05ee6038
fix(mge/distributed): fix single_machine_cache error by
2021-11-18 16:50:04 +0800
0a56a5b1
feat(cuda/comp_node): gets (maximum) GPU memory allocated/reserved by
2021-11-03 16:28:28 +0800
896a6fb0
fix(lite/load_and_run): fix lite load_and_run disable_assert_throw by
2021-11-19 17:33:39 +0800
5c94db16
feat(mge/functional): add groups support for conv_transpose2d & 3d by
2021-11-19 19:11:13 +0800
f2f33565
fix(mge): fix some minor problems by
2021-11-03 20:23:00 +0800
7591718d
feat(mge): add functional test by
2021-11-09 11:37:10 +0800
8a222d2c
fix(mge): replace functional paramter by
2021-11-09 11:34:44 +0800
27aa648b
fix(mge): control compute_mode by context by
2021-11-07 16:55:09 +0800
dfb9d980
feat(ci): remove py format checking by
2021-11-18 14:42:41 +0800
85908148
fix(lar): fix load and run get static memory error by
2021-11-17 18:28:52 +0800
13c7c572
fix(mgb): fix shape infer's condition in lite by
2021-11-03 15:45:42 +0800
8d825246
fix(sdk/load_and_run): remove docs of dump_with_testcase_with_mge.py by
2021-11-01 20:11:21 +0800
38b7cfde
fix(mge/utils): fix module states input is dict or others by
2021-11-16 17:40:25 +0800
16131359
fix(imperative): insert to candidate only if pinned is 0 by
2021-11-16 17:48:19 +0800
846cc13c
fix(mge/functional): fix default dtype of F.full by
2021-11-08 16:04:15 +0800
0ddabb06
fix(imperative/astype): fix astype when target dtype is int4 by
2021-10-26 17:47:26 +0800
a1cba6cc
fix(dnn): fix convbias crash on X86 by
2021-11-03 20:02:46 +0800
9b4cd92b
fix(mgb/dnn): fix cudnnConvBiasActivation crash on nchw32 int8 with oc > 256 by
2021-09-30 16:03:12 +0800
34773ba3
fix(mgb/gopt): tensorcore pass replace BatchConvBias inputs to nchw4 by
2021-11-05 18:54:11 +0800
efe3cfe7
fix(lite): fix the encrypt key in lite by
2021-11-12 20:22:11 +0800
d4a080e1
feat(mge/third_party): update MegRay version by
2021-11-15 14:37:56 +0800
63bb7c29
docs(mge/functional): update functional.arange docstring by
2021-11-03 16:59:47 +0800
89ed7ab2
test(imperative): speed up dtr test by
2021-11-11 17:49:15 +0800
a2a09ef9
fix(imperative): release dtr related resources when disable dtr by
2021-11-11 22:09:27 +0800
2676fb73
refactor(imperative): remove enable_defrag interface as it is enabled by default by
2021-11-09 18:10:43 +0800
d909950f
refactor(imperative): remove swap in dtr by
2021-11-05 13:32:30 +0800
bcf69d8f
refactor(imperative): correctly apply sqrt sampling for dtr by
2021-11-04 19:26:02 +0800
48100781
fix(mgb): support rc version by
2021-11-05 19:27:03 +0800
f00741da
feat(build): bazel build support flatbuffer serialization by
2021-10-28 17:08:31 +0800
829f0907
fix(mge/traced_module): fix insert qat module by
2021-11-09 20:07:25 +0800
8b764934
feat(lite): lite support output var no copy option by
2021-11-01 19:37:09 +0800
7642f66d
feat(core): support output var no copy by
2021-11-01 16:40:11 +0800
7011926c
refactor(sdk): remove old load and run by
2021-10-28 20:39:25 +0800
e02f1e07
fix(ci): replace old lar with new lar by
2021-10-28 20:38:39 +0800
37c1726f
refactor(sdk): refactor load and run with new framework by
2021-10-28 19:47:37 +0800
b75658c8
fix(build): use official build script to build nccl by
2021-11-05 19:27:58 +0800
ec59a5bc
