596dc982
feat(imperative): add rng opdef by
2020-09-09 16:23:36 +0800
471a29a5
feat(log): opt log, enable mgb sdk log at opt build more info: 16cd674c56 by
2021-04-25 17:21:54 +0800
800f119c
feat(mge/tools): module_status-add-functions by
2021-04-28 17:44:18 +0800
3ef4fbdc
fix(mge/module): fix a mistaken in BN docs by
2021-04-25 13:22:06 +0800
6d08be37
fix(data): fix the sampling scale in Lighting by
2021-04-25 13:28:52 +0800
aab6b370
Revert "perf(opr): use pin mem for param_pack_concat" by
2021-04-28 13:54:14 +0800
11de49b9
feat(mge/module): python wrapper for conv_transpose3d by
2021-04-22 16:09:16 +0800
c8f18163
refactor(mgb): simplify fast run proccess with storing algo desc instead of algo name by
2021-04-20 19:34:25 +0800
f4efab3b
feat(lite): add cmake build for rknn and mge in lite by
2021-04-09 15:53:50 +0800
46988d45
refactor(mgb): code refactor of fast run by
2021-04-20 19:34:25 +0800
a1e38342
fix(mgb): fix attribute check compatibility when profiling and read_from_cache in fast run by
2021-04-19 16:02:55 +0800
8109a05a
fix(mgb/dnn): fix cub potential issues by
2021-04-21 19:31:11 +0800
331995e7
refactor(imperative): alloc enum type class on heap by
2021-04-20 16:05:05 +0800
3f5238fb
feat(mgb/dnn): add accuracy shake checker by
2021-04-16 16:38:54 +0800
be6fb6b7
feat(mgb/dnn): add accuracy_depend_on_batch attribute by
2021-04-06 12:08:08 +0800
1cadf9d8
fix(mgb): add usable-depend-on-shape attr by
2021-04-01 16:01:54 +0800
fbd578cc
fix(externcopr): check loader imp dynmaic param by
2021-04-20 16:03:27 +0800
cfbd4f9b
fix(device): fix get_device_count for xpu by
2021-03-30 16:46:28 +0800
3bfe87ec
perf(tensor): always copy to pinned memory before h2d by
2021-03-30 16:38:17 +0800
45a031fe
test(dist): remove redundant assert for reduce_sum and gather by
2021-03-30 16:22:36 +0800
17ef140c
fix(rocm): enable var_releaser for rocm by
2021-03-30 16:07:39 +0800
64a3314b
perf(opr): use pin mem for param_pack_concat by
2021-03-30 15:39:08 +0800
04b96602
feat(opr-mm): add backend argument for remote send/recv by
2021-03-30 15:31:14 +0800
0083f4c4
build(rocm): support rocm-3.9 by
2021-03-23 16:01:32 +0800
d4eb3890
feat(distributed): auto detect device and backend when init group by
2020-11-18 13:33:56 +0800
d3154a45
feat(distributed): support distributed opr for rocm by
2020-11-18 13:30:52 +0800
928a57f8
build(rocm): partially support hcc compilation by
2020-11-18 13:27:15 +0800
100a5027
fix(dnn): replace kernel launch syntax with macro for hcc by
2020-11-18 13:08:29 +0800
07ab8cb6
feat(dnn): add param_pack for rocm by
2020-11-18 12:02:16 +0800
6ac68090
refactor(mgb/opr): make trt batch flag only depend on inputs dimension by
2021-04-16 17:57:56 +0800
00fa02a9
fix(mgb): fix attribute uncomplete filter when get_profile_result_from_cache in fast run by
2021-04-12 17:30:53 +0800
9be7cb89
test(mge/module): add module reset attribute test by
2021-04-14 18:26:52 +0800
7a9d1f57
fix(mge/module): fix module attribute update mistake by
2021-04-14 18:01:21 +0800
38fdd762
build(cuda): link to cuda_stub by
2021-04-09 14:26:57 +0800
4b2b623b
fix(dnn/cuda): fix cutlass matmul splitk limit by
2021-03-31 17:58:34 +0800
f295e358
fix(mgb): fix fast run crash when profile heuristic strategy by
2021-04-12 17:30:53 +0800
e4af4225
