master
[MXNET-912] Refactoring ctc loss operator (#12637)
Add bfloat16 floating-point format support based on AMP (#17265)
Switch to C++17 and modernize toolchain + CI (#17984)
License Googletest and Appendix (#14687)
[LICENSE] Change intgemm to a submodule instead of fetch. (#19408)
Switch npx.savez from libzip to miniz (#19584)
Improve bf16 support (#21002)
Update CUB and include it only for CUDA < 11 (#18799)
[submodule] Upgrade oneDNN to v2.6.1 (#21108)
Fix CI in master (#21026)
Update 3rdparty/ps-lite (#17732)
Update TVM integration to v0.7 (#19613)