fix(build): fix ninja dry run by
2021-11-02 19:27:04 +0800
b18544b4
feat(mge/third_party): update Json version by
2021-11-15 10:39:31 +0800
edfd38be
fix(mge/traced_module): fix node naming in the flattened graph by
2021-11-04 16:33:08 +0800
b8776574
ci(pytest): fixture doctest namespace in pytest config by
2021-11-03 15:01:23 +0800
ba8bd010
fix(mge/traced_module): fix insert module by
2021-11-01 11:59:56 +0800
b8316de5
fix(mge/traced_module): patch missed functional in functional.nn by
2021-11-01 10:25:59 +0800
dbca3270
fix(mge/traced_module): fix module dict by
2021-10-31 21:24:05 +0800
01d2473c
fix(mge/traced_module): fix TracedModule flatten by
2021-10-31 19:45:42 +0800
23c1fda7
perf(arm_common): optimize sigmoid by
2021-10-25 18:15:18 +0800
b20cda6b
fix(pylite): fix pylite callback test bug by
2021-10-28 13:35:56 +0800
c361b193
feat(lite-c): add lite C callback with user_data API by
2021-10-28 13:28:26 +0800
7fa5f6f4
fix(mgb): correct version of fbs serialization by
2021-10-29 19:34:21 +0800
a404c508
feat(mge): support dump with specific format by
2021-10-29 19:33:57 +0800
fba523a1
fix(mge/device): fix the directive content block by
2021-11-02 20:03:50 +0800
25ec2530
feat(whl/api/lar): enable megengine dll on Windows 1: reduce python whl package size 2: unify api link logic on all OS 3: add option: MGE_WINDOWS_BUILD_WITH_STATIC_CRT --- default OFF --- if build CRT(vc runtime) with STATIC with megengine.dll some CRT api will crash, for example, flush, so if you build with static megengine, and do not want to install CRT you can set MGE_WINDOWS_BUILD_WITH_STATIC_CRT TRUE --- how to install CRT: https://docs.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-160 install VC_redist.x64.exe 4: rename megengine_export to megengine_shared(only export needed symbols ), caused by runtime symbols conflict with pytorch by
2021-10-19 10:45:59 +0800
feea43bc
fix(mge/traced_module): associate name with node by
2021-10-26 21:24:02 +0800
a6fe7f7f
refactor(mge/traced_module): refactor Node naming rule and merge GetAttr by
2021-10-08 14:48:43 +0800
c48d58da
feat(dnn/arm_common): add N1HW like elemwise broadcast mode by
2021-10-28 16:23:02 +0800
669c3cda
feat(mge): enable defrag by default by
2021-11-01 20:58:56 +0800
c17d6038
feat(mgb): allow output tensor's ptr change when record by
2021-10-25 20:06:11 +0800
26634db7
fix(dnn): support relayout for non-contigous layout by
2021-10-28 19:26:13 +0800
e0aaf560
docs(lite): delete readme in lite by
2021-10-27 14:27:01 +0800
d9a9d9d4
fix(imperative/fastrun): set workspace limit for imperative rt by
2021-09-07 18:06:27 +0800
a09a2b73
fix(mgb/opr): fix fastrun workspace limit for imperative rt by
2021-08-30 18:32:30 +0800
ac86d644
fix(mge/module): fix prelu error when use_symbolic_shape is true by
2021-10-28 11:27:15 +0800
4f6c5d8f
feat(mge/dump): enable jit.dump to dump with testcase by
2021-10-28 14:42:42 +0800
182ca25d
feat(mge/imperative): add coalesce_all_free_memory mge api by
2021-10-15 14:20:32 +0800
0b2676c6
fix(lite): fix lite profile bug when using packed model with mode info by
2021-10-27 19:59:49 +0800
5207154e
Revert "fix(mgb): fix fastrun no_profiling_on_shape_change" by
2021-09-30 20:07:49 +0800
cf2ae2eb