fix(cmake): fix cmake depends reason: some target call python3 to generate some headers, if PYTHON_EXECUTABLE changed, headers will be invalid, which will lead to rebuild most of obj, because this headers will be include in most cpp/c, what`s more, this target do not depends python3 version(python API) the output is same when use python3.5 or python3.x so change use PYTHON_EXECUTABLE to PYTHON3_EXECUTABLE_WITHOUT_VERSION, when PYTHON_EXECUTABLE/PYTHON_LIBRARY/PYTHON_INCLUDE_DIR changed, can reuse the build obj by
2021-05-10 13:50:56 +0800
6bb6787d
feat(mge): add a tool which can analyze the file generated by compare_binary_iodump.py by
2021-05-10 15:42:25 +0800
43ef9748
feat(mgb): load_and_run can set both fastrun and reproducible by
2021-05-10 15:37:22 +0800
c3f8cf04
feat(dnn): add conv_bwd_data and conv_bwd_filter accuracy shake check by
2021-04-27 16:55:26 +0800
717b88e6
fix(mge/elemwise): fix problem that elemwise.mode is not comparable with string mode by
2021-05-12 11:51:19 +0800
9c17cfc4
fix(imperative/ops): add check_dtype for Elemwise in infer_attrs by
2021-05-11 17:36:06 +0800
05550bc5
fix(mge/quantization): fix a compute problem by
2021-05-11 16:58:09 +0800
813dd2f1
fix(mge/core): support non-Tensor args for `core.tensor.utils.get_device` by
2021-05-11 14:23:28 +0800
9f5bb42d
perf(fastrun): cache persistent cache by
2021-05-12 17:05:04 +0800
6ab1c55d
fix(mgb): fix fastrun workspace limit by
2021-05-12 14:26:50 +0800
7af1ec66
fix(tensor): del valid tensors when compnode finalizing by
2021-03-18 13:17:39 +0800
696d2c2e
fix(mge/autodiff): check tensors to be attached by
2021-05-13 14:56:53 +0800
984d85ca
feat(mge/functional): argmin and argmax support negtive axis by
2021-05-05 11:44:49 +0800
87d6ff22
fix(mge): fix dumping backward graph by
2021-05-06 15:37:26 +0800
f31752d5
feat(mge/module): add __repr__ method for qat and quantized module by
2021-03-30 14:45:06 +0800
9451a961
test(mge/optimizer): update optimizer test to make sure grad not change by
2021-04-23 17:11:13 +0800
92e2ed6e
fix(mge/optimizer): fix optimizer update step inplace add grad by
2021-04-23 16:24:56 +0800
69728969
fix(mge/utils): fix network multiple outputs issue by
2021-05-06 18:26:31 +0800
f36e99d3
fix(build): fix naive build by
2021-04-29 17:24:33 +0800
51676939
fix(mge/imperative): fix proxy graph comp_node finalize by
2020-10-14 19:11:08 +0800
13e6ea34
feat(imperative/opr): rebase rng refactoring to dev & add python module by
2021-02-24 21:16:33 +0800
cded8ef1
feat(imperative): add rng opdef by
2020-09-09 16:23:36 +0800
40bab1ed
feat(log): opt log, enable mgb sdk log at opt build more info: 16cd674c56 by
2021-04-25 17:21:54 +0800
c3a1ac3d
feat(mge/tools): module_status-add-functions by
2021-04-28 17:44:18 +0800
05e4c826
fix(mge/module): fix a mistaken in BN docs by
2021-04-25 13:22:06 +0800
f6e66ec6
fix(data): fix the sampling scale in Lighting by
2021-04-25 13:28:52 +0800
00b48dfe
Revert "perf(opr): use pin mem for param_pack_concat" by
2021-04-28 13:54:14 +0800
e18afa0b
feat(mge/module): python wrapper for conv_transpose3d by
2021-04-22 16:09:16 +0800
fdf7006b
fix(imperative): add __init__.py for tools by
2021-04-19 20:39:49 +0800
f16c9eb9
refactor(mgb): simplify fast run proccess with storing algo desc instead of algo name by
