SciEmbed-FULL
Headline model. Stage 1 DAPT + contrastive on the ~30M-pair Signal A+B pool (1 epoch). The SciRepEval number in the paper is the mean over three seeds; this repo ships the seed-123 weights.
A 149M-parameter ModernBERT-base scientific document embedder trained with citation-context sentences as the primary contrastive signal. Part of the SciEmbed release (paper under double-blind review; author info omitted).
Usage
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("anon-nlp/sciembed-full")
emb = model.encode(["citation-context supervision for scientific embeddings"],
normalize_embeddings=True)
- Context length: 512 tokens
- Pooling: mean · Output dim: 768 (Matryoshka-truncatable to 512/256/128)
- License: MIT
SciRepEval (4-category macro)
| Classif. | Regr. | Prox. | Search | Overall |
|---|---|---|---|---|
| 75.6 | 28.2 | 80.9 | 82.7 | 66.85 ± 0.38 |
Citation
See the repository README. Paper: SciEmbed: Citation-Context Supervision for Scientific Document Embeddings (under review).