---
dataset_info:
  features:
  - name: sentence1
    dtype: string
  - name: sentence2
    dtype: string
  - name: score
    dtype: float32
  splits:
  - name: train
    num_examples: 6948
  - name: validation
    num_examples: 1985
  - name: test
    num_examples: 994
  task_categories:
  - text-regression
  task_ids:
  - semantic-textual-similarity
---

# Samsoup/sickr-sts

This dataset is derived from **`mteb/sickr-sts`** (SICK-R style semantic textual similarity),
which in MTEB is provided as a single split. This script shuffles that split deterministically
and produces **train / validation / test = 70% / 20% / 10%**.

**Fields**
- `sentence1` — first sentence
- `sentence2` — second sentence
- `score` — similarity / relatedness score (float32)

**Processing**
- Input: single split from `mteb/sickr-sts`
- Shuffle with a fixed seed
- 70/20/10 partition
- Keep only (`sentence1`, `sentence2`, `score`)

**Notes**
- This is meant for *regression* STS training (like STS-B style models) but on the SICK-R data.
- Reproducible: split is deterministic.

_Last updated: 2025-10-31_