tnh0527 commited on
Commit
d7fc8e2
·
verified ·
1 Parent(s): ae9f62e

Introduce the models and explain the retrieval pipeline

Browse files
Files changed (1) hide show
  1. README.md +33 -10
README.md CHANGED
@@ -1,10 +1,33 @@
1
- ---
2
- title: README
3
- emoji: 📚
4
- colorFrom: yellow
5
- colorTo: gray
6
- sdk: static
7
- pinned: false
8
- ---
9
-
10
- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ title: Daecore
3
+ sdk: static
4
+ emoji: 📚
5
+ colorFrom: gray
6
+ colorTo: blue
7
+ ---
8
+
9
+ # Small models for local search and passage labeling
10
+
11
+ Daecore's models find and organize evidence in working notes and technical documentation. They run locally through ONNX Runtime, without a hosted model API, and each package can be used on its own.
12
+
13
+ | Model | Use it to | Returns | License |
14
+ |---|---|---|---|
15
+ | [EmbeddingGemma 300M v2](https://huggingface.co/Daecore/embeddinggemma-300m-memory-ft-v2) | Find passages related to a query | A normalized 768-dimensional vector per query or passage | Gemma Terms of Use |
16
+ | [Ettin 150M](https://huggingface.co/Daecore/ettin-150m-memory-reranker-ft-v1) | Put the most useful candidates first | A relevance score per query–passage pair | Apache-2.0 |
17
+ | [GLiClass Base v3](https://huggingface.co/Daecore/gliclass-std-base-v3-5facet-qint8-v2) | Label what a passage contains | Scores for trap, decision, constraint, mechanism and procedure | Apache-2.0 |
18
+
19
+ **Retrieval.** Gemma and Ettin form Daecore's retrieval pipeline. Gemma and BM25 keyword search retrieve candidates, reciprocal-rank fusion merges their rankings, and Ettin reranks up to 50 unique passages. A calibrated selector then returns 3–20 of them when enough candidates exist.
20
+
21
+ Passages are prepared before retrieval: deterministic structural chunking
22
+ follows headings and natural boundaries, carries section context, and keeps
23
+ code and table framing where possible. Token-aware splitting bounds oversized
24
+ material. Training uses frozen snapshots of these passages, including earlier
25
+ parser outputs; changing the chunker changes the evidence available to search.
26
+ Results are source passages for an agent to read and use, and several relevant
27
+ passages can repeat the same fact.
28
+
29
+ **Passage labeling.** GLiClass serves a separate Daecore function: labeling what a passage contains, so stored material can be organized by kind. It is not a retrieval component and needs neither retrieval model; its labels do not judge relevance, truth or authority.
30
+
31
+ Each card has a runnable example near the top, its training recipe and a comparison with the exact upstream checkpoint: on Daecore's evaluation data for Ettin and the classifier, and on five public retrieval datasets for Gemma.
32
+
33
+ Task-specific training and evaluation rely on mostly generated organizational documents and model-generated labels, without a human-adjudicated reference, so reported gains describe that data rather than a general improvement over the upstream models. Serving checks ran on one Windows x64 machine with an NVIDIA RTX 3060 Ti; Linux and AMD or Intel GPUs are untested.