wangwei
/
MegEngine
Not watched
1
0
Fork 0
Code
Releases 31 Wiki Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain

Commit Graph

  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • | *
  • |/
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • *
  • 3bf73ff1 feat(dnn): add cuda preprocess fusion by Megvii Engine Team 2020-12-01 17:26:17 +0800
  • 86cf7490 feat(dnn/aarch64): add quantizeds4 matmul int4x4x16_k8x8x8 by Megvii Engine Team 2020-09-16 17:18:58 +0800
  • bff0fc61 fix(mge/interpreter): fix outputs check on async level0 by Megvii Engine Team 2020-12-07 15:04:09 +0800
  • 142f31a8 perf(dnn/cuda): change conv_bias heu, prefer dnn chanwise impl, dislike dnn batch gemm conv1x1 by Megvii Engine Team 2020-12-07 17:15:55 +0800
  • f214e146 refactor(mgb/cuda): use single implementation of get_device_prop from utils by Megvii Engine Team 2020-12-07 17:30:05 +0800
  • 54e79dd1 perf(mgb/cuda): do not call cudaGetDeviceProperties to avoid io traffic by Megvii Engine Team 2020-12-06 17:01:40 +0800
  • 5f171298 feat(mgb/gopt): add AxisAddRemove opr support for cd4 opt pass by Megvii Engine Team 2020-12-04 14:34:36 +0800
  • 93f4977c feat(mge/imperative): add thread name by Megvii Engine Team 2020-12-04 15:45:41 +0800
  • 98a74e4a refactor(dnn): refactor opr proxy in test by Megvii Engine Team 2020-12-02 22:04:29 +0800
  • 57546b4c test(mge/distributed): fix test skip condition error by Megvii Engine Team 2020-12-04 12:25:08 +0800
  • 90e7cb00 feat(externcopr/lar): imp lar run extern c opr with dynamic param by Megvii Engine Team 2020-11-26 20:20:52 +0800
  • dbb64b46 feat(debug/android): opt android backtrace by Megvii Engine Team 2020-12-03 15:09:59 +0800
  • 3e00e3f6 feat(debug/linux): opt linux backtrace by Megvii Engine Team 2020-12-03 09:04:14 +0800
  • 783a6126 feat(debug/macos/windows): imp macos/windows backtrace, fix mem issue by Megvii Engine Team 2020-11-30 19:21:18 +0800
  • e92670e8 fix(mgb/atlas): when batchsize more than model max batchsize by Megvii Engine Team 2020-12-02 17:32:18 +0800
  • 147dbf8a fix(test): fix a race condition in TestCudaMemAlloc by Megvii Engine Team 2020-12-02 15:25:37 +0800
  • 7066ad5b feat(dnn): add uint16 support by Megvii Engine Team 2020-11-23 14:45:41 +0800
  • a1877ee0 refactor(dnn): refactor algo interface, use algoinfo instead of global algorithm by Megvii Engine Team 2020-11-16 23:08:37 +0800
  • cb59c278 feat(mlir/ir): add more op definitions by Megvii Engine Team 2020-11-27 18:06:08 +0800
  • 9ec8d375 feat(externcopr): add config extern c opr dynamic param by Megvii Engine Team 2020-11-21 18:50:05 +0800
  • ee4ea7fd test(distributed/test): make distributed test more stronger by Megvii Engine Team 2020-11-25 19:29:15 +0800
  • 3ecded74 refactor(distributed/server): use port 0 to get available port by Megvii Engine Team 2020-11-25 19:28:11 +0800
  • 88e918e2 feat(mgb/jit): add scf.ForOp in MgbToGpuLoweringPass by Megvii Engine Team 2020-11-29 17:18:36 +0800
  • 7aa54b0e feat(mge): enable memory swap and drop/recomputation by Megvii Engine Team 2020-11-09 00:40:06 +0800
  • 6f5d0feb perf(dnn/cuda): enhance performance for pooling forward by Megvii Engine Team 2020-11-25 16:51:32 +0800
  • 0560a218 chore(dnn/test): refactor megdnn arm_common test by Megvii Engine Team 2020-11-21 18:05:52 +0800
  • f7731bd4 fix(mgb/jit): fix a pointer bug in mlir executable_cuda by Megvii Engine Team 2020-11-26 17:08:52 +0800
  • 810d8cba fix(mgb/jit): add cmake target MLIRTosa for latest llvm-project by Megvii Engine Team 2020-11-25 17:17:57 +0800
  • 2ad8c5e1 fix(mge/io_remote): fix remote send/recv gradient at trace by Megvii Engine Team 2020-11-25 14:49:52 +0800
  • f470df4f fix(mgb/opr): fix convbias with no bias when weight preprocess by Megvii Engine Team 2020-11-25 18:46:06 +0800
  • 6c4841e8 fix(mge/quantization): `disable_fake_quant` does not work correctly by Megvii Engine Team 2020-11-23 13:50:26 +0800