doc(mge):update readme by
2021-11-03 10:39:37 +0800
d18d3b5d
doc(mge): add bibtex msg by
2021-11-03 10:37:29 +0800
b426723e
feat(mgb/third_party): add gflags submodule by
2021-10-26 17:18:13 +0800
9d05738d
Merge pull request #225 from chenjiahui0131:pixel_shuffle_module by
2021-10-27 15:20:02 +0800
46b7c327
feat(lite): add get static mem info function in pylite by
2021-10-15 19:03:48 +0800
5e8a9a95
feat(lite): add get static mem info function in lite c by
2021-10-15 18:46:20 +0800
8bdcf6b5
feat(lite): add get static mem info function in lite c++ by
2021-10-15 16:32:42 +0800
b84d2893
fix(lite): fix lite c thread local by
2021-10-13 11:11:12 +0800
936bb237
fix(mge/tensor): fix is/is not SyntaxWarning occurs in python3.8 by
2021-10-26 15:43:05 +0800
8d507cc3
docs(mge/module): add note about definition of momentum in BatchNorm by
2021-10-25 18:22:54 +0800
056fd6bc
feat(dnn/arm64): support stride_m in arm64 relayout by
2021-10-18 17:15:50 +0800
ec75cd86
add: add gflags submodule for megengine by
2021-10-20 17:21:32 +0800
bd42cb06
refactor(mgb/lite): refactor lite InfilePersistentCache with core impl by
2021-10-20 19:06:07 +0800
676b205b
style(imperative): add bypass of clang-format for cg options by
2021-10-25 13:39:29 +0800
2c5a007a
fix(mge/optimizer): allow lr to be 0 by
2021-10-18 10:56:33 +0800
c50858ee
fix(dnn): specialize pow to make it consistent by
2021-10-14 15:06:44 +0800
d898838e
fix(src/gopt/test): remove device info in the cache_data by
2021-10-19 22:29:50 +0800
c8a9a3fd
feat(imperative): add PixelShuffle to module.__init__.py by
2021-10-21 18:28:57 +0800
a2a46b56
fix(lite): fix rknn error in lite by
2021-10-13 20:43:58 +0800
849f0ece
fix(dnn): drop batched matmul cublas algo when batch is 1 by
2021-10-14 15:37:59 +0800
b5bf56e0
style(dnn): add bypass of clang-format for dnn foreach_opr macro by
2021-10-15 16:14:14 +0800
5af52746
fix(mgb): fix bug caused by conv filter size is too big by
2021-09-26 19:51:34 +0800
b0ba6d32
Merge pull request #207 from togetherwhenyouwant:feat-x86-matmul-6x16x2 by
2021-10-18 19:44:19 +0800
a4ac5e7e
fix(mgb/gopt): fix profiler cache when build without opr type info name by
2021-10-11 13:24:35 +0800
012de769
feat(mgb/gopt): add profiler cache by
2021-09-28 13:58:48 +0800
b33ec46e
fix(mgb/gopt): fix layout gtrans when graph partition has opr with different format by
2021-09-30 13:41:39 +0800
fe15239a
fix(imperative): fix error message for tensors with intbx data type by
2021-10-14 12:44:04 +0800
10af44ab
fix(dnn/cuda): fix cudnn conv impl for nchw4_nchw hybrid layout by
2021-10-14 16:59:22 +0800
feb813bc
chore(release): bump version by
2021-10-14 18:46:57 +0800
5885b137
feat(dnn/arm): support layout like NHWC channel like broadcast on arm by
2021-10-08 10:47:57 +0800
565466c2
feat(lite): auto deduce output tensor shape before model forward by
2021-10-11 16:16:08 +0800
29f9935d
fix(imperative/python): add layer_norm doc and rm useless param by
2021-10-12 14:44:46 +0800
30095514
fix(traced_module/doc): fix warning for traced_module docstring by
2021-10-08 16:25:15 +0800
4cb7fa8e
ci(check): force check clang-format version by
2021-10-13 10:51:04 +0800
f7d05db7
fix(mge/module): fix named_tensors by
2021-10-12 13:12:11 +0800