https://github.com/torch/cunn
Revision 0d15053cb3aaaedafdb1572e50d693816f1e6f3f authored by Gregory Chanan on 14 October 2016, 23:06:21 UTC, committed by Gregory Chanan on 08 November 2016, 21:01:06 UTC
Half types fail on backward, probably because we don't consistently accumulate in accreal. This is difficult because gradInput is accumulated directly (either with atomicAdd or not) rather than in another variable.
1 parent c077b3f
Tip revision: 0d15053cb3aaaedafdb1572e50d693816f1e6f3f authored by Gregory Chanan on 14 October 2016, 23:06:21 UTC
Add generic support for SpatialSubSampling.
Add generic support for SpatialSubSampling.
Tip revision: 0d15053
File | Mode | Size |
---|---|---|
doc | ||
lib | ||
rocks | ||
.gitignore | -rw-r--r-- | 19 bytes |
CMakeLists.txt | -rw-r--r-- | 850 bytes |
CONTRIBUTING.md | -rw-r--r-- | 4.9 KB |
DataParallelTable.lua | -rw-r--r-- | 21.3 KB |
LICENSE | -rw-r--r-- | 1.9 KB |
README.md | -rw-r--r-- | 3.1 KB |
THCUNN.lua | -rw-r--r-- | 5.2 KB |
init.lua | -rw-r--r-- | 185 bytes |
test.lua | -rw-r--r-- | 207.5 KB |
test_DataParallelTable.lua | -rw-r--r-- | 23.1 KB |
Computing file changes ...