ranksaga-optimized-e5-v2

This is a fine-tuned embedding model optimized by RankSaga for information retrieval tasks. It's based on intfloat/e5-base-v2 and has been optimized using advanced fine-tuning techniques on BEIR benchmark datasets.

Model Details

Base Model

• Base Model: intfloat/e5-base-v2
• Architecture: E5 (Embeddings from bidirectional Encoder representations)
• Model Type: Sentence Transformer

Training

• Fine-tuning Method: Multiple Negatives Ranking Loss
• Training Datasets: BEIR datasets (scifact, nfcorpus, scidocs, quora)
• Epochs: 5
• Batch Size: 32
• Learning Rate: 1e-5
• Mixed Precision: FP16

Optimization Results

The model was evaluated on BEIR benchmark datasets and shows significant improvements on technical domains:

NFE Corpus (Medical Information Retrieval):

• NDCG@10: +15.25% improvement
• NDCG@100: +32.62% improvement
• MAP@100: +49.49% improvement
• Recall@100: +51.03% improvement

SciDocs (Scientific Document Retrieval):

• NDCG@10: +3.14% improvement
• NDCG@100: +11.82% improvement
• MAP@100: +7.70% improvement
• Recall@100: +20.21% improvement

Quora (General Semantic Similarity):

• Maintained high baseline performance (NDCG@10: 0.8472)

Usage

Using Sentence Transformers

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("RankSaga/ranksaga-optimized-e5-v2")

# Encode sentences
sentences = [
    "What is the capital of France?",
    "Paris is the capital of France."
]
embeddings = model.encode(sentences)

# Compute similarity
from sentence_transformers.util import cos_sim
similarity = cos_sim(embeddings[0], embeddings[1])
print(f"Similarity: {similarity.item():.4f}")

Using for Information Retrieval

from sentence_transformers import SentenceTransformer
from sentence_transformers.util import cos_sim

model = SentenceTransformer("RankSaga/ranksaga-optimized-e5-v2")

# Encode documents
documents = [
    "Machine learning is a subset of artificial intelligence.",
    "Python is a popular programming language.",
    "Deep learning uses neural networks with multiple layers."
]
doc_embeddings = model.encode(documents)

# Encode query
query = "What is machine learning?"
query_embedding = model.encode(query)

# Find most similar documents
similarities = cos_sim(query_embedding, doc_embeddings)[0]
top_result_idx = similarities.argmax().item()

print(f"Query: {query}")
print(f"Most relevant document: {documents[top_result_idx]}")
print(f"Similarity: {similarities[top_result_idx].item():.4f}")

Using with Hugging Face Transformers

from transformers import AutoTokenizer, AutoModel
import torch

tokenizer = AutoTokenizer.from_pretrained("RankSaga/ranksaga-optimized-e5-v2")
model = AutoModel.from_pretrained("RankSaga/ranksaga-optimized-e5-v2")

# Encode
inputs = tokenizer("What is machine learning?", return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)
    embeddings = outputs.last_hidden_state.mean(dim=1)

Evaluation

The model was evaluated on the BEIR benchmark suite:

Dataset	NDCG@10	NDCG@100	MAP@100	Recall@100
NFE Corpus	0.3921	0.4187	0.2373	0.4830
SciDocs	0.1767	0.2726	0.1246	0.4782
Quora	0.8472	0.8631	0.8121	0.9865
SciFact	0.5137	0.5563	0.4658	0.8684

For detailed results and comparisons, see our benchmarking blog post and GitHub repository.

Limitations

• The model performs best on technical and domain-specific content (medical, scientific)
• Performance on general tasks may be similar to or slightly lower than the base model
• Model size: ~110M parameters
• Requires sentence-transformers library for optimal usage

Training Data

The model was fine-tuned on:

• SciFact: Scientific fact-checking dataset (300 queries, 5K documents)
• NFE Corpus: Medical information retrieval (323 queries, 3.6K documents)
• SciDocs: Scientific document retrieval (1K queries, 25K documents)
• Quora: Duplicate question detection (10K queries, 523K documents)

All datasets are part of the BEIR benchmark suite.

Citation

If you use this model, please cite:

@misc{ranksaga-optimized-e5-v2,
  title={RankSaga Optimized E5-v2: Fine-tuned Embedding Model for Information Retrieval},
  author={RankSaga},
  year={2026},
  url={https://huggingface.co/RankSaga/ranksaga-optimized-e5-v2}
}

License

This model is licensed under the MIT License. See the LICENSE file for details.

Contact

For questions, issues, or commercial inquiries:

• Website: https://ranksaga.com
• Contact: Contact RankSaga
• GitHub: RankSaga/beir-benchmarking

Acknowledgments

• Base model: intfloat/e5-base-v2
• BEIR Benchmark: BEIR Paper
• Sentence Transformers: sentence-transformers