mjuicem/

MM-HELIX-100K

Project Page Paper Code MM-HELIX-100K

The 42 tasks in the MM-HELIX-100K

To train models for complex reasoning, a large-scale, high-quality dataset is essential. We introduce MM-HELIX-100K, a dataset of 100,000 instruction-tuning instances with detailed, reflective reasoning paths. This dataset was created using our Step-Elicited Response Generation (SERG) pipeline, which efficiently generates high-quality… See the full description on the dataset page: https://huggingface.co/datasets/mjuicem/MM-HELIX-100K.