r-three/tulu3-sft-random7-seed123

# Tülu-3 SFT Mixture — Random 7-Cluster Baseline

This is a **randomly clustered version** of the `allenai/tulu-3-sft-mixture` dataset, where each example is assigned uniformly at random to one of **7 clusters**.
Unlike the original category-based clustering, this split contains **no semantic or source-based logic** — it serves as a **baseline** for evaluating whether principled clustering methods outperform naive random partitioning.

A fixed random seed ensures that the split is **stable and reproducible**.

**Disclaimer**: All examples are preserved exactly as in the original Tülu-3 SFT mixture.
Only an additional `random_cluster` indicator was added, and items were partitioned by split.
No text, labels, or metadata were changed.

---

## Dataset Card

- **Source base dataset:** `allenai/tulu-3-sft-mixture`
- **This repo:** 7-way random clustering with reproducible seed
- **Format:** HuggingFace `datasets` (Arrow/Parquet); conversational examples in `messages` format

---

## Splits / Random Clusters

Each split corresponds to one of seven uniform random clusters:

- `random0`
- `random1`
- `random2`
- `random3`
- `random4`
- `random5`
- `random6`

Each example includes a `random_cluster` field in the range `0–6` corresponding to its assigned split.

---

## Sizes

Each random cluster contains approximately the same number of examples:

- `random0` 134,000 examples
- `random1` 134,000 examples
- `random2` 134,000 examples
- `random3` 134,000 examples
- `random4` 134,000 examples
- `random5` 134,000 examples
- `random6` 134,000 examples

**Total:** 939,343 examples (identical to the original Tülu-3 SFT mixture)

---

## Usage

```python
from datasets import load_dataset

base = "r-three/tulu3-sft-random7"

# Load a random cluster
ds = load_dataset(base, split="random3")

# Inspect an example
ex = ds[0]
print(ex["random_cluster"])
for turn in ex["messages"]:
print(turn["role"], ":", turn["content"])
```