trunk/542330ebe79bfc3e9275d7610f9adcf489e3abf9: Fix scaled-MM DeepSeek indexing and contraction validation (#197028)
- PyTorch 近 90 天出现 1030 次
- PyTorch 近 90 天第 1019 次发版
- 上一次：同一天稍早 · trunk/7083f3a16277e0e4f58b5064dcc5d9840bf2c24c: [cuda][nccl] Drop the dead `NCCL_ALLTOALL_SUPPORTED` arm (#196400)
发生了什么
Human note I found this repeatedly failing on blackwell CI Agent note The DeepSeek Triton main loop assumed row-major scale storage, advanced scale indices incorrectly for GEMM tiles smaller than a scale block, and interleaved outer/K dimensions when expanding 128x128 scales. Honor actual strides and logical axes, compute scale coordinates from element offsets, and preserve outer/K order during expansion. Explicit v…
摘要按规则整理自下方来源原文