qql-edge 0.4.2

Embedded QQL executor backed by qdrant-edge and optional local FastEmbed inference
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
//! Trait dispatch: `Embedder` implementation over the configured slots.
//!
//! Pure move from `embedder.rs` (size hygiene split). This file holds the
//! single `impl Embedder for FastEmbedder` block — trait coherence forbids
//! splitting it further, so it is the one file that stays just over 400
//! lines; every other module in this directory is under the limit.

use async_trait::async_trait;
use qql_core::error::QqlError;
use qql_embed::{Bm25Params, Bm25TextConfig, Embedder, JointEmbeddingOutput, SparseVector};

use super::FastEmbedder;
use super::bm25::{embed_sparse_fastembed_batch, ensure_sparse_model_allowed, to_qql_sparse};
use super::err;

#[async_trait]
impl Embedder for FastEmbedder {
    fn dimension(&self) -> Option<usize> {
        Some(self.dense.dim)
    }

    fn multi_dimension(&self) -> Option<usize> {
        self.multi.as_ref().map(|m| m.dim)
    }

    fn accepts_model(&self, model: &str) -> bool {
        self.accepts_model(model)
    }

    /// Document-side BM25 hyperparameters the fallback encoder was built with.
    /// When a fastembed sparse model is configured, sparse inference is
    /// ONNX-backed and these values are not used.
    fn bm25_params(&self) -> Bm25Params {
        self.bm25_params
    }

    /// Full text configuration the fallback encoder was built with. The
    /// embedding itself runs in Qdrant's engine; this mirror keeps estimators
    /// and default-path consumers on the same language/tokenizer.
    fn bm25_text_config(&self) -> Bm25TextConfig {
        self.bm25_text.clone()
    }

    // image_dimension is not on Embedder trait; use dimension() for dense CLIP text.
    // Image dim available via FastEmbedder::image_dimension().

    async fn embed_dense(&self, text: &str, model: &str) -> Result<Vec<f32>, QqlError> {
        if !self.accepts_dense_model(model) {
            // Multi-only model id on dense path is a clear mistake.
            if self.accepts_multi_model(model) {
                return Err(err(format!(
                    "model '{model}' is the multi/ColBERT model on this edge embedder; \
                     use it with AS MULTI / multivector RERANK, not dense embedding. \
                     Dense model is '{}' ({}).",
                    self.dense.model_name, self.dense.model_code
                )));
            }
            return Err(err(format!(
                "local embedder is locked to dense model '{}' ({}); cannot satisfy USING MODEL '{model}'. \
                 Create the executor with model='{model}' (or omit MODEL to use the locked one).",
                self.dense.model_name, self.dense.model_code
            )));
        }

        let model = self.dense.model.clone();
        let texts = vec![text.to_string()];

        let mut embeddings = tokio::task::spawn_blocking(move || {
            let mut model = model
                .lock()
                .map_err(|e| err(format!("fastembed mutex poisoned: {e}")))?;
            model
                .embed(texts, None)
                .map_err(|e| err(format!("fastembed failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        embeddings
            .pop()
            .ok_or_else(|| err("fastembed returned empty result"))
    }

    async fn embed_dense_batch(
        &self,
        texts: &[String],
        model: &str,
    ) -> Result<Vec<Vec<f32>>, QqlError> {
        if texts.is_empty() {
            return Ok(vec![]);
        }
        if !self.accepts_dense_model(model) {
            if self.accepts_multi_model(model) {
                return Err(err(format!(
                    "model '{model}' is the multi/ColBERT model on this edge embedder; \
                     use it with AS MULTI / multivector RERANK, not dense embedding. \
                     Dense model is '{}' ({}).",
                    self.dense.model_name, self.dense.model_code
                )));
            }
            return Err(err(format!(
                "local embedder is locked to dense model '{}' ({}); cannot satisfy USING MODEL '{model}'. \
                 Create the executor with model='{model}' (or omit MODEL to use the locked one).",
                self.dense.model_name, self.dense.model_code
            )));
        }

        let model = self.dense.model.clone();
        let batch = texts.to_vec();

        let embeddings = tokio::task::spawn_blocking(move || {
            let mut model = model
                .lock()
                .map_err(|e| err(format!("fastembed mutex poisoned: {e}")))?;
            model
                .embed(batch, None)
                .map_err(|e| err(format!("fastembed batch failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        Ok(embeddings)
    }

