sfc-gh-goliaro/kb-nano-prefill-heavy

kb-nano prefill-heavy工作负载是一个预计算的LLM吞吐量基准测试数据集，用于kb_nano项目的bench_vllm.py测试。数据集来源于LongBench，使用meta-llama/Llama-3.1-8B-Instruct作为参考分词器。包含400个请求，每个请求的提示词上限为4096个令牌，解码上限为256个令牌。数据以原始文本形式存储，以便基准测试运行器可以使用任何模型的分词器重新分词。数据集还提供了详细的令牌长度统计信息。The kb-nano prefill-heavy workload is a precomputed LLM throughput-benchmark dataset used by kb_nanos bench_vllm.py tests. The dataset is sourced from LongBench and uses meta-llama/Llama-3.1-8B-Instruct as the reference tokenizer. It contains 400 requests, with a prompt cap of 4096 tokens and a decode cap of 256 tokens per request. The data is stored as raw text to allow the benchmark runner to re-tokenize the workload with any models tokenizer. The dataset also provides detailed token-length statistics.