The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
Tiny Base 300M English Pilot Tokens
This public pilot contains packed uint16 token IDs for pretraining the Tiny Base 300M, 2,048-context language model. It contains 50,000,000 training, 1,000,000 validation, and 1,000,000 test tokens. It does not redistribute raw documents.
Files and use
train.bin, validation.bin, and test.bin are little-endian uint16 streams. Use tokenizer.json with the Hugging Face tokenizers library. manifest.json records exact token counts, split policy, source revisions, mixture, and SHA-256 checksums; settings.json records preparation settings.
The deterministic split is SHA-256 of normalized text modulo 1,000: 0–9 test, 10–19 validation, and 20–999 training.
Sources and provenance
The mixture is 70% FineWeb-Edu, 25% English Wikipedia, and 5% TinyStories. Exact dataset revisions and the FineWeb configuration are pinned in manifest.json.
Upstream terms remain applicable to this derived token dataset: FineWeb-Edu is distributed under ODC-By and subject to Common Crawl terms; English Wikipedia is CC BY-SA 3.0/GFDL; TinyStories is CDLA-Sharing-1.0. Users must review and comply with those terms, including applicable attribution and share-alike obligations.
Limitations
This is a small, English-first pilot rather than a broadly representative corpus. It can carry source-data biases, inaccuracies, and copyrighted or otherwise problematic material. Do not treat it as a benchmark, safety filter, or factual authority.
- Downloads last month
- -