ef9aa800
fix(mgb/dnn): fix cuda naive matmul algo by
2021-04-15 15:44:58 +0800
fe99cdc7
feat(interpreter): add dynamic sublinear by
2021-03-02 15:59:33 +0800
62c394ca
feat(cuda/comp_node): enable to query adjacent free blocks size by
2021-03-09 14:56:05 +0800
fd61f095
feat(cuda/comp_node): enable to directly query memory status by
2021-03-02 14:18:15 +0800
a3ab9f1d
feat(blob_manager): make all memory allocation go through blob manager by
2021-03-02 11:56:32 +0800
3ea00eba
fix(cmake): fix cmake build failed at soft link workspace by
2021-04-15 17:15:09 +0800
a8108522
refactor(imperative): refactor enum param type caster by
2021-04-14 18:15:12 +0800
dcff115e
fix(distributed/launcher): fetch data early to avoid dead lock in launcher by
2021-04-13 19:10:32 +0800
7e22e9f0
feat(optimzer): add AdamW by
2021-04-14 16:01:49 +0800
7995f857
(release-1.3)
chore(release): bump version by
2021-04-14 15:30:41 +0800
f3f9acf1
fix(tree): fix copybara by
2021-03-22 19:16:09 +0800
36fbd5a6
fix(mgb): fix spell error by
2021-04-07 10:55:20 +0800
957d1d40
fix(src/gopt): add replace func of typecvt opr for nhwcd4 pass by
2021-04-07 20:46:59 +0800
865fbe16
feat(imperative/utils): add optimize-for-inference interface for opgraph by
2021-04-07 12:35:55 +0800
c92317ed
fix(mge/utils): filter out parameter "arg_names" and "output_name" in network.dump by
2021-04-07 12:53:11 +0800
24b91b98
feat(mge/utils): add array method for varnode by
2021-04-04 12:50:38 +0800
5e54fdc3
fix(imperative/utils): fix name filter of network varnode by
2021-04-09 18:23:55 +0800
2369c4f6
fix(mge/utils): fix auto naming bug when expanding structure by
2021-04-09 18:00:16 +0800
d275a823
chore(scripts): clarify and fix default value of bit combined enum by
2021-04-02 15:20:33 +0800
7c715bd4
fix(mge/utils): fix module stats calculate flops bug for group conv and remove model status change by
2021-04-08 17:53:44 +0800
245a3f81
fix(mge/tools): fix module stats' receptive field bug for Module by
2021-04-08 15:54:54 +0800
007a2376
fix(mge/tools): fix node display bug in tensorboard by
2021-03-31 17:32:24 +0800
b10238ac
feat(mge/tools): add support of receptive_field stats for NetworkNode by
2021-03-26 19:15:25 +0800
84c2a5c2
feat(mge/tools): add summary print for module_stats and network_visualize by
2021-03-26 16:21:18 +0800
edea528b
feat(mge/tools): set network_visualize's log_path as optional flag by
2021-03-23 16:57:26 +0800
e6d77604
refactor(imperative): refactor tablegen code generator by
2021-03-29 17:59:53 +0800
cad8568c
fix(mge/optimizer): fix optimizer's state_dict bug by
2021-03-31 12:01:56 +0800
0ed36998
fix(mge/quantization): fix quantized concat forward problem by
2021-03-30 11:16:03 +0800
888c7f1f
fix(mge/module): fix auto-naming error when there are containers in the module by
2021-03-24 11:34:21 +0800
d13b6a4a
fix(mgb/core): fix allocate_task_block_unsafe in thread_impl_1.h by
2021-03-24 15:10:50 +0800
1edcfa19
fix(imperative/opr): fix apply_on_var_node for broadcast by
2021-03-25 15:11:52 +0800
2d18074a
fix(mgb): fix spell error by
2021-04-07 10:55:20 +0800
65af9cfc
refactor(mge): use lower case for default string parameters in functional and module by
2021-03-30 17:32:21 +0800
d0aa9b41
refactor(mge/functional): move nvof to vision, compatible with old usage by
