--- language: - el - en license: other license_name: openmdw-1.1 base_model: nvidia/Nemotron-3-Embed-1B-BF16 pipeline_tag: sentence-similarity library_name: sentence-transformers tags: - greek - english - retrieval - rag - embeddings - sentence-transformers - nemotron model-index: - name: Sophea-Nemo-Embedding results: - task: type: sentence-similarity name: Greek+English retrieval (5 domains, cosine IR) dataset: type: retrieval name: Internal Greek+English retrieval benchmark metrics: - type: nDCG@10 name: nDCG@10 (mean of 5 domains) value: 0.8176 - type: recall@10 name: Recall@10 (mean of 5 domains) value: 0.9310 ---

Sophea-Nemo-Embedding

Greek + English dense retriever — fine-tuned NVIDIA Nemotron-3-Embed-1B

**Sophea-Nemo-Embedding** is a Greek + English **dense retrieval embedder**, fine-tuned from **[NVIDIA Nemotron-3-Embed-1B](https://huggingface.co/nvidia/Nemotron-3-Embed-1B-BF16)** for Greek and English retrieval. It is the first stage of a two-stage Greek RAG stack; the matching second stage is **[Sophea-Nemo-Reranker](https://huggingface.co/KIEFERSA/Sophea-Nemo-Reranker)**. - **Creator:** Kiefer SA - **Base model:** `nvidia/Nemotron-3-Embed-1B-BF16` (Ministral-3 backbone, **bidirectional**, mean pooling with `include_prompt=true`, 2048-d Matryoshka-sliceable, L2-normalized) - **Languages:** Greek + English - **Prompts:** `query: ` on queries, `passage: ` on documents (both sides prompted) - **Training:** full fine-tune, InfoNCE (in-batch + 7 hard negatives), 1 epoch, max-len 4096, bf16 > **Greek is out-of-distribution for the base.** Nemotron-3-Embed lists 34 languages and **Greek is > not one of them** — which is exactly why the base retrieves poorly on Greek and why this fine-tune > exists. ## Why it matters — base vs fine-tuned On our held-out 5-domain Greek+English retrieval eval (cosine IR, `query: `/`passage: ` prompts), the fine-tune roughly **2.5×'s** the base embedder's retrieval quality — on **every** domain. ![Greek+English retrieval — base vs fine-tuned](base_vs_ft.png)

metric	Val	Energy	Legal	Finance	Medical	Mean
nDCG@10 — base	0.382	0.345	0.526	0.194	0.209	0.331
nDCG@10 — fine-tuned	0.875	0.794	0.950	0.722	0.747	0.818
Δ nDCG@10	+0.493	+0.448	+0.424	+0.528	+0.538	+0.486
Recall@10 — base	0.465	0.464	0.599	0.283	0.314	0.425
Recall@10 — fine-tuned	0.963	0.898	0.984	0.903	0.906	0.931

**Read.** Mean **nDCG@10 0.331 → 0.818 (+0.486)**, with the largest gains on the hardest domains (finance +0.53, medical +0.54). Recall@10 climbs from a mean 0.425 to **0.931** — i.e. the right passage is now almost always in the top-10. ## Size comparison — vs base Qwen3-Embedding (off-the-shelf) Same 5-domain retrieval eval, same corpus/prompts. Your **fine-tuned 1B** embedder beats the **base Qwen3-Embedding-8B by a wide margin** (nDCG@10):

model	Val	Energy	Legal	Finance	Medical	Mean
Sophea-Nemo-Embedding (ours · 1B · fine-tuned)	0.875	0.794	0.950	0.722	0.747	0.818
Qwen3-Embedding-8B (base, off-the-shelf)	0.740	0.580	0.910	0.490	0.602	0.664
Qwen3-Embedding-4B (base, off-the-shelf)	0.738	0.579	0.919	0.501	0.586	0.665
Qwen3-Embedding-0.6B (base, off-the-shelf)	0.658	0.509	0.863	0.426	0.467	0.585

**Read.** Mean nDCG@10 **0.818** (ours, 1B) vs **0.664** (8B), **0.665** (4B), **0.585** (0.6B). Note the Qwen 8B ≈ 4B — scaling the off-the-shelf model barely helps on Greek, whereas a 1B **fine-tuned** on-domain wins by **+0.153** over the 8×-larger model. ## End-to-end two-stage pipeline The two models are designed to run **together**: the embedder retrieves the top-50 candidates, the reranker re-orders them. Measured end-to-end on the same 5-domain Greek+English eval (150 queries/set, correct `query:`/`passage:` prompts): ![End-to-end two-stage stack — embedder alone vs + reranker](stack.png)

metric	Val	Energy	Legal	Finance	Medical	Mean
nDCG@10 — embedder alone	0.863	0.802	0.939	0.702	0.689	0.799
nDCG@10 — + reranker (stack)	0.867	0.780	0.976	0.763	0.703	0.818
Δ from reranking	+0.004	-0.022	+0.037	+0.061	+0.013	+0.019
Recall@10 — full stack	0.987	0.873	0.993	0.920	0.860	0.927

**Read.** The full stack reaches **mean nDCG@10 0.818** (Recall@10 0.927) — the strongest configuration measured. The reranker adds the most where the embedder is weakest (**finance +0.061, legal +0.037**), i.e. it sharpens precision on the hard cases. On **energy** it slightly regresses (-0.022) and on `val` the gain is marginal, because the embedder alone is already very strong there and leaves little headroom — for the easiest, already-high-recall domains you can skip the rerank stage if latency matters. ## Usage (sentence-transformers) ```python from sentence_transformers import SentenceTransformer m = SentenceTransformer("KIEFERSA/Sophea-Nemo-Embedding") q = m.encode(["Ποια είναι η πρωτεύουσα της Ελλάδας;"], prompt="query: ", normalize_embeddings=True) d = m.encode(["Η Αθήνα είναι η πρωτεύουσα της Ελλάδας."], prompt="passage: ", normalize_embeddings=True) print((q @ d.T)) # cosine similarity ``` Both sides **must** be prompted (`query: ` / `passage: `) — the base uses `include_prompt=true`, so dropping the prefix is a distribution mismatch. Embeddings are L2-normalized (use cosine / dot). ## Serving with vLLM ```bash vllm serve KIEFERSA/Sophea-Nemo-Embedding --runner pooling ``` Exposes an OpenAI-compatible `/v1/embeddings` endpoint. vLLM does **not** auto-apply the model's ST prompts, so prepend them **client-side** — `query: ` on queries, `passage: ` on documents: ```python from openai import OpenAI c = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY") name = "KIEFERSA/Sophea-Nemo-Embedding" q = c.embeddings.create(model=name, input=["query: Ποια είναι η πρωτεύουσα της Ελλάδας;"]).data[0].embedding d = c.embeddings.create(model=name, input=["passage: Η Αθήνα είναι η πρωτεύουσα της Ελλάδας."]).data[0].embedding ``` ## License Fine-tune of `nvidia/Nemotron-3-Embed-1B-BF16` — governed by **OpenMDW-1.1** (base Ministral-3 is Apache-2.0); commercial use permitted under those terms. Review the base model's license before use.