Skip to page content
SAVRN
Search Plan a project

Dataset · Text generation

ncm-datasets

by HazyResearch hazyresearch/ncm-datasets

Evaluation sets for the released NCM checkpoints.

Rows—
Configurations—
Size22.7 GB
Licensemit
AccessPublicly accessible
Monthly Downloads108

Dataset Card

By HazyResearch, published under mit, revision 6580ba41a4b0.

Evaluation sets for the released NCM checkpoints.

folder files contents
xsum/ passages.pickle, reference_summaries.pickle the complete XSum test set as tokenized article rows with summary slots, and the untruncated reference summaries as strings
wmt14/ sources.pickle, reference_translations.pickle the WMT newstest German sources and English references as strings
sudoku/{easy,medium,hard}/ puzzles.pickle, solutions.pickle held-out boards per difficulty: puzzle prefixes and completed grids as token lists
lm1b/ train_bert_30522.bin, valid_bert_30522.bin and their _offsets.npy files One Billion Word training and validation token streams with sequence offsets
tinystories/ train_bpe_8192.bin, train_bpe_8192.pickle, valid_bpe_8192.bin, valid_bpe_8192.pickle, offsets and the BPE vocabulary and merges TinyStories training and validation token streams under the custom BPE tokenizer
owt/ train_gpt2.bin, valid_gpt2.bin and their _offsets.npy files OpenWebText training and validation token streams with document offsets

Citation

@misc{kumbong2026nextconcept,
  title   = {Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text Generation},
  author  = {Kumbong, Hermann and Khare, Ishan S. and Mlauzi, Iddah and Woen, Andrew and Ding, Charles and Buchanan, E. Kelly and Arora, Simran and Mirhoseini, Azalia and R{\'e}, Christopher},
  year    = {2026},
  eprint={2610.14162},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2610.14162},
}

Details

Repository
hazyresearch/ncm-datasets
Publisher
HazyResearch
Task category
Text generation
Tags
Not stated by the source
Size category
Not stated by the source
Languages
Not stated by the source
Revision
6580ba41a4b0a976dde876fbe84d25327617b0e6
Last updated
2026-10-16

Files

30 files, 22.7 GB in total.

Documentation4 files · 2.5 KB
Other25 files · 22.7 GB
Repository1 file · 2.5 KB
Every file
FileTypeSizeSHA-256
README.mdDocumentation2.0 KB—
lm1b/README.mdDocumentation155 B—
owt/README.mdDocumentation142 B—
tinystories/README.mdDocumentation245 B—
lm1b/train_bert_30522.binOther1.8 GBd434e454d3c6
lm1b/train_bert_30522_offsets.npyOther56.7 MB454624ab4bbc
lm1b/valid_bert_30522.binOther18.4 MBacfaa554bbb7
lm1b/valid_bert_30522_offsets.npyOther574.7 KBd328d2509876
owt/train_gpt2.binOther18.1 GBb24b78a07f56
owt/train_gpt2_offsets.npyOther64.1 MB7e84564946b8
owt/valid_gpt2.binOther9.2 MB1f2ee17e0832
owt/valid_gpt2_offsets.npyOther32.1 KB09a748ce1dce
sudoku/easy/puzzles.pickleOther372.1 KB403353dd5d11
sudoku/easy/solutions.pickleOther364.1 KBb52de75f5e46
sudoku/hard/puzzles.pickleOther372.1 KBdd7b6763e559
sudoku/hard/solutions.pickleOther364.1 KBb52de75f5e46
sudoku/medium/puzzles.pickleOther372.1 KBd49cd2b380bd
sudoku/medium/solutions.pickleOther364.1 KBb52de75f5e46
tinystories/bpe_merges_8192.pickleOther108.3 KB0d4cab13f2e7
tinystories/bpe_vocab_8192.pickleOther95.1 KBfb176b4c6810
tinystories/train_bpe_8192.binOther1.1 GBb1dccdf7562f
tinystories/train_bpe_8192.pickleOther1.5 GB278847cc2cd4
tinystories/train_bpe_8192_offsets.npyOther21.7 MBc45c0471bc3f
tinystories/valid_bpe_8192.binOther11.0 MBe5c9825a84cf
tinystories/valid_bpe_8192.pickleOther15.6 MBf4723c2b0e4c
wmt14/reference_translations.pickleOther366.3 KB3dcef81d8aed
wmt14/sources.pickleOther406.1 KB62254653c32c
xsum/passages.pickleOther18.5 MB39b820f0a71d
xsum/reference_summaries.pickleOther1.5 MBdb624e065b66
.gitattributesRepository2.5 KB—

License and Download

License
mit
Access
No access gate
Download from HazyResearch

Released by HazyResearch through its official repository on Hugging Face. Read the license.