  • aa953c3b fix(mge/module): fix missing import by Megvii Engine Team 2020-11-25 11:23:21 +0800
  • 5a01de78 fix(mge): fix scalar transpose by Megvii Engine Team 2020-11-24 17:55:46 +0800
  • 6b9ac894 fix(mgb/topk): fix topk grad by Megvii Engine Team 2020-11-20 20:38:50 +0800
  • 6856ce9c feat(dnn): support conv bias activation for nchw4 input tensor format and nchw output tensor format by Megvii Engine Team 2020-11-12 17:01:50 +0800
  • 85368643 chore(version): fix dev version to a large number by Megvii Engine Team 2020-11-23 11:02:33 +0800
  • 61c5c9cf chore(cmake): normlize some cmake message level by Megvii Engine Team 2020-10-20 11:48:30 +0800
  • c2255914 (tag: v1.1.0, release-1.1) chore(release): bump version by Megvii Engine Team 2020-11-23 14:40:51 +0800
  • 2e874208 fix(module): fix docs in normalization by Megvii Engine Team 2020-11-21 17:32:08 +0800
  • 638ab52f feat(mge/imperative): simulates scalar by Megvii Engine Team 2020-09-30 17:11:26 +0800
  • 7167fdbd feat(mge/module): add normalization module includes group_norm, instance_norm and layer_norm by Megvii Engine Team 2020-08-11 14:58:57 +0800
  • 94dba16f perf(mge/imperative): misc optimizations by Megvii Engine Team 2020-11-20 17:48:50 +0800
  • 9f139562 test(mge): fix requires-test.txt for doctest by Megvii Engine Team 2020-11-19 21:40:00 +0800
  • 442058ae feat(mge/quantize): add batch matmul activation module for inference by Megvii Engine Team 2020-11-16 17:56:27 +0800
  • 81e0dae4 docs(mge): update GradManger docs by Megvii Engine Team 2020-11-19 18:15:42 +0800
  • 66671006 feat(mge): use weakref for GradManger.attach by Megvii Engine Team 2020-11-19 18:14:45 +0800
  • 75ca5bfe feat(mge): remove GradManager.detach until it is ready by Megvii Engine Team 2020-11-17 16:25:49 +0800
  • 544b7983 revert: feat(mge/grad_manager): add `clear_grad` method for GradManager by Megvii Engine Team 2020-11-17 15:28:34 +0800
  • 176268d2 test(mge): disable gc in leak check by Megvii Engine Team 2020-11-20 21:35:13 +0800
  • 5d0f8da4 feat(mgb/jit): add Dimshuffle and lowering passes in jit mlir backend by Megvii Engine Team 2020-11-19 23:34:06 +0800
  • 0007b9e0 build(third_party): update llvm-project by Megvii Engine Team 2020-11-16 14:35:49 +0800
  • 404ef808 feat(mgb/jit): adapt jit mlir backend to new mgb dialect and add typecvt by Megvii Engine Team 2020-11-13 06:24:02 +0800
  • cc85047b fix(mge/trace): fix sublinear in trace by Megvii Engine Team 2020-11-19 12:34:21 +0800
  • b9918c32 feat(mge/distributed): support distributed key-value store by Megvii Engine Team 2020-11-18 14:13:40 +0800
  • ab82c8da feat(mge/device): add python binding for get_mem_status_bytes by Megvii Engine Team 2020-11-18 19:31:49 +0800
  • ab6328c5 feat(imperative): port persistent cache by Megvii Engine Team 2020-10-29 13:39:08 +0800
  • 60c6d59f feat(mbg/core): support bias preprocess in conv_bias by Megvii Engine Team 2020-11-13 15:12:38 +0800
  • ff8ef9ed docs(dnn): add comments of weight prerpocess interface by Megvii Engine Team 2020-11-11 15:27:57 +0800
  • 13272eaa fix(mge/trace): fix random op in symbolic trace by Megvii Engine Team 2020-11-18 21:48:26 +0800
  • 1fed5929 perf(mge/optimizer): close conver_inputs for optimizer step by Megvii Engine Team 2020-10-28 21:42:28 +0800
  • 1f75c7ad ci(midout): fix midout and reopen midout test we just test function, do not check size by Megvii Engine Team 2020-10-28 15:19:05 +0800
  • 49547295 fix(trace): link io-op to avoid deadlock by Megvii Engine Team 2020-11-18 11:39:59 +0800
  • 064c774f feat(imperative): impl hashable for SendRecv and add virtual input for Recv by Megvii Engine Team 2020-11-18 11:38:06 +0800
  • 798f7b3e feat(imperative): add virtual_deps op by Megvii Engine Team 2020-11-18 11:31:51 +0800
  • 1e71e0af refactor(dnn): refactor deconv algo by Megvii Engine Team 2020-11-17 00:52:11 +0800
