6 results

domain-aware

inall models datasets apps agents people projects

Datasets

All datasets matching “domain-aware”

MikhailRudenko /domain-aware-sd-synthetic Domain-Aware Speculative Decoding — Synthetic Dataset Synthetic distillation data for training domain-specific draft models in speculative decoding. Overview Generated by running TurboSparse-Mistral-Instruct (7B) in teacher-forcing mode on Flan tasks and capturing top-10 token distributions at each position. Structure train/v3/ — 66 NPZ files (2.2 GB), one per Flan cluster validation/v3/ — 66 JSONL files (93 MB) test/v3/ — 66 NPZ files (598… See the full description on the dataset page: https://huggingface.co/datasets/MikhailRudenko/domain-aware-sd-synthetic.text-generation1M<n<10M0 likes283 downloads28d agoHugging Face