Update README.md
Browse files
README.md
CHANGED
|
@@ -36,10 +36,6 @@ The training recipe has two stages:
|
|
| 36 |
For the second stage, a frozen listwise teacher (`jinaai/jina-reranker-v3.5`) scores each
|
| 37 |
32-document candidate set jointly. GLINT distils that ordering with a sharpened listwise KL
|
| 38 |
objective, while a false-negative-masked InfoNCE term preserves a direct retrieval signal.
|
| 39 |
-
The result is still one compact, first-stage late-interaction retriever: it is neither an
|
| 40 |
-
ensemble nor a re-ranker at inference time.
|
| 41 |
-
|
| 42 |
-
This is a private research release.
|
| 43 |
|
| 44 |
## Usage
|
| 45 |
|
|
|
|
| 36 |
For the second stage, a frozen listwise teacher (`jinaai/jina-reranker-v3.5`) scores each
|
| 37 |
32-document candidate set jointly. GLINT distils that ordering with a sharpened listwise KL
|
| 38 |
objective, while a false-negative-masked InfoNCE term preserves a direct retrieval signal.
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
|
| 40 |
## Usage
|
| 41 |
|