optivision-rag 0.2.0 → 0.4.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +14 -3
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -21,9 +21,14 @@ page image
|
|
|
21
21
|
|
|
22
22
|
| setting | encoder / corpus | compression | nDCG@5 retained |
|
|
23
23
|
|---|---|---|---|
|
|
24
|
-
| prune + binary | ColSmol-256M, generated pages | 113.5x | 86.7% |
|
|
25
|
-
| prune + int8 | ColSmol-256M, generated pages | 14.2x | 96.0% |
|
|
26
|
-
| prune + binary | ColPali-v1.3, ViDoRe (
|
|
24
|
+
| prune + binary | ColSmol-256M, 60 generated pages (72 queries) | 113.5x | 86.7% |
|
|
25
|
+
| prune + int8 | ColSmol-256M, 60 generated pages (72 queries) | 14.2x | 96.0% |
|
|
26
|
+
| prune + binary | ColPali-v1.3, 4 ViDoRe splits (1,780 pages, 1,325 queries) | 53-60x | 94.6-103.4% |
|
|
27
|
+
|
|
28
|
+
Each row is one fixed pipeline scored on every query of its corpus, with no choice made
|
|
29
|
+
from those queries. The 60-page corpus has 72 queries, so its figures carry intervals of
|
|
30
|
+
about 6 points (a sanity check, not evidence). Ratios and retention are specific to the
|
|
31
|
+
encoder and corpus named in each row.
|
|
27
32
|
|
|
28
33
|
Full results, the ablations and the analysis are in the
|
|
29
34
|
[repository](https://github.com/Daemon-VI/optivision-rag).
|
|
@@ -42,6 +47,12 @@ Full results, the ablations and the analysis are in the
|
|
|
42
47
|
> | `OPTIVISION_HOME` | where the environments are kept |
|
|
43
48
|
>
|
|
44
49
|
> Every `optivision` command below also works as `npx optivision-rag <command>`.
|
|
50
|
+
>
|
|
51
|
+
> The launcher installs the `corpus` extra by default, which does not include SciPy. The
|
|
52
|
+
> `optivision certify` command (added in 0.4.0) needs SciPy for its
|
|
53
|
+
> default candidate list, so through npx set `OPTIVISION_EXTRAS=corpus,merge` (the `merge` extra
|
|
54
|
+
> installs SciPy), or pass `--allow-reduced` to use the smaller candidate list. Each set of extras
|
|
55
|
+
> gets its own environment. Running `certify` from a pip install is the more direct route.
|
|
45
56
|
|
|
46
57
|
## Install
|
|
47
58
|
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "optivision-rag",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.4.0",
|
|
4
4
|
"description": "OptiVision RAG: extreme token compression for vision-language document retrieval (npx launcher for the Python CLI)",
|
|
5
5
|
"keywords": ["rag", "colpali", "vision-language", "retrieval", "quantization", "token-pruning", "qdrant", "cli"],
|
|
6
6
|
"homepage": "https://github.com/Daemon-VI/optivision-rag#readme",
|