn distribution(&self) -> Option<DistributionShift> {
        self.options.distribution()
    }
}