a12a471b2d
suppress compiler warning.
2015-08-28 14:02:53 -07:00
ea0c7afa49
Delete db.cc
...
db.cc has been moved to caffe2/core and this is no longer used.
2015-08-27 10:45:41 -07:00
561bf8eb1b
Merge branch 'master' of https://github.com/Yangqing/caffe2
2015-08-15 07:20:19 -07:00
7d021a0346
context change and dropout bugfix
2015-08-15 07:17:07 -07:00
d6bebc4824
lrn bugfix - for specific cuda devices a single if is not enough, need to do kernel loop.
2015-08-15 07:15:14 -07:00
10ffe1132f
image input: support caffe datum format
2015-08-08 13:04:02 -07:00
b4656c77b3
prefetch op bugfix
2015-08-08 13:01:12 -07:00
4b32534e84
bugfix for dropout and filler
2015-08-06 13:39:44 -07:00
a07c255d16
Some utility function changes
2015-07-29 09:21:02 -07:00
d829950eff
change arg order of Copy/Memcpy to follow inputs-then-outputs convention
...
instead of C memcpy order -- from (dst, src, n) to (n, src, dst)
2015-07-27 21:19:32 -07:00
b856798020
disable CannotAccessRepeatedParameterWithWrongType test broken by
...
removal of check (PR #7 )
2015-07-27 18:28:44 -07:00
75b1f2f868
Merge pull request #8 from jeffdonahue/tensor-init-values-const
...
Tensor constructor: values arg is const
2015-07-27 18:14:16 -07:00
091aabeaf0
operator.cc: remove check for empty repeated argument (to allow empty
...
shapes and other use cases)
2015-07-27 18:09:58 -07:00
4aad861f9f
allow Tensor with empty dims (a scalar)
...
use in loss functions for scalar loss output
2015-07-27 18:08:06 -07:00
7b481f5a8f
Tensor constructor: values arg is const
2015-07-27 13:38:54 -07:00
7f5ee34b9a
add back gflags dependency
2015-07-23 20:40:39 -07:00
a6d20495c2
[gflags] sort out the gflags namespace issue.
2015-07-23 20:12:35 -07:00
c3ba30a537
blob templating
2015-07-22 20:59:22 -07:00
aaead5d6a5
reorg mpi_ops
2015-07-20 21:24:05 -07:00
90affff039
broadcast op: it is an in-place op with both input and output set
2015-07-19 20:15:50 -07:00
745d8ed969
run_plan_mpi: check the returned state
2015-07-19 20:14:43 -07:00
d4eab84548
conv_op: during backward, bias is not needed.
2015-07-19 20:14:22 -07:00
691986ec21
Add an execution chain option to force operator orders.
2015-07-19 20:14:05 -07:00
a5254881f2
Some more MPI examples
2015-07-19 14:21:44 -07:00
571ee16b44
minor change
2015-07-19 09:13:07 -07:00
cfab4ed865
zmq feeder bugfix
2015-07-18 18:50:45 -07:00
43aaadbef4
zmq feeder: catch error when setting up the socket.
2015-07-18 17:38:36 -07:00
ecf0dceef6
zmqdb: pass key as well.
2015-07-18 17:26:29 -07:00
05ba5b0527
Use c++ to do zmqdb, and added a simple zmq feeder example.
2015-07-18 14:56:34 -07:00
47c70a43b4
(1) minidb bugfix
...
(2) blob serialization comments
(3) cudnn: putting it under a separate device name
so we can explicitly choose cudnn instead of
having CUDA device prioritizing it.
(4) note that mint is not available with ipython
due to zeromq conflict
(5) db_throughput utility
(6) added gprofiler
2015-07-18 07:23:09 -07:00
c5166e578c
Several changes:
...
[misc] Update license and readme.
[binary] some enhancement to move over caffe databases.
[python] added an alexnet2 example notebook.
2015-07-08 19:42:02 -07:00
85a40a0b96
zmqdb: make clear error message that zmq 3+ is required.
2015-07-06 22:01:48 -07:00
f15492f303
make things compile for gcc
2015-07-06 22:00:02 -07:00
16c253e62e
Some non-trivial refactoring:
...
(1) added blob serialization.
(2) registry can now use key types other than string.
(3) changed load_save_op so they interface with a db.
(4) change sgd iter op: it does increments so we can resume an iter.
(5) mnist linear classifier tests snapshot functionality.
(6) added protodb which is a small wrapper over TensorProtos.
2015-07-06 21:17:18 -07:00
dc41b23e41
[bugfix] static initialization order fiasco.
2015-07-06 21:17:18 -07:00
97f4b9f3e7
[bugfix] missing dependency
2015-07-06 21:17:18 -07:00
e078bf4e81
[op] iter_op: instead of producing an int, produces it in a wrapped tensor.
2015-07-06 21:17:18 -07:00
2a3fda4d60
[interface] OperatorBase::OutputIsType
2015-07-06 21:17:15 -07:00
8ba95403fd
minor format issue
2015-07-06 11:37:31 +08:00
036229c889
[style] Finishing name changes for the rest of the fields in the protobuf.
2015-07-01 18:16:43 -07:00
c1077400c9
[style] Massive name change from plural to singular. This one changes operator’s “inputs” and “outputs” to “input” and “output”, and "args" to "arg".
2015-07-01 15:31:01 -07:00
df6fd55cce
[minor] net.cc: in parallelnet, if the user forgot to set a num_workers, warn instead of quitting.
2015-07-01 14:09:43 -07:00
241cad91f2
[LegacySupport] Fixed a bug in the caffe pooling case: pad_tail is changed on first run so we can only use pad_head for the legacy padding value.
2015-07-01 14:08:57 -07:00
e94be296b7
Fix some bug in the old code: the net device option overwite used to not work.
2015-07-01 13:38:25 -07:00
8834e3eb13
Adding <memory> to the header. Clang is fine with missing this but gcc complains.
2015-07-01 12:37:24 -07:00
8bedbde1c1
More flag changes… and adding back run_plan_mpi
2015-07-01 12:35:34 -07:00
314696a262
explicitly make end_to_end_test depend on core_gpu
2015-07-01 12:21:58 -07:00
fdf6066e45
namespace google -> gflags for gflags functions.
2015-07-01 11:22:23 -07:00
2807aac523
enable optional_deps so that any non-crucial failures will not break the whole system.
2015-06-30 20:41:41 -07:00
2abd5e263e
GoogleNet adaption - added yet another legacy padding support.
2015-06-30 19:40:05 -07:00
