README.md

# ms-marco-MiniLM-L-6-v2

## Model description
This model is a fine-tuned version of ms-marco-MiniLM-L-6-v2 for relevancy evaluation in RAG scenarios.

## Training Data

The model was trained on a specialized dataset for evaluating RAG responses,
containing pairs of (context, response) with relevancy labels.
Dataset size: 4505 training examples, 5006 validation examples.


## Performance Metrics
```

Validation Metrics:
- NDCG: 0.9996 ± 0.0001
- MAP: 0.9970 ± 0.0009
- Accuracy: 0.9766 ± 0.0033

```

## Usage Example
```python

from sentence_transformers import CrossEncoder

# Load model
model = CrossEncoder('xtenzr/ms-marco-MiniLM-L-6-v2_finetuned_20241120_2220')

# Prepare inputs
texts = [
    ["Context: {...}
Query: {...}", "Response: {...}"],
]

# Get predictions
scores = model.predict(texts)  # Returns relevancy scores [0-1]

```

## Training procedure
- Fine-tuned using sentence-transformers CrossEncoder
- Trained on relevancy evaluation dataset
- Optimized for RAG response evaluation