    async fn embed_sparse_query(&self, text: &str, model: &str) -> Result<SparseVector, QqlError> {
        ensure_sparse_model_allowed(self, model)?;
        if self.sparse.is_some() {
            let mut out = embed_sparse_fastembed_batch(self, vec![text.to_string()]).await?;
            return out
                .pop()
                .ok_or_else(|| err("fastembed SparseTextEmbedding returned no result"));
        }
        Ok(to_qql_sparse(self.bm25.embed_query(text)))
    }

    async fn embed_sparse_document(
        &self,
        text: &str,
        model: &str,
    ) -> Result<SparseVector, QqlError> {
        ensure_sparse_model_allowed(self, model)?;
        if self.sparse.is_some() {
            let mut out = embed_sparse_fastembed_batch(self, vec![text.to_string()]).await?;
            return out
                .pop()
                .ok_or_else(|| err("fastembed SparseTextEmbedding returned no result"));
        }
        Ok(to_qql_sparse(self.bm25.embed_document(text)))
    }

    async fn embed_sparse_document_batch(
        &self,
        texts: &[String],
        model: &str,
    ) -> Result<Vec<SparseVector>, QqlError> {
        ensure_sparse_model_allowed(self, model)?;
        if self.sparse.is_some() {
            return embed_sparse_fastembed_batch(self, texts.to_vec()).await;
        }
        Ok(texts
            .iter()
            .map(|text| to_qql_sparse(self.bm25.embed_document(text)))
            .collect())
    }

    /// Single-pass BGE-M3 joint embedding: one `Bgem3Embedding::embed` call
    /// yields dense, sparse, and ColBERT together. Falls back to the default
    /// three-call implementation when no BGE-M3 multi model is configured.
    async fn embed_joint(&self, text: &str, model: &str) -> Result<JointEmbeddingOutput, QqlError> {
        let Some(ref multi) = self.multi else {
            // No BGE-M3 model: delegate to default per-call impl (non-optimal
            // but correct — no error suppression).
            let dense = self.embed_dense(text, model).await?;
            let sparse = self.embed_sparse_document(text, model).await?;
            let multi_vec = self.embed_multi(text, model).await?;
            return Ok(JointEmbeddingOutput {
                dense: Some(dense),
                sparse: Some(sparse),
                multi: Some(multi_vec),
            });
        };

        if !(self.accepts_multi_model(model)
            || model.is_empty()
            || model.eq_ignore_ascii_case("default"))
        {
            return Err(err(format!(
                "local joint embedder uses BGE-M3 '{}' ({}); cannot satisfy MODEL '{model}'",
                multi.model_name, multi.model_code
            )));
        }

        let model_arc = multi.model.clone();
        let texts = vec![text.to_string()];

        let output = tokio::task::spawn_blocking(move || {
            let mut m = model_arc
                .lock()
                .map_err(|e| err(format!("fastembed joint mutex poisoned: {e}")))?;
            m.embed(texts, None)
                .map_err(|e| err(format!("fastembed BGE-M3 joint failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        let dense = output.dense.into_iter().next();
        let sparse = output.sparse.into_iter().next().map(|e| SparseVector {
            indices: e.indices.iter().map(|&i| i as u32).collect(),
            values: e.values.clone(),
        });
        let colbert = output.colbert.into_iter().next();

        Ok(JointEmbeddingOutput {
            dense,
            sparse,
            multi: colbert,
        })
    }

    async fn embed_multi(&self, text: &str, model: &str) -> Result<Vec<Vec<f32>>, QqlError> {
        let Some(ref multi) = self.multi else {
            return Err(qql_embed::multi_unsupported_error(model));
        };
        if !self.accepts_multi_model(model) && !self.accepts_dense_model(model) {
            // Accept dense model id only when multi is configured and model is defaulted;
            // explicit wrong model still errors.
            if !model.is_empty() && !model.eq_ignore_ascii_case("default") {
                return Err(err(format!(
                    "local multi embedder is locked to '{}' ({}); cannot satisfy MODEL '{model}'",
                    multi.model_name, multi.model_code
                )));
            }
        }

        let model_arc = multi.model.clone();
        let texts = vec![text.to_string()];

        let mut output = tokio::task::spawn_blocking(move || {
            let mut model = model_arc
                .lock()
                .map_err(|e| err(format!("fastembed multi mutex poisoned: {e}")))?;
            model
                .embed(texts, None)
                .map_err(|e| err(format!("fastembed multi (BGE-M3) failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        output
            .colbert
            .pop()
            .filter(|rows| !rows.is_empty())
            .ok_or_else(|| err("fastembed multi returned empty ColBERT bag"))
    }

