| configs: | |
| - config_name: upsample-1 | |
| data_files: upsample-1/*.parquet | |
| - config_name: upsample-2 | |
| data_files: upsample-2/*.parquet | |
| # MixtureVitae-v1 upsampled | |
| Approximately 125B tokens of synthetic data generated based on MV1-decontaminated. Synthetic dialogues / study-strategy data generated from MixtureVitae-v1 with Qwen3-8B, | |
| MIND-style ([arXiv:2410.12881](https://arxiv.org/abs/2410.12881)), Swallow and Finephrase. "Synthpile" refers to the Finephrase strategies, such as tables. | |
| ## Configs | |
| - `upsample-1` — full generation pass over MixtureVitae-v1 shards. | |
| - `upsample-2` — second generation pass. | |
| ```python | |
| from datasets import load_dataset | |
| load_dataset("ontocord/MixtureVitae-v1-upsampled", "upsample-2") # default "train" split | |
| ``` | |