Tienson Qin
38e049e1c5
fix: stale embedding check ( #12049 )
...
* fix: slow stale embedding check
* fix: mark all blocks for embedding when full embedding
Also, store hnsw index for each batch so that embedding can continue
when aborted by switching graph or closing the app.
* remove embedding/re-embedding-graph-data!
* enhance: mod+c mod+s to rebuild both keyword and semantic search
2025-08-13 13:47:52 +08:00
Tienson Qin
9d594637c4
fix: don't infer hidden blocks
2025-08-05 20:27:18 +08:00
Tienson Qin
be69292346
set *publishing?
2025-08-03 16:37:45 +08:00
Tienson Qin
e8d6ae1a2f
disable qwen3 embedding
2025-07-31 22:02:13 +08:00
Tienson Qin
a9e9149350
fix: add score for non-matched results
2025-07-29 01:17:22 +08:00
Tienson Qin
d1cabafcc2
ensure db/id is not great than 2147483647 for embedding
2025-07-29 00:53:24 +08:00
Tienson Qin
74fb1194e9
adjust weights for keyword search and semantic search
2025-07-29 00:08:46 +08:00
Tienson Qin
d488c53fd6
adjust page/object sorting weight
2025-07-23 22:12:20 +08:00
Tienson Qin
6a22385f0b
replace uuid ref with block title when embedding
2025-07-23 20:54:29 +08:00
Tienson Qin
0f48207329
add tags to embedding
2025-07-23 20:48:55 +08:00
Tienson Qin
aeeaad3eb0
fix: embedding status
2025-07-23 20:12:19 +08:00
Tienson Qin
0722688fd8
feat: hybrid search
2025-07-21 21:47:36 +08:00
Tienson Qin
337b28e5d4
set batch size to 500
2025-07-21 20:53:35 +08:00
Tienson Qin
2c27aee09d
use db/id for hnsw label
...
Because db/id is unique, auto-increment so that old ids are not re-used.
2025-07-21 20:14:14 +08:00
Tienson Qin
f1d8e526fc
enhance: set embedding distance to 0.3 to keep search result related
2025-07-21 17:44:45 +08:00
Tienson Qin
bf3af0fedd
perf: skip refresh ui for embedding data transactions
2025-07-21 17:21:20 +08:00
Tienson Qin
690b92c975
enhance: adjust batch embedding size to cut down resource usage
2025-07-21 17:15:06 +08:00
Tienson Qin
2e65b639dc
fix: cancel index when switching to another model
2025-07-21 16:37:35 +08:00
Tienson Qin
b23f9e28a2
feat: add semantic search result when searching blocks
2025-07-21 13:30:45 +08:00
Tienson Qin
91d3ab70fb
enhance: don't update index info if there's no block update
2025-07-19 15:15:24 +08:00
Tienson Qin
de80615c5f
enhance: index embedding for updated blocks every 30s
2025-07-19 14:36:11 +08:00
Tienson Qin
88f62e86ec
perf improvement
2025-07-16 14:40:18 +08:00
Tienson Qin
e1147c3264
fix: c.m/run-task args
2025-07-15 15:59:01 +08:00
rcmerci
bb226a4147
feat(vec-search): debug ui show load-model-progress
2025-03-21 01:38:39 +08:00
rcmerci
a7273847bc
feat(vec-search): debug ui support selecting model
2025-03-20 22:27:56 +08:00
rcmerci
47c24177a6
feat(vec-search): update debug ui
2025-03-18 23:37:10 +08:00
rcmerci
cd44e77ac9
feat(vec-search): add debug-ui at sidebar
2025-03-18 21:55:51 +08:00
rcmerci
328b38db30
feat(embedding): add vector-search-state-flows
2025-03-18 18:37:25 +08:00
rcmerci
6b8c1dde93
feat(embedding): check webgpu available
2025-03-18 15:43:44 +08:00
rcmerci
b19f0d91ca
enhance(embedding): add user-defined class/prop
2025-03-15 23:48:26 +08:00
rcmerci
0c82a420c7
fix: use d/entity instead of d/pull
...
because :block/title in entity will replace :block/uuid by its block/title
2025-03-15 22:56:45 +08:00
rcmerci
2c88070fe0
fix: remove outdated hnsw-label
2025-03-15 22:43:28 +08:00
rcmerci
a5db52bf37
feat(embedding): text-embedding graph blocks and query hnsw
...
new properties:
:logseq.property.embedding/hnsw-label,
:logseq.property.embedding/hnsw-label-updated-at
add ns frontend.worker.embedding
2025-03-15 20:56:22 +08:00
rcmerci
d48a30000d
enhance(text-embedding): add api delete-labels
2025-03-14 17:30:25 +08:00