stsb-roberta-base

Model Overview

This model is a cross-encoder fine-tuned for semantic textual similarity (STS) on the STS benchmark dataset. It accepts a pair of sentences and outputs a similarity score between 0 and 1. The underlying architecture is RoBERTa-base, making it well-suited for re-ranking tasks where fine-grained relevance judgments are needed.

Key Strengths

• Directly learns pairwise similarity, yielding higher accuracy than bi-encoder approaches for re-ranking.
• Optimized for the STS benchmark, a standard evaluation for sentence similarity.
• Outputs a continuous score (0–1) suitable for downstream ranking or thresholding.

Usage Through gigarouter

Gigarouter hosts this model as a managed, OpenAI-compatible API. You send sentence pairs and receive similarity scores — no need to manage transformers or inference infrastructure.

Intended Application

• Re-ranking results from an initial retrieval step (e.g., in search or question-answering).
• Any task requiring accurate, pairwise semantic similarity measurement.