VSP Agents: Biomedical Dense Embedding Engine (A100) 🧠⚡

Built and benchmarked by VSP Agents on an NVIDIA A100-SXM4 (80GB VRAM).

This is a specialized biomedical sentence and document embedding model designed for sub-second semantic search, biomedical retrieval-augmented generation (RAG), and clinical textual similarity.

---

📊 Official MTEB Benchmark Results

Evaluated using the official MTEB (Massive Text Embedding Benchmark) test suite:

---

🚀 Quickstart & Usage

Using SentenceTransformers:

from sentence_transformers import SentenceTransformer

# Load model directly from Hugging Face
model = SentenceTransformer("vishnusureshperumbavoor/vsp-agent-biomed-embedding-a100")

sentences = [
    "GLP-1 receptor agonists reduce cardiovascular mortality in type 2 diabetes.",
    "Semaglutide demonstrates significant reduction in adverse cardiac events.",
    "Statins are first-line pharmacological therapy for hypercholesterolemia."
]

# Generate normalized dense embeddings (768 dimensions)
embeddings = model.encode(sentences, normalize_embeddings=True)

# Compute cosine similarity
similarity = model.similarity(embeddings[0], embeddings[1])
print(f"Clinical Semantic Similarity: {similarity.item():.4f}")

Using Hugging Face Transformers & PyTorch:

from transformers import AutoTokenizer, AutoModel
import torch
import torch.nn.functional as F

tokenizer = AutoTokenizer.from_pretrained("vishnusureshperumbavoor/vsp-agent-biomed-embedding-a100")
model = AutoModel.from_pretrained("vishnusureshperumbavoor/vsp-agent-biomed-embedding-a100")

inputs = tokenizer(["Cardiac troponin I elevation indicates myocardial necrosis."], return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs)
    # Mean pooling
    embeddings = outputs.last_hidden_state.mean(dim=1)
    normalized_embeddings = F.normalize(embeddings, p=2, dim=1)

---

🛠️ Model Architecture & Specifications

• Team / Author: VSP Agents (Vishnu Suresh Perumbavoor)
• Hardware: NVIDIA A100-SXM4 (80GB VRAM)
• Embedding Dimension: 768
• Max Sequence Length: 512 tokens
• Pooling Mode: Mean Pooling with L2 Unit Normalization
• Underlying Backbone: BioLinkBERT (Bidirectional Biomedical Pretrained Encoder)