168 Commits (2e320079d35e140d2c9b01f859bf386fd3cf9304)

Author	SHA1	Message	Date
yi.wu	8d04d0e2a3	update	7 years ago
yi.wu	6f0107126a	fix broadcast bug	7 years ago
yi.wu	8e48c77b54	wip	7 years ago
yi.wu	3d69a82b83	fix dist train broadcasting bug	7 years ago
fengjiayi	964f515e9a	fix mac compile	7 years ago
Yancey1989	7e6518e8ca	fix compile warning	7 years ago
Yancey1989	7d1b146939	Merge branch 'develop' of github.com:PaddlePaddle/Paddle into overlap_memcpy_with_dist	7 years ago
Qiyang Min	046bb5c8cb	Fix NCCLBcast hang up bug in Parallel Executor (#11377) ... * 1. Create buddy allocator in each places before NcclBcast the variables 2. Check the memory usage of ALL gpus rather than the first one * 1. Make NCCLGroupGuard guards only the ncclBcast part, which avoid ncclGroupEnd blocking the exception throwing 2. NOTE the usage of NCCLGroupGuard * Remove the memory usage check of gpus * Fix code style	7 years ago
Yancey1989	6d752bafd8	use get_appropriate_dev to schedule rpc op	7 years ago
Yancey1989	4444e79e46	Merge branch 'develop' of github.com:PaddlePaddle/Paddle into overlap_memcpy_with_dist	7 years ago
chengduoZH	173d72b481	Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into enable_cpu_on_pe	7 years ago
chengduoZH	aadaadf735	replace use_event with use_cuda, because use_event means the program running with CUDA, so use_cuda maybe more intuitive.	7 years ago
chengduoZH	1e731f5964	small fix	7 years ago
chengduoZH	5a3c8bf813	fix in c++ side	7 years ago
chengduoZH	0c851cab22	add SSA graph checker	7 years ago
Yancey1989	d5a88b9340	Merge branch 'develop' of github.com:PaddlePaddle/Paddle into overlap_memcpy_with_dist	7 years ago
chengduoZH	8291b916d6	replace graph_builder_factory with ssa_graph_builder_factory	7 years ago
Yancey1989	23433def4b	Merge branch 'develop' of github.com:PaddlePaddle/Paddle into overlap_memcpy_with_dist	7 years ago
yuyang18	d9af153232	SSA Graph Builder Factory ... * Use Builder Chain to decorate new builders. It is easy to extend builders. * Make graphviz path as a build strategy, not a FLAGS.	7 years ago
Yancey1989	e533a4b4ab	Merge branch 'develop' of github.com:PaddlePaddle/Paddle into overlap_memcpy_with_dist	7 years ago
Yancey1989	cb3861538d	fix compile failed with CPU	7 years ago
Yancey1989	93401c98e1	overlap rpc op memcpy in distributed training	7 years ago
yuyang18	86a61c177f	Add ScopeBufferedSSAGraphExecutor	7 years ago
yuyang18	7c777dd549	Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into feature/exec_strategy	7 years ago
yuyang18	08295f9877	Add build strategy	7 years ago
yuyang18	e5281b3c2d	Clean code & add execution strategy	7 years ago
typhoonzero	928418a9ac	Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into gen_nccl_id_op	7 years ago
typhoonzero	f5840d8925	follow comments	7 years ago
chengduoZH	97cb5479ae	change PE strategy	7 years ago
typhoonzero	a529d790b6	Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into gen_nccl_id_op	7 years ago
typhoonzero	d9320dcd94	complete code	7 years ago
chengduoZH	c891189568	update sparse gradient parameter with reduce and broadcast	7 years ago
chengduoZH	5ff1ef36ee	update sparse parameter	7 years ago
yangyaming	82571deb89	Change `customize_loss_grad` to `use_default_grad_scale`.	7 years ago
Yu Yang	54ada9449e	Add demo for recordio train/test and parallel executor	7 years ago
Yu Yang	7a395881d4	Add customize_loss_grad option to PE	7 years ago
Yu Yang	5305c5f845	Correctly implement destructor of ParallelExecutor	7 years ago
fengjiayi	fbe562478d	Merge pull request #9994 from reyoung/feature/debug ... Fix bugs in local_scopes	7 years ago
Yu Yang	06fb055a2f	New group	7 years ago
Yu Yang	71a2e6b73c	Reverse create var	7 years ago
Yu Yang	89728f8e66	update	7 years ago
Yu Yang	eb2e4eeade	Debug	7 years ago
Yu Yang	b4aaa00a8a	Polish logic of ParallelExecutor	7 years ago
Yu Yang	ad73b331c7	Eagerly drop local scope in iteration (#9838) ... * Eagerly drop local scope in iteration * Correct create var * Fix typo * Debug	7 years ago
fengjiayi	90084a25d2	Merge pull request #9743 from JiayiFeng/modify_readers_to_fit_parallel_executor ... Modify readers to fit the parallel executor	7 years ago
wanghaoshuang	19c1a68ee9	Fix lost of LoD while splitting tensor in parallel executor.	7 years ago
JiayiFeng	ee178d5aeb	fix bugs	7 years ago
chengduoZH	7e7611d067	when the number of samples of current batch is less than the count of devices, let it crash.	7 years ago
qingqing01	2b7e5bd366	Support testing during training by ParallelExecutor. (#9738) ... * Support testing during training by ParallelExecutor. * Add unit test. * Improve the interface. * Follow comments.	7 years ago
Xin Pan	4bbfa9eccb	Add feed to ParallelExecutor	7 years ago