Skip to content

v2.0.1
Compare
Choose a tag to compare
@milakov milakov released this 23 Nov 21:39
· 128 commits to master since this release
v2.0.1
0586d45
  • Multiple improvements to reduce total buffer sizes, allows running larger chunks, (3x for ImageNet):
    • Taking buffer sizes into account when coloring graph
    • Maxout, ReLU, and MaxSubsampling layers consume much less memory in CUDA backend
    • Action graph is optimized to exclude unnecessary concurrency
  • Migrated to cuDNN v3
  • Reusing CUDA streams
  • Allocating chunk of mem for fixed working buffers - improves perf
  • Few bug-fixes
Assets 2
Loading