    async fn embed_multi_batch(
        &self,
        texts: &[String],
        model: &str,
    ) -> Result<Vec<Vec<Vec<f32>>>, QqlError> {
        if texts.is_empty() {
            return Ok(vec![]);
        }
        let Some(ref multi) = self.multi else {
            return Err(qql_embed::multi_unsupported_error(model));
        };
        if !(self.accepts_multi_model(model)
            || model.is_empty()
            || model.eq_ignore_ascii_case("default"))
        {
            return Err(err(format!(
                "local multi embedder is locked to '{}' ({}); cannot satisfy MODEL '{model}'",
                multi.model_name, multi.model_code
            )));
        }

        let model_arc = multi.model.clone();
        let batch = texts.to_vec();

        let output = tokio::task::spawn_blocking(move || {
            let mut model = model_arc
                .lock()
                .map_err(|e| err(format!("fastembed multi mutex poisoned: {e}")))?;
            model
                .embed(batch, None)
                .map_err(|e| err(format!("fastembed multi batch failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        if output.colbert.iter().any(|rows| rows.is_empty()) {
            return Err(err("fastembed multi returned an empty ColBERT bag"));
        }
        Ok(output.colbert)
    }

    async fn embed_image(&self, source: &str, model: &str) -> Result<Vec<f32>, QqlError> {
        let Some(ref image) = self.image else {
            return Err(qql_embed::image_unsupported_error(model));
        };
        if !(self.accepts_image_model(model)
            || model.is_empty()
            || model.eq_ignore_ascii_case("default"))
        {
            return Err(err(format!(
                "local image embedder is locked to '{}' ({}); cannot satisfy MODEL '{model}'",
                image.model_name, image.model_code
            )));
        }

        let model_arc = image.model.clone();
        let path = source.to_string();

        let mut embeddings = tokio::task::spawn_blocking(move || {
            let mut model = model_arc
                .lock()
                .map_err(|e| err(format!("fastembed image mutex poisoned: {e}")))?;
            model
                .embed(vec![path], None)
                .map_err(|e| err(format!("fastembed image embed failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        embeddings
            .pop()
            .ok_or_else(|| err("fastembed image returned empty result"))
    }

    async fn embed_image_batch(
        &self,
        sources: &[String],
        model: &str,
    ) -> Result<Vec<Vec<f32>>, QqlError> {
        if sources.is_empty() {
            return Ok(vec![]);
        }
        let Some(ref image) = self.image else {
            return Err(qql_embed::image_unsupported_error(model));
        };
        if !(self.accepts_image_model(model)
            || model.is_empty()
            || model.eq_ignore_ascii_case("default"))
        {
            return Err(err(format!(
                "local image embedder is locked to '{}' ({}); cannot satisfy MODEL '{model}'",
                image.model_name, image.model_code
            )));
        }

        let model_arc = image.model.clone();
        let batch = sources.to_vec();

        let embeddings = tokio::task::spawn_blocking(move || {
            let mut model = model_arc
                .lock()
                .map_err(|e| err(format!("fastembed image mutex poisoned: {e}")))?;
            model
                .embed(batch, None)
                .map_err(|e| err(format!("fastembed image batch failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        Ok(embeddings)
    }

    async fn rerank_pairs(
        &self,
        query: &str,
        documents: &[String],
        model: &str,
    ) -> Result<Vec<f32>, QqlError> {
        if documents.is_empty() {
            return Ok(vec![]);
        }
        let Some(ref reranker) = self.reranker else {
            return Err(qql_embed::cross_rerank_unsupported_error(model));
        };
        if !(self.accepts_reranker_model(model)
            || model.is_empty()
            || model.eq_ignore_ascii_case("default"))
        {
            return Err(err(format!(
                "local reranker is locked to '{}' ({}); cannot satisfy MODEL '{model}'",
                reranker.model_name, reranker.model_code
            )));
        }

        let model_arc = reranker.model.clone();
        let q = query.to_string();
        let docs = documents.to_vec();

        let ranked = tokio::task::spawn_blocking(move || {
            let mut model = model_arc
                .lock()
                .map_err(|e| err(format!("fastembed rerank mutex poisoned: {e}")))?;
            model
                .rerank(q, docs, false, None)
                .map_err(|e| err(format!("fastembed TextRerank failed: {e}")))
        })
        .await
        .map_err(|e| err(format!("spawn_blocking failed: {e}")))??;

        // Unpermute to original document order.
        let mut scores = vec![0.0f32; documents.len()];
        for item in ranked {
            if item.index < scores.len() {
                scores[item.index] = item.score;
            }
        }
        Ok(scores)
    }
}