Tags: MJxiaomings/pytorch
Tags
[ROCm] fix ROCm 5.5 nightly build after hipblas change
Update on "fix bf16 constant accuracy" This PR aims to sort out the data type for `constant`. The constant should be promoted to float pytorch#105440. So there are serval changes to do: - Data type propagation should propagate constant node to `float` dtype if original dtype is `bfloat16` - We do not need to insert `to_dtype` after the `constant` node, directly init an `fp32` constant is faster. ``` vectorized<bfloat16> tmp(value); vectorized <float> tmp1 = cvt_bf16_fp32(tmp); -> vectorized<float> tmp(value); ``` - move `constant` out of the list for `all operations can support bf16 without converting to fp32` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 ipiszy ngimel yf225 chenyang78 kadeng muchulee8 aakhundov [ghstack-poisoned]
Update on "Move ASAN to clang12 and Ubuntu-22.04 (Jammy)" Modify `install_conda` to remove libstdc++ from libstdcxx-ng to use one from OS Modify `install_torchvision` to workaround weird glibc bug, where malloc interposers (such as ASAN) are causing a hang in internationalization library, see https://sourceware.org/bugzilla/show_bug.cgi?id=27653 and https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90589 Extracted from pytorch#105260 [ghstack-poisoned]
[ROCm] fix ROCm 5.5 nightly build after hipblas change
[dynamo] Support dict.get with no default specified
Update on "adding mixed_dtype_mm to torch._inductor" Summary: if torch._inductor.config.use_mixed_mm then we can convert torch.mm(a, b.to(some_dtype)) into a triton kernel where the casting b is fused into the matmul rather than needing to instantiate the casted b tensor. This is necessary for weight-only quantization where we don't want to load memory with a tensor 4x the size of our quantized one. Test Plan: python test/inductor/test_pattern_matcher.py -k "mixed_mm" python test/inductor/test_torchinductor.py -k "mixed_mm" Reviewers: Subscribers: Tasks: Tags: cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 ipiszy ngimel yf225 chenyang78 kadeng muchulee8 aakhundov [ghstack-poisoned]
AOTInductor compile in prod env (pytorch#106442) Summary: Pull Request resolved: pytorch#106442 This diff updates the Inductor internal compile workflow Reviewed By: houseroad, wushirong Differential Revision: D47958727 fbshipit-source-id: be9660319c2400cb8c2d247fa6255cab14c4546c
reword error message, import TEST_WITH_TORCHDYNAMO
PreviousNext