roihu_gen
收藏官方服务：
资源简介：
# 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 ```
提供机构：
maas创建时间：
2026-05-12
