searchengines-web/MeiliSearch
searchengines-web/MeiliSearch
1
0
Fork 0
mirror of https://github.com/meilisearch/MeiliSearch synced 2025-07-04 04:17:10 +02:00
Code Issues Releases Wiki Activity

Add embedder stats in batches

Browse source
This commit is contained in:
Mubelotix 2025-06-20 12:42:22 +02:00
parent fc6cc80705
commit 4cadc8113b
No known key found for this signature in database
GPG key ID: 89F391DBCC8CE7F0
26 changed files with 188 additions and 73 deletions
Download patch file Download diff file Expand all files Collapse all files

15
crates/milli/src/vector/ollama.rs
View file

@ -1,3 +1,4 @@
use std::sync::Arc;
use std::time::Instant;
use rayon::iter::{IntoParallelIterator as _, ParallelIterator as _};
@ -7,6 +8,7 @@ use super::error::{EmbedError, EmbedErrorKind, NewEmbedderError, NewEmbedderErro
use super::rest::{Embedder as RestEmbedder, EmbedderOptions as RestEmbedderOptions};
use super::{DistributionShift, EmbeddingCache, REQUEST_PARALLELISM};
use crate::error::FaultSource;
use crate::progress::EmbedderStats;
use crate::vector::Embedding;
use crate::ThreadPoolNoAbort;
@ -104,8 +106,9 @@ impl Embedder {
&self,
texts: &[S],
deadline: Option<Instant>,
embedder_stats: Option<Arc<EmbedderStats>>
) -> Result<Vec<Embedding>, EmbedError> {
match self.rest_embedder.embed_ref(texts, deadline) {
match self.rest_embedder.embed_ref(texts, deadline, embedder_stats) {
Ok(embeddings) => Ok(embeddings),
Err(EmbedError { kind: EmbedErrorKind::RestOtherStatusCode(404, error), fault: _ }) => {
Err(EmbedError::ollama_model_not_found(error))
@ -118,15 +121,16 @@ impl Embedder {
&self,
text_chunks: Vec<Vec<String>>,
threads: &ThreadPoolNoAbort,
embedder_stats: Option<Arc<EmbedderStats>>,
) -> Result<Vec<Vec<Embedding>>, EmbedError> {
// This condition helps reduce the number of active rayon jobs
// so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
if threads.active_operations() >= REQUEST_PARALLELISM {
text_chunks.into_iter().map(move |chunk| self.embed(&chunk, None)).collect()
text_chunks.into_iter().map(move |chunk| self.embed(&chunk, None, embedder_stats.clone())).collect()
} else {
threads
.install(move || {
text_chunks.into_par_iter().map(move |chunk| self.embed(&chunk, None)).collect()
text_chunks.into_par_iter().map(move |chunk| self.embed(&chunk, None, embedder_stats.clone())).collect()
})
.map_err(|error| EmbedError {
kind: EmbedErrorKind::PanicInThreadPool(error),
@ -139,13 +143,14 @@ impl Embedder {
&self,
texts: &[&str],
threads: &ThreadPoolNoAbort,
embedder_stats: Option<Arc<EmbedderStats>>
) -> Result<Vec<Vec<f32>>, EmbedError> {
// This condition helps reduce the number of active rayon jobs
// so that we avoid consuming all the LMDB rtxns and avoid stack overflows.
if threads.active_operations() >= REQUEST_PARALLELISM {
let embeddings: Result<Vec<Vec<Embedding>>, _> = texts
.chunks(self.prompt_count_in_chunk_hint())
.map(move |chunk| self.embed(chunk, None))
.map(move |chunk| self.embed(chunk, None, embedder_stats.clone()))
.collect();
let embeddings = embeddings?;
@ -155,7 +160,7 @@ impl Embedder {
.install(move || {
let embeddings: Result<Vec<Vec<Embedding>>, _> = texts
.par_chunks(self.prompt_count_in_chunk_hint())
.map(move |chunk| self.embed(chunk, None))
.map(move |chunk| self.embed(chunk, None, embedder_stats.clone()))
.collect();
let embeddings = embeddings?;