RewriterConfig.MemOptType

public static final enum RewriterConfig.MemOptType

Protobuf enum tensorflow.RewriterConfig.MemOptType

Constants

int DEFAULT_MEM_OPT_VALUE
 The default setting (SCHEDULING and SWAPPING HEURISTICS only)
 
DEFAULT_MEM_OPT = 0;
int HEURISTICS_VALUE
 Use any combination of swapping and recomputation heuristics.
int MANUAL_VALUE
 Driven by manual op-level annotations.
int NO_MEM_OPT_VALUE
 Disabled in the meta-optimizer.
int RECOMPUTATION_HEURISTICS_VALUE
 Recomputation heuristics will recompute ops (such as Relu activation)
 during backprop instead of storing them, reducing peak memory usage.
int 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.
int 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.

Inherited Methods

Enum Values

public static final RewriterConfig.MemOptType DEFAULT_MEM_OPT

 The default setting (SCHEDULING and SWAPPING HEURISTICS only)
 
DEFAULT_MEM_OPT = 0;

public static final RewriterConfig.MemOptType HEURISTICS

 Use any combination of swapping and recomputation heuristics.
 
HEURISTICS = 3;

public static final RewriterConfig.MemOptType MANUAL

 Driven by manual op-level annotations.
 
MANUAL = 2;

public static final RewriterConfig.MemOptType NO_MEM_OPT

 Disabled in the meta-optimizer.
 
NO_MEM_OPT = 1;

public static final RewriterConfig.MemOptType RECOMPUTATION_HEURISTICS

 Recomputation heuristics will recompute ops (such as Relu activation)
 during backprop instead of storing them, reducing peak memory usage.
 
RECOMPUTATION_HEURISTICS = 5;

public static final RewriterConfig.MemOptType SCHEDULING_HEURISTICS

 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;

public static final RewriterConfig.MemOptType SWAPPING_HEURISTICS

 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;

public static final RewriterConfig.MemOptType UNRECOGNIZED

Constants

public static final int DEFAULT_MEM_OPT_VALUE

 The default setting (SCHEDULING and SWAPPING HEURISTICS only)
 
DEFAULT_MEM_OPT = 0;

Constant Value: 0

public static final int HEURISTICS_VALUE

 Use any combination of swapping and recomputation heuristics.
 
HEURISTICS = 3;

Constant Value: 3

public static final int MANUAL_VALUE

 Driven by manual op-level annotations.
 
MANUAL = 2;

Constant Value: 2

public static final int NO_MEM_OPT_VALUE

 Disabled in the meta-optimizer.
 
NO_MEM_OPT = 1;

Constant Value: 1

public static final int 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;

Constant Value: 5

public static final int 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;

Constant Value: 6

public static final int 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;

Constant Value: 4