05 2022 档案
摘要:A.scatter_(dim, index, B) # 基本用法, tensor A 被就地scatter到 tensor B 如果dim = 0,意思就是在保持行索引不变的情况下,针对每一行,把每个原色的列索引转换为index里面的数值。 可以用于生成one - hot embedding 参考:
阅读全文
摘要:autocast op reference Op Eligibility Ops that run in float64 or non-floating-point dtypes are not eligible, and will run in these types whether or not
阅读全文
摘要:CTA: Cooperative Thread Array 即 CUDA BLOCK https://github.com/NVIDIA/cuda-samples/blob/2e41896e1b2c7e2699b7b7f6689c107900c233bb/Samples/3_CUDA_Feature
阅读全文