tddg/tensorsketch-qwen35

该数据集包含了10个Qwen3.5和Qwen3模型（参数≤9B）中每个张量的6,063个Count Sketch指纹（TensorSketch），按HuggingFace下载量排序。这些指纹是对张量的紧凑表示，具有确定性、紧凑性、相似性保持和内存高效等特点。数据集提供了每个模型的详细信息，包括模型名称、张量数量、总参数数量和计算时间。此外，还详细介绍了TensorSketch算法的实现、参数设置、数据格式以及使用方法。

This dataset contains 6,063 Count Sketch fingerprints (TensorSketch) for every tensor in 10 Qwen3.5 and Qwen3 models (≤9B parameters), sorted by HuggingFace downloads. These fingerprints are compact representations of tensors with properties like determinism, compactness, similarity-preserving, and memory efficiency. The dataset provides detailed information for each model, including model name, number of tensors, total parameters, and compute time. Additionally, it details the TensorSketch algorithm implementation, parameter settings, data format, and usage instructions.