Update community links: point to qovaryx.jehorizon.com (decoder no longer on HF)
Browse files
README.md
CHANGED
|
@@ -31,16 +31,6 @@ tags:
|
|
| 31 |
|
| 32 |
# Q-RAG-50M-Sovereign — the sovereign retrieval head that punches above its weight
|
| 33 |
|
| 34 |
-
<!-- QOVARYX_BACKLINK_START -->
|
| 35 |
-
> ### Q-RAG — sovereign retrieval head
|
| 36 |
-
>
|
| 37 |
-
> Sovereign retrieval-augmented-generation specialist trained on our own
|
| 38 |
-
> base. Published for research transparency.
|
| 39 |
-
>
|
| 40 |
-
> - Read the research: https://qovaryx.jehorizon.com/research
|
| 41 |
-
> - Main site: https://qovaryx.jehorizon.com
|
| 42 |
-
<!-- QOVARYX_BACKLINK_END -->
|
| 43 |
-
|
| 44 |
> **A 50M-parameter relevance scorer that beats BGE-reranker-large (560M, 11× larger) on in-distribution refusal and ties or beats 4 of 11 tested rerankers/embeddings on out-of-distribution BEIR — at 50M params, on CPU, fully sovereign.**
|
| 45 |
|
| 46 |
## What this model does, in one sentence
|
|
@@ -228,11 +218,15 @@ All ten specialists share the `qovaryx-50m-scratch-base` and the same audit disc
|
|
| 228 |
|
| 229 |
If you run Q-RAG against a model not in our table — Cohere Rerank, Voyage Rerank, jina-reranker-v2, ColBERT, or anything else — please open a discussion on this repo with the numbers. We'll add it to the card, honestly, whichever direction the result falls. The benchmark script + holdouts are in this repo.
|
| 230 |
|
| 231 |
-
##
|
|
|
|
|
|
|
| 232 |
|
| 233 |
-
- **
|
| 234 |
-
- **
|
|
|
|
|
|
|
| 235 |
- **Ko-fi (we cover GPU bills):** https://ko-fi.com/tjarvis91
|
| 236 |
-
- **
|
| 237 |
|
| 238 |
If you find a failure mode this card doesn't cover, open a discussion or come to the Discord — that's how the next crystal corpus gets written.
|
|
|
|
| 31 |
|
| 32 |
# Q-RAG-50M-Sovereign — the sovereign retrieval head that punches above its weight
|
| 33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
> **A 50M-parameter relevance scorer that beats BGE-reranker-large (560M, 11× larger) on in-distribution refusal and ties or beats 4 of 11 tested rerankers/embeddings on out-of-distribution BEIR — at 50M params, on CPU, fully sovereign.**
|
| 35 |
|
| 36 |
## What this model does, in one sentence
|
|
|
|
| 218 |
|
| 219 |
If you run Q-RAG against a model not in our table — Cohere Rerank, Voyage Rerank, jina-reranker-v2, ColBERT, or anything else — please open a discussion on this repo with the numbers. We'll add it to the card, honestly, whichever direction the result falls. The benchmark script + holdouts are in this repo.
|
| 220 |
|
| 221 |
+
## Official site & community
|
| 222 |
+
|
| 223 |
+
The full Qovaryx runtime that orchestrates this specialist alongside the other nine ships from:
|
| 224 |
|
| 225 |
+
- **Site:** https://qovaryx.jehorizon.com
|
| 226 |
+
- **Download (desktop beta):** https://qovaryx.jehorizon.com/download.html
|
| 227 |
+
- **Research devlog:** https://qovaryx.jehorizon.com/research
|
| 228 |
+
- **Community Discord:** https://discord.gg/PtuHZDv5ju
|
| 229 |
- **Ko-fi (we cover GPU bills):** https://ko-fi.com/tjarvis91
|
| 230 |
+
- **Open research repo:** https://github.com/thron-j/qovaryx-ai-research
|
| 231 |
|
| 232 |
If you find a failure mode this card doesn't cover, open a discussion or come to the Discord — that's how the next crystal corpus gets written.
|