  • 89ad33ae feat(dnn/cuda): support weight preprocessing for cutlass algorithms by Megvii Engine Team 2020-11-04 10:41:54 +0800
  • 33e8879a feat(mge/quantization): support distributed qat by Megvii Engine Team 2020-11-16 15:21:26 +0800
  • 8e11204a perf(mge/module): optimize conv_bn qat module to improve performance by Megvii Engine Team 2020-11-17 11:22:34 +0800
  • 51fa530d fix(mge/interpreter): add check for invalid tensor ptr by Megvii Engine Team 2020-11-10 15:16:16 +0800
  • 634de590 feat(mge/imperative): add valid flag of `infer_output_attrs_fallible` by Megvii Engine Team 2020-11-06 18:27:43 +0800
  • 50c4daac feat(mge/interpreter): add async_level mechanism for Interpreter by Megvii Engine Team 2020-11-02 17:41:20 +0800
  • 82b0f677 fix(mge/core): fix dtype promotion issue for quantized dtype by Megvii Engine Team 2020-11-16 11:41:11 +0800
  • 8118a594 fix(mge/utils): fix get_oprs_seq of cgtools by Megvii Engine Team 2020-11-13 18:09:59 +0800
  • ae8c3c81 feat(mge/functional): add python wrapper for fake quant opr by Megvii Engine Team 2020-11-10 18:25:23 +0800
  • b60cc8ca feat(mgb/opr): add megbrain fake quant opr by Megvii Engine Team 2020-11-10 18:23:14 +0800
  • c03249c0 feat(dnn/opr): add megdnn fake quant opr by Megvii Engine Team 2020-11-10 18:22:16 +0800
  • 2e530779 fix(mge/trace): use xpux device when dump by Megvii Engine Team 2020-11-11 11:52:32 +0800
  • 739f927c feat(dnn/cuda): opt dp4a conv for small channel base on cutlass by Megvii Engine Team 2020-11-09 18:01:02 +0800
  • 4f9948d0 chore(mbg/core): add MGB_WORKER_SHORT_SPIN env variable to set short spin by Megvii Engine Team 2020-11-11 17:47:57 +0800
  • 4f527af4 chore(mbg/core): fix the repeat code by Megvii Engine Team 2020-11-11 17:45:18 +0800
  • 1f8e4075 fix(mkl): fix windows mkl LOG compute exception by Megvii Engine Team 2020-11-13 12:43:58 +0800
  • fea5cf62 fix(sdk/loader): debug sstar loader by Megvii Engine Team 2020-10-14 22:26:30 +0800
  • 9cb3c07c feat(mge/functional): add elemwise mode support string input by Megvii Engine Team 2020-11-12 11:26:17 +0800
  • 8fad00a1 feat(mge/functional/nn): add conv1d padding by Megvii Engine Team 2020-10-21 22:45:41 +0800
  • 4aa277a2 refactor(dnn/cuda): misc by Megvii Engine Team 2020-11-06 17:47:01 +0800
  • 094601e8 feat(mge/distributed): allow remote grad by using grad manager by Megvii Engine Team 2020-11-12 14:45:05 +0800
  • f7b2bdae refactor(dnn): refactor algorithm type interface by Megvii Engine Team 2020-11-07 21:34:09 +0800
  • d793c87c refactor(mlir/dialect): redefine mgb dialect by Megvii Engine Team 2020-11-11 16:27:57 +0800
  • 005ead5a docs(cpu/comp_node): note cpu_default() by Megvii Engine Team 2020-11-10 14:45:04 +0800
  • 9415ba58 feat(src/core): free weight preprocessed weight by Megvii Engine Team 2020-10-12 21:17:56 +0800
  • 7cd71c31 fix(mgb/gopt): fix cd4 elewise transform by Megvii Engine Team 2020-11-03 20:40:50 +0800
  • cae8c8a4 test(mge/parampack): add test for parampack by Megvii Engine Team 2020-11-10 12:54:00 +0800
  • 02438ee6 refactor(mge/distributed): use thread.Threading to create Server by Megvii Engine Team 2020-11-09 15:49:37 +0800
  • a1efd4d5 fix(mge/parampack): fix param pack when no param left by Megvii Engine Team 2020-11-09 15:33:52 +0800
  • e0e18964 docs(mge): formatting math in docstring by Megvii Engine Team 2020-11-09 12:10:32 +0800
  • 18ec5341 refactor(dnn): remove unused costmodel in cuda by Megvii Engine Team 2020-11-10 23:12:13 +0800
  • e39f9386 refactor(dnn): remove ProfileCache and matmul algo in x86 by Megvii Engine Team 2020-11-10 22:20:27 +0800
  • 28dbadf7 feat(imperative): add remap opr by Megvii Engine Team 2020-09-29 15:13:51 +0800
  • 280861ae feat(atlas): add acl.json support, use for profile by Megvii Engine Team 2020-08-05 15:16:54 +0800
  • 1e0fb127 perf(mge/functional): reduce matmul python overhead by Megvii Engine Team 2020-11-09 00:09:51 +0800