bge_m3_embedding_server/handler/
sparse.rs

1// Copyright (c) 2026 J. Patrick Fulton
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7//     http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15//! `POST /v1/sparse-embeddings` handler — BGE-M3 SPLADE-style sparse embeddings.
16
17use std::sync::Arc;
18use std::time::Instant;
19
20use axum::{Json, extract::State, http::HeaderMap};
21
22use super::common::{check_ready, collect_x_headers, validate_input};
23use crate::error::AppError;
24use crate::models::{SparseEmbeddingData, SparseRequest, SparseResponse, SparseValues};
25use crate::state::AppState;
26
27/// Handles `POST /v1/sparse-embeddings` — returns sparse (SPLADE-style) embeddings.
28///
29/// # Errors
30///
31/// - [`AppError::ServiceUnavailable`] if the model is not ready or no workers are live.
32/// - [`AppError::InvalidRequest`] if the batch is empty, exceeds `max_batch`, or any
33///   text exceeds the per-string character limit.
34/// - [`AppError::Internal`] if the embedding pool returns an inference error.
35///
36/// # Panics
37///
38/// Panics if the request semaphore has been closed — should not occur in normal operation.
39#[allow(clippy::cast_possible_truncation)]
40#[tracing::instrument(
41    skip(state, req, headers),
42    fields(
43        batch_size,
44        chunks,
45        max_chunk_seq,
46        tokenize_ms,
47        inference_ms,
48        queue_wait_ms,
49        total_ms,
50    )
51)]
52pub async fn sparse_embeddings(
53    State(state): State<Arc<AppState>>,
54    headers: HeaderMap,
55    Json(req): Json<SparseRequest>,
56) -> Result<Json<SparseResponse>, AppError> {
57    check_ready(&state)?;
58    let x_headers = collect_x_headers(&headers);
59    let texts = req.input.0;
60    validate_input(&texts, state.max_batch)?;
61    let batch_size = texts.len();
62    tracing::Span::current().record("batch_size", batch_size);
63
64    let t0 = Instant::now();
65
66    let _permit = Arc::clone(&state.request_permits)
67        .acquire_owned()
68        .await
69        .expect("request semaphore is never closed");
70
71    let queue_wait_ms = u64::try_from(t0.elapsed().as_millis()).unwrap_or(u64::MAX);
72
73    let (embeddings, embed_stats) = state.pool.sparse(texts).await?;
74
75    let total_ms = u64::try_from(t0.elapsed().as_millis()).unwrap_or(u64::MAX);
76    tracing::Span::current()
77        .record("chunks", embed_stats.chunks)
78        .record("max_chunk_seq", embed_stats.max_chunk_seq)
79        .record("tokenize_ms", embed_stats.tokenize_ms)
80        .record("inference_ms", embed_stats.inference_ms)
81        .record("queue_wait_ms", queue_wait_ms)
82        .record("total_ms", total_ms);
83    let x_headers_val =
84        (!x_headers.is_empty()).then(|| serde_json::to_string(&x_headers).unwrap_or_default());
85    tracing::info!(
86        route = "sparse",
87        batch_size,
88        chunks = embed_stats.chunks,
89        max_chunk_seq = embed_stats.max_chunk_seq,
90        total_token_positions = embed_stats.total_token_positions,
91        seq_len_min = embed_stats.seq_len_min,
92        seq_len_max = embed_stats.seq_len_max,
93        seq_len_mean = embed_stats.seq_len_mean,
94        seq_len_p95 = embed_stats.seq_len_p95,
95        tokenize_ms = embed_stats.tokenize_ms,
96        inference_ms = embed_stats.inference_ms,
97        queue_wait_ms,
98        total_ms,
99        x_headers = x_headers_val,
100        "embedding request complete"
101    );
102
103    let data = embeddings
104        .into_iter()
105        .enumerate()
106        .map(|(index, emb)| SparseEmbeddingData {
107            index,
108            sparse_values: SparseValues {
109                indices: emb.indices.iter().map(|i| *i as u32).collect(),
110                values: emb.values,
111            },
112        })
113        .collect();
114
115    Ok(Json(SparseResponse { data }))
116}
bge_m3_embedding_server/handler/sparse.rs

bge_m3_embedding_server/handler/
sparse.rs
