metadata
license: apache-2.0
This dataset contains dataset splits for evaluating speculative decoding algorithms on different tasks.
File:
- Coding: HumanEval.jsonl
- Math: math_reasoning.jsonl
- Question Answering: qa.jsonl
- MT_bench: question.jsonl
- Retrieval-Augmented Generation: rag.jsonl
- Summarization: summarization.jsonl
- Translation (German to English): translation.jsonl
- Writing: writing.jsonl
The data comes from two sources:
- https://github.com/openai/human-eval (1). (The MIT License)
- https://github.com/hemingkx/Spec-Bench (2-8) (Apache 2.0)