2021-03-31 14:57:24 +0800
ff755451
refactor(mgb): move algo's name from info to desc and delete some algo's unnecessary param() method by
2021-04-08 18:02:55 +0800
a437ec8e
fix(src/gopt): add replace func of typecvt opr for nhwcd4 pass by
2021-04-07 20:46:59 +0800
b1baee60
feat(imperative/utils): add optimize-for-inference interface for opgraph by
2021-04-07 12:35:55 +0800
86598c82
fix(mge/utils): filter out parameter "arg_names" and "output_name" in network.dump by
2021-04-07 12:53:11 +0800
9dbe71dd
feat(mge/utils): add array method for varnode by
2021-04-04 12:50:38 +0800
9b0bd695
fix(imperative/utils): fix name filter of network varnode by
2021-04-09 18:23:55 +0800
af62cf6f
(tag: v1.3.1)
fix(mge/utils): fix auto naming bug when expanding structure by
2021-04-09 18:00:16 +0800
58ebb261
fix(imperative/tensor): fix ConstTensorCache by
2021-04-07 14:32:32 +0800
756c1eb7
fix(mgb/dnn): add cuda float naive matmul algo by
2020-10-13 17:49:33 +0800
4f0e6eae
fix(mge/trace): re-open gopt level in trace by
2021-03-29 15:22:08 +0800
8494a152
chore(scripts): clarify and fix default value of bit combined enum by
2021-04-02 15:20:33 +0800
da167cbc
fix(mge/utils): fix module stats calculate flops bug for group conv and remove model status change by
2021-04-08 17:53:44 +0800
6bb9a255
fix(mge/tools): fix module stats' receptive field bug for Module by
2021-04-08 15:54:54 +0800
acf28603
fix(cmake): export more sym caused by we change exception namespace when disable rtti by
2021-04-08 18:38:37 +0800
44bafd3f
fix(imperative/quantization): fix zero scale bug of easy quant by
2021-04-02 18:17:46 +0800
8aced67b
chore(release): bump version by
2021-04-08 18:34:29 +0800
e474994f
feat(imperative/jit): catch input tensors name when tracing by
2021-04-02 16:59:58 +0800
aed681d3
feat(imperative/utils): optimize the naming rules by
2021-03-31 18:04:30 +0800
c6bbc478
feat(imperative/utils): add __repr__ for network node by
2021-03-31 18:08:12 +0800
04b1a45a
fix(dnn): fix cudnn crash when finalize called after cudnn dtor by
2021-04-06 14:00:36 +0800
14a089c4
fix(dnn): change ci to cudnn804, reopen testcase by
2021-04-06 14:00:36 +0800
c338e876
refactor(mgb/dnn): add negative attribute for algo by
2021-03-31 16:47:14 +0800
ec1a99ac
refactor(mgb/dnn): replace reproducible with attribute by
2021-03-04 11:05:08 +0800
6af0299c
test(mge/ioremote): add REQUIRE_GPU for ioremote opr test by
2021-04-01 14:31:20 +0800
bab0caab
fix(mgb/opr): add layout constraint on the input of Cumsum by
2021-03-31 17:08:12 +0800
0d165399
fix(mgb): fix fastrun for imperative by
2021-04-01 21:01:30 +0800
94401ce4
chore(dotprod): dotprod is enabled by default on the android platform by
2021-03-10 11:23:49 +0800
610fecc9
feat(cmake): fix android cmake build issue detail: NDK do not support find_package(Threads) but NDK support threads! by
2021-04-01 23:36:07 +0800
2df84754
fix(mge/tools): fix node display bug in tensorboard by
2021-03-31 17:32:24 +0800
13481fd2
feat(mge/tools): add support of receptive_field stats for NetworkNode by
2021-03-26 19:15:25 +0800
5a7c30e0
feat(mge/tools): add summary print for module_stats and network_visualize by
2021-03-26 16:21:18 +0800
31035391
feat(mge/tools): set network_visualize's log_path as optional flag by
