---
language:
- tr
license: mit
library_name: sentence-transformers
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- semantic-search
- information-retrieval
- turkish
- matryoshka-embeddings
- variable-dimensions
- hard-negatives
- mrl
datasets:
- GoktugD/DUSUNEN-HardNegatives-50K-v1
base_model: intfloat/multilingual-e5-base
model-index:
- name: DUSUNEN Atlas 278M v1
  results:
  - task:
      type: sentence-similarity
      name: Held-out hard-negative triplet discrimination
    dataset:
      type: GoktugD/DUSUNEN-HardNegatives-50K-v1
      name: DUSUNEN hard-negative validation
      split: validation
    metrics:
    - type: accuracy
      value: 0.9475
      name: Triplet accuracy at 128 dimensions
---

# DUSUNEN Atlas 278M v1

**One Turkish retrieval model, six useful embedding sizes.** This model adapts
`intfloat/multilingual-e5-base` with Matryoshka Representation Learning on
50,000 curated Turkish hard-negative triplets. Applications can select 768,
512, 384, 256, 128, or 64 dimensions at inference time without training or
storing a second model.

The practical target is a smaller vector index: 128-dimensional float32
vectors use **83.3% less storage** than 768-dimensional vectors, while
improving held-out hard-negative triplet accuracy from 92.65% to 94.75% over
the truncated base model.

## Measured dimension sweep

All rows use the same 2,000-example held-out validation split and cosine
similarity. The base and fine-tuned model were evaluated with identical E5
query/passage prefixes.

| Dimensions | Base accuracy | Atlas accuracy | Gain | Float32 bytes/vector | Index reduction |
|---:|---:|---:|---:|---:|---:|
| 768 | 95.25% | **95.50%** | +0.25 pp | 3,072 | 0% |
| 512 | 95.30% | **95.65%** | +0.35 pp | 2,048 | 33.3% |
| 384 | 95.05% | **95.35%** | +0.30 pp | 1,536 | 50.0% |
| 256 | 94.15% | **95.00%** | +0.85 pp | 1,024 | 66.7% |
| 128 | 92.65% | **94.75%** | +2.10 pp | 512 | 83.3% |
| 64 | 88.75% | **92.50%** | +3.75 pp | 256 | 91.7% |

These are held-out in-domain triplet results, not a claim of state of the art.
Broader Turkish MTEB evaluation is planned. Machine-readable results are in
`dimension-sweep.json`.

## Usage

```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer(
    "GoktugD/DUSUNEN-Atlas-278M-v1",
    truncate_dim=128,
)

queries = ["query: Türkiye'nin başkenti neresidir?"]
passages = [
    "passage: Türkiye'nin başkenti Ankara'dır.",
    "passage: İstanbul Türkiye'nin en kalabalık şehridir.",
]

query_embeddings = model.encode(queries, normalize_embeddings=True)
passage_embeddings = model.encode(passages, normalize_embeddings=True)
scores = query_embeddings @ passage_embeddings.T
print(scores)
```

Use the `query: ` prefix for queries and `passage: ` for documents. Apply the
same `truncate_dim` to both sides. Recommended operating points:

- **128 dimensions:** best storage/quality balance for large indexes.
- **256 dimensions:** conservative production default.
- **768 dimensions:** maximum measured held-out accuracy.
- **64 dimensions:** extremely compact indexes and edge experiments.

## Training

- Base revision: `d128750597153bb5987e10b1c3493a34e5a4502a`
- Data: 50,000 training and 2,000 held-out Turkish hard-negative triplets
- Loss: cached multiple-negatives ranking loss wrapped by Matryoshka loss
- Dimensions: 768, 512, 384, 256, 128, 64
- Effective batch size: 64
- Sequence length: 256
- Epochs: 1
- Precision: bfloat16
- Seed: 3407
- Hardware: one NVIDIA GeForce RTX 3090 24 GB
- Training time: 2,039 seconds
- Peak memory allocated by the training process: 2.475 GiB

Exact hyperparameters are available in `training_config.yaml`; the recorded
environment is in `training_environment.json`.

## Limitations

- Evaluation here is limited to an in-domain held-out hard-negative split.
- Training data is Turkish retrieval data; quality on other languages may
  differ from the multilingual base model.
- Prefix-free encoding was not evaluated.
- Similarity thresholds should be calibrated for each application.

## Reproducibility and integrity

The published `model.safetensors` SHA-256 is:

`66c65d4109a646d3f8a6dc6c19b20efda4d9c5ed9cd5054386812da0d6b263dd`

Developed and released by [Göktuğ Düşünen](https://huggingface.co/GoktugD).