README.md

---
license: other
license_name: baa-proprietary
library_name: sentence-transformers
tags:
- retrieval
- embeddings
- reranker
- cross-encoder
- rag
- sentence-similarity
pipeline_tag: sentence-similarity
---

# baa.ai · Merino-Small

**One model that does both halves of RAG retrieval — bi-encoder embedding *and* cross-encoder reranking — over a single shared word-embedding table.** A 384-dimensional English model, ~55M parameters, by BAA AI (Black Sheep AI).

## Get the optimal model for *your* data

Merino-Small is a strong, cost-efficient **default**. But the best embedder + reranker is **corpus-specific** — the ideal choice depends on your documents and your notion of relevance. **baa.ai offers exclusive tooling that identifies the optimal embedding and reranking models for your specific data**, so you ship the smallest models that maximize document recovery on your corpus. For a tailored recommendation, **reach out to baa.ai**.

## What it is

A two-role retrieval model over a **shared input word-embedding matrix** (stored once). The bi-encoder embedder and a 12-layer cross-encoder reranker are built on the same `MiniLM-L6-H384-uncased` backbone, so their word-embedding table is stored a single time and injected into the reranker at load — a smaller download at **no measured quality loss**, with no retraining.

- **Embed role:** bi-encoder, 384-d, L2-normalized. Prepend `"Represent this sentence for searching relevant passages: "` to queries.
- **Rerank role:** cross-encoder, single relevance logit per (query, document) pair.
- **Router:** call `.embed(...)` or `.rerank(...)`.

## Usage

```python
from modeling_baa import BaaEmbeddingReranker   # included in this repo

m = BaaEmbeddingReranker("baa-ai/Merino-Small")
qv = m.embed(["how does a cross-encoder reranker work?"], is_query=True)[0]
dv = m.embed(["a cross-encoder scores a (query, document) pair jointly",
              "bi-encoders embed query and document separately for fast retrieval"])
ranked = m.rerank("how does a cross-encoder reranker work?",
                  ["a cross-encoder scores a (query, document) pair jointly",
                   "the mitochondria is the powerhouse of the cell"])
# -> [(doc, score), ...] sorted best-first
```

## Specs

| | |
|---|---|
| Embedding dim | 384 |
| Parameters | ~55M (embedder + reranker, shared word-embedding table) |
| Languages | English |
| Max sequence length | 512 |
| Hardware | CPU / edge / GPU |

## License & attribution

- **BAA Contributions** (shared-embedding architecture, router/loader code, packaging, weights, docs) are **proprietary to BAA AI (Black Sheep AI)** — see `LICENSE`.
- Incorporates the `MiniLM-L6-H384-uncased` backbone under the **MIT License** — see `LICENSE-minilm.txt`.

© 2026 BAA AI (Black Sheep AI) — baa.ai. Provided "as is" without warranty.

## Certification & corpus fit (2026-07)

**Position Balance (PB): 0.17** — PB measures how findable a chunk is through its *second* fact when two
facts share one embedding (second-fact / first-fact top-1 retrieval on an adversarial 1,300-chunk audit;
fleet range 0.16–0.69). Compact deployment tier: use with strictly atomic chunking (one claim per embedded chunk) and parent-document retrieval.

**4-bit quantization: certified lossless under distractor stress.** Paired contested-region robustness
(gold document injected into pools of up to 100 near-topical distractors, n=300 queries, bootstrap CIs)
is statistically indistinguishable from fp16 — an axis standard hit@k benchmarks do not measure.

**Chunking prescription:** embed one atomic claim per chunk and lead with its key entity; retrieve small,
return the parent section for context. Basis: single-vector embeddings preserve ~one independent fact per
chunk regardless of encoder family (measured across 12 encoders).