searchengines-web/MeiliSearch
searchengines-web/MeiliSearch
1
0
Fork 0
mirror of https://github.com/meilisearch/MeiliSearch synced 2024-11-09 22:48:54 +01:00
Code Issues Releases Wiki Activity

ollama and openai use new EmbedderOptions

Browse Source
This commit is contained in:
Louis Dureuil 2024-07-16 15:17:49 +02:00
parent a1beddd5d9
commit d731fa661b
No known key found for this signature in database
2 changed files with 40 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
milli/src/vector/ollama.rs
View File

@ -28,19 +28,22 @@ impl EmbedderOptions {
impl Embedder { impl Embedder {
pub fn new(options: EmbedderOptions) -> Result<Self, NewEmbedderError> { pub fn new(options: EmbedderOptions) -> Result<Self, NewEmbedderError> {
let model = options.embedding_model.as_str(); let model = options.embedding_model.as_str();
let rest_embedder = match RestEmbedder::new(RestEmbedderOptions { let rest_embedder = match RestEmbedder::new(
api_key: options.api_key, RestEmbedderOptions {
dimensions: None, api_key: options.api_key,
distribution: options.distribution, dimensions: None,
url: options.url.unwrap_or_else(get_ollama_path), distribution: options.distribution,
query: serde_json::json!({ url: options.url.unwrap_or_else(get_ollama_path),
"model": model, request: serde_json::json!({
}), "model": model,
input_field: vec!["prompt".to_owned()], "prompt": super::rest::REQUEST_PLACEHOLDER,
path_to_embeddings: Default::default(), }),
embedding_object: vec!["embedding".to_owned()], response: serde_json::json!({
input_type: super::rest::InputType::Text, "embedding": super::rest::RESPONSE_PLACEHOLDER,
}) { }),
},
super::rest::ConfigurationSource::Ollama,
) {
Ok(embedder) => embedder, Ok(embedder) => embedder,
Err(NewEmbedderError { Err(NewEmbedderError {
kind: kind:

41
milli/src/vector/openai.rs
View File

@ -26,20 +26,21 @@ impl EmbedderOptions {
} }
} }
pub fn query(&self) -> serde_json::Value { pub fn request(&self) -> serde_json::Value {
let model = self.embedding_model.name(); let model = self.embedding_model.name();
let mut query = serde_json::json!({ let mut request = serde_json::json!({
"model": model, "model": model,
"input": [super::rest::REQUEST_PLACEHOLDER, super::rest::REPEAT_PLACEHOLDER]
}); });
if self.embedding_model.supports_overriding_dimensions() { if self.embedding_model.supports_overriding_dimensions() {
if let Some(dimensions) = self.dimensions { if let Some(dimensions) = self.dimensions {
query["dimensions"] = dimensions.into(); request["dimensions"] = dimensions.into();
} }
} }
query request
} }
pub fn distribution(&self) -> Option<DistributionShift> { pub fn distribution(&self) -> Option<DistributionShift> {
@ -180,17 +181,23 @@ impl Embedder {
let url = options.url.as_deref().unwrap_or(OPENAI_EMBEDDINGS_URL).to_owned(); let url = options.url.as_deref().unwrap_or(OPENAI_EMBEDDINGS_URL).to_owned();
let rest_embedder = RestEmbedder::new(RestEmbedderOptions { let rest_embedder = RestEmbedder::new(
api_key: Some(api_key.clone()), RestEmbedderOptions {
distribution: None, api_key: Some(api_key.clone()),
dimensions: Some(options.dimensions()), distribution: None,
url, dimensions: Some(options.dimensions()),
query: options.query(), url,
input_field: vec!["input".to_owned()], request: options.request(),
input_type: crate::vector::rest::InputType::TextArray, response: serde_json::json!({
path_to_embeddings: vec!["data".to_owned()], "data": [{
embedding_object: vec!["embedding".to_owned()], "embedding": super::rest::RESPONSE_PLACEHOLDER
})?; },
super::rest::REPEAT_PLACEHOLDER
]
}),
},
super::rest::ConfigurationSource::OpenAi,
)?;
// looking at the code it is very unclear that this can actually fail. // looking at the code it is very unclear that this can actually fail.
let tokenizer = tiktoken_rs::cl100k_base().unwrap(); let tokenizer = tiktoken_rs::cl100k_base().unwrap();
@ -201,7 +208,7 @@ impl Embedder {
pub fn embed(&self, texts: Vec<String>) -> Result<Vec<Embeddings<f32>>, EmbedError> { pub fn embed(&self, texts: Vec<String>) -> Result<Vec<Embeddings<f32>>, EmbedError> {
match self.rest_embedder.embed_ref(&texts) { match self.rest_embedder.embed_ref(&texts) {
Ok(embeddings) => Ok(embeddings), Ok(embeddings) => Ok(embeddings),
Err(EmbedError { kind: EmbedErrorKind::RestBadRequest(error), fault: _ }) => { 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."); 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) self.try_embed_tokenized(&texts)
} }
@ -225,7 +232,7 @@ impl Embedder {
let embedding = self.rest_embedder.embed_tokens(tokens)?; let embedding = self.rest_embedder.embed_tokens(tokens)?;
embeddings_for_prompt.append(embedding.into_inner()).map_err(|got| { embeddings_for_prompt.append(embedding.into_inner()).map_err(|got| {
EmbedError::openai_unexpected_dimension(self.dimensions(), got.len()) EmbedError::rest_unexpected_dimension(self.dimensions(), got.len())
})?; })?;
all_embeddings.push(embeddings_for_prompt); all_embeddings.push(embeddings_for_prompt);