SpecBlock-train-data-llama

# SpecBlock-train-data-llama Training data for SpecBlock draft models, distilled from the target model [`meta-llama/Llama-3.1-8B-Instruct`](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct). ## Preparation pipeline 1. Source prompts: ShareGPT (Vicuna unfiltered) + UltraChat. 2. Re-generate each prompt's answer with the target model (greedy decoding) to obtain a clean target-aligned distribution. 3. Split into `train.jsonl` and `eval.jsonl`. Full preparation scripts: https://github.com/shiweijiezero/SpecBlock (`scripts/prepare_data.py`, `scripts/offline_generate_data.py`). ## Files - `train.jsonl` — training split (one conversation per line) - `eval.jsonl` — evaluation split ## Citation ```bibtex @misc{shi2026specblockblockiterativespeculativedecoding, title={SpecBlock: Block-Iterative Speculative Decoding with Dynamic Tree Drafting}, author={Weijie Shi and Qiang Xu and Fan Deng and Yaguang Wu and Jiarun Liu and Yehong Xu and Hao Chen and Jia Zhu and Jiajie Xu and Xiangjun Huang and Jian Yang and Xiaofang Zhou}, year={2026}, eprint={2605.07243}, archivePrefix={arXiv}, primaryClass={cs.CL}, url={https://arxiv.org/abs/2605.07243} } ```