Dataset Viewer
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.
DragonCode Coding-Tokens — Pre-tokenized Code Corpus
Pre-tokenized code-domain training corpus for the DragonCode LLM family.
🟢 LIVE STATUS — READ FIRST
Last updated: 2026-08-29 ~09:45 HKT (Sat)
- ✅ COMPLETE / STABLE — 58 shards finalised (
data/train-00000..00057),dataset_info.jsonpresent (num_shards: 58,total_tokens: 7,740,201,908). - Used by the 150M and 387M tiers for their streaming code pretrain (both now done).
- 787M will use this repo's 58 code shards plus the in-progress finance corpus (
DragonCode-Finance-Tokens) for its mixed code+finance pretrain.
No action needed here — the corpus is finalised and read-only.
Format
- Shards:
data/train-*.bin, flatint32little-endian token-id stream. - Tokenizer:
bigcode/starcoder2-3b(vocab 49152, shared family tokenizer;add_special_tokens=False). - Corpus recorded in
dataset_info.json—num_shards: 58,total_tokens: 7,740,201,908.
Sources (streaming, permissive only)
| Source | Field | License filter |
|---|---|---|
codeparrot/codeparrot-clean |
content |
permissive only (MIT/ISC/Apache-2.0/BSD/CC0/Unlicense) |
Non-permissive / empty samples dropped. General web corpora excluded per project spec.
Consumption (for pretrain)
Read by scripts/dragoncode_pretrain.py → _download_tokenized → merge to a flat
mmap, sequence length 4096, single forward pass. Resume-from-HF only.
- Downloads last month
- 141