Datasets:

CausalNLP
/

clt_gpt2_tokenized_control

license: odc-by
task_categories:
  - text-generation
configs:
  - config_name: default
    data_files:
      - split: arb_Arab
        path: data/arb_Arab/*.parquet
      - split: cmn_Hani
        path: data/cmn_Hani/*.parquet
      - split: fra_Latn
        path: data/fra_Latn/*.parquet
      - split: deu_Latn
        path: data/deu_Latn/*.parquet
      - split: eng
        path: data/eng/*.parquet

Fresh multilingual GPT-2 CLT control data

Sequential, unshuffled control sample for CLT null experiments. For each language, complete source documents were tokenized with CausalNLP/gpt2-hf_multilingual-20 at revision 0afbb31b2db3f394270d42d6a4cb7f8fceeca3d8. The first 100,000,000 tokenizer tokens were discarded (including the complete document that crossed the threshold), after which complete documents were retained until at least 100,000,000 tokens were collected.

Data are uploaded in approximately 10,000,000-token Parquet shards. Filenames record cumulative source rows, source tokens, and retained tokens for auditability and resumption. There is no shuffling.

Split	Source	Config	Revision
arb_Arab	HuggingFaceFW/fineweb-2	arb_Arab	af9c13333eb981300149d5ca60a8e9d659b276b9
cmn_Hani	HuggingFaceFW/fineweb-2	cmn_Hani	af9c13333eb981300149d5ca60a8e9d659b276b9
fra_Latn	HuggingFaceFW/fineweb-2	fra_Latn	af9c13333eb981300149d5ca60a8e9d659b276b9
deu_Latn	HuggingFaceFW/fineweb-2	deu_Latn	af9c13333eb981300149d5ca60a8e9d659b276b9
eng	HuggingFaceFW/fineweb	default	9bb295ddab0e05d785b879661af7260fed5140fc

Schema: text: string, input_ids: list<int32>.