kenga-embed-prophet2
44M-parameter Russian/English sentence encoder from the Kenga project. Bidirectional Z-factored transformer, SentencePiece-16k tokenizer, mean pooling, 768-d L2-normalised output, 512-token context. Prefix protocol is the FRIDA / BERTA one, so it drops into any pipeline that already uses them.
kenga-embed-prophet2 starts from kenga-embed-z2 and adds a supervised stage: InfoNCE with mined hard negatives on 80k retrieval pairs (RuBQ / MIRACL-style QA, wiki title-passage), CoSENT on 20k STS pairs, plus an anchor loss that keeps the embedding close to the teacher so the distilled knowledge is not overwritten. This is the Kenga Prophet stage: learn from verified pairs, penalise confident mistakes (hard negatives).
Official MTEB(rus, v1.1) numbers
Run with mteb==2.20.5, all splits and subsets as defined by the benchmark, no task was skipped or re-weighted.
Reference columns are the models' own leaderboard submissions
(embeddings-benchmark/results). 23/23 tasks done.
| task | kenga-embed-prophet2 | Giga-Embeddings-instruct-480M | BERTA-128M | USER2-small-34M | rubert-tiny-turbo-29M |
|---|---|---|---|---|---|
| GeoreviewClassification | 43.6 | 55.4 | 54.8 | 41.1 | 41.4 |
| HeadlineClassification | 79.8 | 89.0 | 89.0 | 74.3 | 68.9 |
| InappropriatenessClassification | 58.9 | 86.1 | 74.8 | 60.7 | 59.1 |
| KinopoiskClassification | 59.6 | 73.0 | 67.8 | 52.2 | 50.5 |
| MassiveIntentClassification | 54.2 | 85.3 | 74.0 | 66.1 | 58.0 |
| MassiveScenarioClassification | 64.0 | 90.9 | 84.5 | 70.3 | 62.9 |
| RuReviewsClassification | 66.7 | 76.3 | 72.3 | 60.8 | 60.7 |
| RuSciBenchGRNTIClassification | 58.3 | 74.0 | 69.0 | 63.1 | 52.9 |
| RuSciBenchOECDClassification | 44.7 | 59.9 | 54.8 | 49.2 | 40.8 |
| CEDRClassification | 46.8 | 69.8 | 73.0 | 39.4 | 39.0 |
| SensitiveTopicsClassification | 22.6 | 44.3 | 39.9 | 27.5 | 25.2 |
| GeoreviewClusteringP2P | 41.8 | 73.8 | 73.8 | 66.2 | 59.7 |
| RuSciBenchGRNTIClusteringP2P | 55.9 | 70.5 | 65.0 | 56.4 | 48.1 |
| RuSciBenchOECDClusteringP2P | 48.6 | 58.1 | 55.6 | 48.6 | 41.1 |
| TERRa | 57.0 | 79.6 | 65.7 | 54.0 | 56.3 |
| RuBQReranking | 57.8 | 80.5 | 75.2 | 66.0 | 62.2 |
| MIRACLReranking | 36.7 | 67.5 | 64.3 | 50.5 | 47.7 |
| RiaNewsRetrievalHardNegatives.v2 | 35.9 | 88.9 | 84.5 | 74.5 | 52.3 |
| RuBQRetrieval | 36.5 | 80.6 | 71.0 | 61.1 | 51.7 |
| MIRACLRetrievalHardNegatives.v2 | 28.7 | 74.7 | 65.9 | 46.1 | 42.4 |
| RUParaPhraserSTS | 64.5 | 78.3 | 77.8 | 69.6 | 72.1 |
| RuSTSBenchmarkSTS | 70.0 | 83.6 | 82.2 | 81.0 | 78.5 |
| STS22 | 50.4 | 65.3 | 61.1 | 66.1 | 64.6 |
| --- | --- | --- | --- | --- | --- |
| Classification (mean) | 58.9 | 76.7 | 71.2 | 59.8 | 55.0 |
| MultilabelClassification (mean) | 34.7 | 57.1 | 56.5 | 33.5 | 32.1 |
| Clustering (mean) | 48.8 | 67.5 | 64.8 | 57.1 | 49.6 |
| PairClassification (mean) | 57.0 | 79.6 | 65.7 | 54.0 | 56.3 |
| Reranking (mean) | 47.3 | 74.0 | 69.7 | 58.3 | 54.9 |
| Retrieval (mean) | 33.7 | 81.4 | 73.8 | 60.6 | 48.8 |
| STS (mean) | 61.6 | 75.7 | 73.7 | 72.2 | 71.7 |
| mean over tasks | 51.4 | 74.2 | 69.4 | 58.5 | 53.7 |
| mean over task types (leaderboard) | 48.8 | 73.1 | 67.9 | 56.5 | 52.6 |
| tasks done | 23 | 23 | 23 | 23 | 23 |
Leaderboard-style mean (average of task-type means): 48.8.
What this is and is not: a 44M model, roughly 10x smaller than Giga-Embeddings-instruct-480M, meant to be compared
with the 30-40M Russian encoders (USER2-small, rubert-tiny-turbo). On the 23 finished tasks (plain mean) it scores 51.4 vs 74.2 for Giga-Embeddings-instruct-480M (+22.7 gap) and 58.5 for USER2-small-34M (-7.0). It does not beat Giga.
The numbers above are the whole story; the raw result files are in mteb_results/ of the training tree.
Usage
import sys; sys.path.insert(0, "<this folder>") # or trust_remote_code-style import after download
from modeling_kenga_embed_v2 import KengaEmbedV2HF
m = KengaEmbedV2HF.from_pretrained("<this folder>", device="cuda") # cpu works too
q = m.encode(["??? ??????? ????? ? ????"], prefix="search_query")
d = m.encode(["????? ???? ???????? ...", "?????? ?????"], prefix="search_document")
print(q @ d.T) # cosine, embeddings are L2-normalised
a = m.encode(["??? ???? ?? ??????."], prefix="paraphrase")
b = m.encode(["?? ?????? ???? ?????."], prefix="paraphrase")
Prefixes ("<prefix>: <text>" is prepended for you):
| use | prefix |
|---|---|
| retrieval query | search_query |
| retrieval document | search_document |
| STS / paraphrase (both sides) | paraphrase |
| classification / clustering | categorize, categorize_sentiment, categorize_topic |
| NLI / entailment (TERRa) | categorize_entailment |
Download with huggingface_hub.snapshot_download("GermannM/kenga-embed-prophet2"); the folder contains pytorch_model.bin,
config.json, kenga_spm.model and the self-contained modeling_kenga_embed_v2.py (torch + sentencepiece only).
Architecture
d=768, layers=8, heads=12, dff=3072, factorised token embedding (16385 x 128 -> 768), Z-factored attention/FF projections with rank 192/512, learned positions up to 512. 44.2M parameters, fp32 checkpoint 177 MB. Checkpoint step 1000.
Training code
PyTorch trainers live in the z-system lab tree (embed_v2/: build_segments.py, teacher.py, distill.py,
build_ft_data.py, mine_hard.py, finetune_prophet.py, run_mteb.py), not in the public kenga-lang repo;
the recipe and the Prophet contract are documented in
docs/PROPHETS.md. Trained on one GTX 1660 (6 GB).
License
MIT.