metadata
dataset_info:
features:
conversations:
type: list
feature:
type: object
fields:
content: string
role: string
source: string
splits:
- name: train
num_bytes: 110
num_examples: 2
download_size: 1448
dataset_size: 110
configs:
- config_name: default
data_files:
- split: train
path: interleaved_output-chunk-*.jsonl
task_categories:
- text-generation
size_categories:
- 1M<n<10M
Turngen
A large scale turnwise conversational dataset, with a focus on roleplay data
Data breakdown
Datasets included:
- PygmalionAI/PIPPA license: apache-2.0
- 873380 samples from HuggingfaceTB/smoltalk see repo for licensing information; different subsets have different licenses
- 2,380 samples from everyday-conversations
- 50,000 samples from metamathqa-50k
- 112,000 samples from numina-cot-100k
- 50,000 samples from self-oss-instruct
- 36,000 samples from smol-constraints
- 431,000 samples from smol-magpie-ultra
- 56,000 samples from smol-rewrite
- 101,000 samples from smol-summarize
- 35,000 samples from systemchats-30k
- openerotica/mixed-rp license: apache-2.0
- openerotica/long-roleplay-v0.1 license: apache-2.0
- anthracite-org/stheno-filtered-v1.1 license: apache-2.0
- Open-Orca/SlimOrca license: MIT
- PJMixers/hieunguyenminh_roleplay-deduped-ShareGPT license: unknown, derived from a cc-by-4.0 dataset (hieunguyenminh/roleplay)
- nvidia/Daring-Anteater license: cc-by-4.0
- WizardLMTeam/WizardLM_evol_instruct_V2_196k license: MIT
- Open-Orca/SlimOrca license: MIT
- CollectiveCognition/chats-data-2023-09-22 license: MIT
- NewEden/Roleplay-Logs-V2 license: MIT
- Nopm/Opus_WritingStruct license: apache-2.0
- allenai/WildChat-1M license: odc-by
- allenai/tulu-3-sft-personas-instruction-following license: odc-by
- HuggingFaceH4/ultrachat_200k license: MIT
- both train_sft and train_gen splits
- 250,000 samples from soda-chatml license: unknown, derived from a cc-by-4.0 dataset (allenai/soda)
- AlekseyKorshuk/oasst1-chatml license: unknown, derived from an apache-2.0 dataset (OpenAssistant/oasst1)
- 500,00 samples from nvidia/OpenMathInstruct-2 license: cc-by-4.0
- glaiveai/glaive-code-assistant license: apache-2.0
- theblackcat102/alexa-qa license: MIT
- dinalt/roleplay_build license: cc-by-4.0
If a quantity of samples is not specified, approximately all of the dataset is included Counts may not be exact
Notes
- Some turns contain no text
- Some roleplays have a system prompt with no content (None) which may cause code using the dataset to crash if not accounted for
- These conversations are not strictly system/user/assistant, some put character name as role