2021-04-20 19:34:25 +0800
3e172f3b
feat(lite): add cmake build for rknn and mge in lite by
2021-04-09 15:53:50 +0800
5d637d07
refactor(mgb): code refactor of fast run by
2021-04-20 19:34:25 +0800
f6bd4f59
fix(mgb): fix attribute check compatibility when profiling and read_from_cache in fast run by
2021-04-19 16:02:55 +0800
0a86a070
fix(mgb/dnn): fix cub potential issues by
2021-04-21 19:31:11 +0800
282dfc62
refactor(imperative): alloc enum type class on heap by
2021-04-20 16:05:05 +0800
1e6ef377
feat(mgb/dnn): add accuracy shake checker by
2021-04-16 16:38:54 +0800
a5a29826
feat(mgb/dnn): add accuracy_depend_on_batch attribute by
2021-04-06 12:08:08 +0800
4b141f8d
fix(mgb): add usable-depend-on-shape attr by
2021-04-01 16:01:54 +0800
15b647ae
fix(externcopr): check loader imp dynmaic param by
2021-04-20 16:03:27 +0800
5c131d2a
fix(device): fix get_device_count for xpu by
2021-03-30 16:46:28 +0800
9eb4ba88
perf(tensor): always copy to pinned memory before h2d by
2021-03-30 16:38:17 +0800
4d9073fc
test(dist): remove redundant assert for reduce_sum and gather by
2021-03-30 16:22:36 +0800
3bd0df8e
fix(rocm): enable var_releaser for rocm by
2021-03-30 16:07:39 +0800
21c6c437
perf(opr): use pin mem for param_pack_concat by
2021-03-30 15:39:08 +0800
1a711299
feat(opr-mm): add backend argument for remote send/recv by
2021-03-30 15:31:14 +0800
69a146c8
build(rocm): support rocm-3.9 by
2021-03-23 16:01:32 +0800
dee5a10a
feat(distributed): auto detect device and backend when init group by
2020-11-18 13:33:56 +0800
1bec737d
feat(distributed): support distributed opr for rocm by
2020-11-18 13:30:52 +0800
a31b7c6e
build(rocm): partially support hcc compilation by
2020-11-18 13:27:15 +0800
621ae0a1
fix(dnn): replace kernel launch syntax with macro for hcc by
2020-11-18 13:08:29 +0800
78fff72a
feat(dnn): add param_pack for rocm by
2020-11-18 12:02:16 +0800
6de3e4ba
refactor(mgb/opr): make trt batch flag only depend on inputs dimension by
2021-04-16 17:57:56 +0800
ce610ca3
fix(mgb): fix attribute uncomplete filter when get_profile_result_from_cache in fast run by
2021-04-12 17:30:53 +0800
c992afa9
test(mge/module): add module reset attribute test by
2021-04-14 18:26:52 +0800
93bfda51
fix(mge/module): fix module attribute update mistake by
2021-04-14 18:01:21 +0800
c9348b16
build(cuda): link to cuda_stub by
2021-04-09 14:26:57 +0800
33ab2454
fix(whl): support wheel version without sdk info, which caused by python official PYPI do not support LOCAL_VERSION; sync lite setup.py by
2021-04-19 15:36:15 +0800
8163ed15
fix(dnn/cuda): fix cutlass matmul splitk limit by
2021-03-31 17:58:34 +0800
50f182a0
(tag: v1.4.0-rc1)
fix(whl): support wheel version without sdk info, which caused by python official PYPI do not support LOCAL_VERSION; sync lite setup.py by
2021-04-19 15:36:15 +0800
976b351a
feat(functional/nn): support F.warp_perspective with `mat_idx` by
2021-04-19 13:57:37 +0800
899d70a8
fix(imperative): add __init__.py for tools by
2021-04-19 20:39:49 +0800
933188aa
feat(functional/nn): support F.warp_perspective with `mat_idx` by
2021-04-19 13:57:37 +0800
8585aa61
fix(mgb): fix fast run crash when profile heuristic strategy by
2021-04-12 17:30:53 +0800
4a8b01fc
chore(release): bump version by
2021-04-19 18:55:46 +0800