---
dataset_info:
  features:
  - name: conversations
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
    - name: turn_identifier
      dtype: int64
  - name: conversation_id
    dtype: string
  - name: source
    dtype: string
  splits:
  - name: train
    num_bytes: 137076991
    num_examples: 37152
  download_size: 74078361
  dataset_size: 137076991
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

Regeneration of [shisa-ai/rewild-set](https://huggingface.co/datasets/shisa-ai/rewild-set) using DeepSeek-V3-0324

37152/180490 samples (20.58% of the original set)

~13.5h on 10 x H100 nodes