README.md

metadata

license: apache-2.0
language:
  - en
task_categories:
  - text-generation
tags:
  - BlackTarot
  - uncensored
  - instruct
  - curated
  - synthetic-big
  - sharegpt
  - alpaca
  - instruction-tuning
  - sft
  - supervised-fine-tuning
  - chatml
  - reasoning
  - text-generation
  - synthetic-data
pretty_name: BlackTarot Curated
size_categories:
  - 1K<n<10K
viewer: true
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/sharegpt/train.parquet
      - split: test
        path: data/sharegpt/test.parquet
  - config_name: sharegpt
    data_files:
      - split: train
        path: data/sharegpt/train.parquet
      - split: test
        path: data/sharegpt/test.parquet
  - config_name: alpaca
    data_files:
      - split: train
        path: data/alpaca/train.parquet
      - split: test
        path: data/alpaca/test.parquet
  - config_name: text
    data_files:
      - split: train
        path: data/text/train.parquet
      - split: test
        path: data/text/test.parquet

BlackTarot Curated

BlackTarot Uncensored — Curated core by Grimxlock. 5,070 records (4,871 train / 199 test) from unbound-instruct canonical pipeline: 78 human-authored + 4,992 synthetic-big (0.9559 mean quality, 0.16% dedup).

• sharegpt: system + conversations[{from: human/gpt, value}] — Axolotl/LLaMA-Factory
• alpaca: instruction/input/output — Aisha shape
• text: text with <|system|>/<|user|>/<|response|> — dolphin-2 shape

from datasets import load_dataset
ds = load_dataset("Grimxlock/BlackTarot-Curated", "sharegpt") # or alpaca, text
print(ds["train"][0])
# streaming
ds = load_dataset("Grimxlock/BlackTarot-Curated", "alpaca", streaming=True)

Built via python -m unbound.build (schema validate → quality gate → MinHash dedup → deterministic split → 3 formats). See manifest.json for sha256 + stats. Parent BIG synthetic: Grimxlock/BlackTarot-Uncensored (250k).

Why Competitive with Popular

Same quality bar as BIG (0.956) but curated for fine-tuning: 5,070 hand-verified + synthetic mix, 0.16% dedup, 3 configs vs single-format popular python_code_instructions_18k (347 likes). To reach 29k downloads tier, add to HF Collections and enable trending.

Citation

@dataset{blacktarot_curated_2026,
  title={BlackTarot Curated},
  author={Grimxlock},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/Grimxlock/BlackTarot-Curated}
}