遇见数据集

nunchaku-ai/cuda-challenge-flux-dump-holdout

收藏
Hugging Face2026-04-17 更新2026-05-10 收录
官方服务:

资源简介:

--- license: apache-2.0 pretty_name: CUDA Challenge — FLUX.1-schnell dump (holdout) tags: - cuda - quantization - int4 - gemm - flux --- # FLUX.1-schnell activation & weight dump (holdout) Holdout set for the W4A4 Quantized GEMM CUDA Challenge. Same layer / step / shape / prompt as the [public dump](https://huggingface.co/datasets/samuelt0207/cuda-challenge-flux-dump), different seed — used to validate that submitted kernels generalize rather than hard-coding to specific activation values. Seed: `42`. Otherwise identical extraction to the public dump. ## Files - `weights.pt` — bit-identical to the public dump (weights are seed-independent) - `activations_1024x1024.pt` — different from the public dump (different noise → different hidden states) Shapes match the public dump. See that README for the GEMM shape table.

许可证:Apache-2.0 展示名称:CUDA挑战赛——FLUX.1-schnell 留存测试集(holdout) 标签: - CUDA - 量化(quantization) - int4 - GEMM(通用矩阵乘法) - FLUX # FLUX.1-schnell 激活值与权重留存测试集(holdout) 本数据集为W4A4量化GEMM CUDA挑战赛的留存测试集。其层结构、迭代步数、张量形状与提示词均与[公开数据集](https://huggingface.co/datasets/samuelt0207/cuda-challenge-flux-dump)一致,仅随机种子不同——该数据集用于验证提交的计算内核具备泛化能力,而非针对特定激活值进行硬编码。 随机种子固定为`42`,其余数据提取流程与公开数据集完全一致。 ## 文件列表 - `weights.pt`:与公开数据集的权重文件比特完全一致(权重与随机种子无关) - `activations_1024x1024.pt`:与公开数据集的激活值文件不同(不同的噪声会生成不同的隐状态) 张量形状与公开数据集完全一致,GEMM的形状表请参阅该数据集的README文件。

提供机构:
nunchaku-ai
二维码
社区交流群
二维码
科研交流群
商业服务