enum akhir statis publik RewriterConfig.MemOptType
Protobuf termasuk tensorflow.RewriterConfig.MemOptType
Konstanta
ke dalam | DEFAULT_MEM_OPT_VALUE | The default setting (SCHEDULING and SWAPPING HEURISTICS only) DEFAULT_MEM_OPT = 0; |
ke dalam | HEURISTICS_VALUE | Use any combination of swapping and recomputation heuristics. |
ke dalam | MANUAL_VALUE | Driven by manual op-level annotations. |
ke dalam | NO_MEM_OPT_VALUE | Disabled in the meta-optimizer. |
ke dalam | RECOMPUTATION_HEURISTICS_VALUE | Recomputation heuristics will recompute ops (such as Relu activation) during backprop instead of storing them, reducing peak memory usage. |
ke dalam | SCHEDULING_HEURISTICS_VALUE | Scheduling will split big ops such as AddN and try to enforce a schedule of the new computations that decreases peak memory usage. |
ke dalam | SWAPPING_HEURISTICS_VALUE | Swapping heuristic will move a tensor from the GPU to the CPU and move it back when needed to reduce peak memory usage. |
Metode Warisan
Nilai Enum
RewriterConfig.MemOptType DEFAULT_MEM_OPT final statis publik
The default setting (SCHEDULING and SWAPPING HEURISTICS only)
DEFAULT_MEM_OPT = 0;
HEURISTICS RewriterConfig.MemOptType final statis publik
Use any combination of swapping and recomputation heuristics.
HEURISTICS = 3;
MANUAL RewriterConfig.MemOptType final publik statis
Driven by manual op-level annotations.
MANUAL = 2;
RewriterConfig.MemOptType NO_MEM_OPT final statis publik
Disabled in the meta-optimizer.
NO_MEM_OPT = 1;
RewriterConfig.MemOptType RECOMPUTATION_HEURISTICS final statis publik
Recomputation heuristics will recompute ops (such as Relu activation) during backprop instead of storing them, reducing peak memory usage.
RECOMPUTATION_HEURISTICS = 5;
RewriterConfig.MemOptType SCHEDULING_HEURISTICS final statis publik
Scheduling will split big ops such as AddN and try to enforce a schedule of the new computations that decreases peak memory usage.
SCHEDULING_HEURISTICS = 6;
RewriterConfig.MemOptType SWAPPING_HEURISTICS final statis publik
Swapping heuristic will move a tensor from the GPU to the CPU and move it back when needed to reduce peak memory usage.
SWAPPING_HEURISTICS = 4;
RewriterConfig.MemOptType final statis publik TIDAK DIAKUI
Konstanta
int final statis publik DEFAULT_MEM_OPT_VALUE
The default setting (SCHEDULING and SWAPPING HEURISTICS only)
DEFAULT_MEM_OPT = 0;
Nilai Konstan: 0
int akhir statis publik HEURISTICS_VALUE
Use any combination of swapping and recomputation heuristics.
HEURISTICS = 3;
Nilai Konstan: 3
int akhir statis publik MANUAL_VALUE
Driven by manual op-level annotations.
MANUAL = 2;
Nilai Konstan: 2
int akhir statis publik NO_MEM_OPT_VALUE
Disabled in the meta-optimizer.
NO_MEM_OPT = 1;
Nilai Konstan: 1
int akhir statis publik RECOMPUTATION_HEURISTICS_VALUE
Recomputation heuristics will recompute ops (such as Relu activation) during backprop instead of storing them, reducing peak memory usage.
RECOMPUTATION_HEURISTICS = 5;
Nilai Konstan: 5
int akhir statis publik SCHEDULING_HEURISTICS_VALUE
Scheduling will split big ops such as AddN and try to enforce a schedule of the new computations that decreases peak memory usage.
SCHEDULING_HEURISTICS = 6;
Nilai Konstan: 6
int akhir statis publik SWAPPING_HEURISTICS_VALUE
Swapping heuristic will move a tensor from the GPU to the CPU and move it back when needed to reduce peak memory usage.
SWAPPING_HEURISTICS = 4;
Nilai Konstan: 4