2021-03-23 16:57:26 +0800
eeeddbbc
refactor(imperative): refactor tablegen code generator by
2021-03-29 17:59:53 +0800
0ad85a41
test(mge): do not ignore data test by
2021-03-23 17:20:07 +0800
85b41a90
feat(dnn): add checksum opr and test by
2021-01-27 14:29:53 +0800
a49f4a66
feat(dnn): add indexing_one_hot and indexing_set_one_hot opr by
2020-12-22 13:33:24 +0800
1310ad49
feat(dnn): add memory manager for enflame by
2020-09-16 15:19:57 +0800
2fd3fa88
feat(cmake): update for enflame cmake compile by
2020-10-26 16:25:20 +0800
9f2af209
feat(mgb): add enflame comp node by
2020-09-24 14:45:18 +0800
15d3b3b9
fix(mgb): fix mgb still profiling matmul even when no-profiling-on-shape-change by
2021-03-22 18:02:39 +0800
31e4bf2c
fix(mge/optimizer): fix optimizer's state_dict bug by
2021-03-31 12:01:56 +0800
dc96f6aa
fix(mge/quantization): fix quantized concat forward problem by
2021-03-30 11:16:03 +0800
33da8de1
build(dnn/cuda): split compilation for cutlass wrapper by
2021-03-18 13:25:39 +0800
63cc5743
fix(windows): remove windows build path warning by
2021-03-23 13:26:08 +0800
48a7d333
fix(mge/module): fix auto-naming error when there are containers in the module by
2021-03-24 11:34:21 +0800
68f2e597
fix(mgb(ci)): fix tx1 ci testcase by
2021-03-26 19:41:22 +0800
420672be
fix(mgb/dnn): fix x86 matmul midout decl by
2021-03-18 18:41:32 +0800
988f897e
test(mge): make test stable by
2021-03-26 14:13:32 +0800
bcbfbbd7
fix(mgb): fix TensorRT runtime opr profiling by
2021-03-17 18:05:12 +0800
702ed9ee
feat(mge): enable cg dump for debugging by
2021-01-29 10:27:58 +0800
ef18a640
fix(mgb/core): fix allocate_task_block_unsafe in thread_impl_1.h by
2021-03-24 15:10:50 +0800
fa1ca0ea
fix(imperative/opr): fix apply_on_var_node for broadcast by
2021-03-25 15:11:52 +0800
e1c83d8d
fix(mgb/core): add warning information about const_var_shape when record mode by
2021-03-05 13:29:24 +0800
fc8b501b
refactor(mgb/core): refactor cpu compnode so that default cpu has no ability to record by
2021-02-26 10:37:29 +0800
b7176069
fix(dnn/cuda): add block size limit for culass gemm algo by
2021-03-24 15:24:21 +0800
55974e8c
feat(log): opt log * opt log at release mode * add MGE_OVERRIDE_LOG_LEVEL for runtime debug //! env to config LogLevel //! DEBUG = 0, INFO = 1, WARN = 2, ERROR = 3, NO_LOG = 4 //! for example , export MGE_OVERRIDE_LOG_LEVEL=0, means set LogLevel to DEBUG by
2021-03-19 11:15:05 +0800
c85eefde
fix(tree): fix copybara by
2021-03-22 19:16:09 +0800
26a81b89
(tag: v1.3.0)
chore(release): bump version by
2021-03-24 22:29:37 +0800
37c3e0a9
chore(mge): remove old version.py and use version_template by
2021-03-23 19:37:39 +0800
efe264b7
test(mge/utils): skip network node convbias test in gpu by
2021-03-22 11:43:28 +0800
c6069d02
chore(scripts): remove manylinux2010 build enviroment by
2021-03-22 18:53:48 +0800
b730547b
chore(third_party): update upstream of cutlass by
2021-03-21 18:53:03 +0800
e9982d61
chore(mge/functional): add compatible code for functional api by
2021-03-17 16:14:37 +0800
8928c77c
fix(neuropilot): fix loader header mismatch by
2021-03-17 17:43:29 +0800
221ec38a
feat(imperative): reduce profiler overhead by
2021-03-18 13:40:35 +0800
1c01128f
fix SASS cu111 by
2021-03-19 04:21:14 +0800