def _validate_config_before_canonical(self, model: Module, config_list: List[Dict]):
schema_list = [deepcopy(NORMAL_SCHEMA), deepcopy(INTERNAL_SCHEMA)]
for schema in schema_list:
schema.update({SchemaOptional('rho'): And(float, lambda n: n > 0)})
schema_list.append(deepcopy(EXCLUDE_SCHEMA))
schema = CompressorSchema(schema_list, model, _logger)
schema.validate(config_list)
def _validate_config_before_canonical(self, model: Module, config_list: List[Dict]):
schema_list = [deepcopy(NORMAL_SCHEMA), deepcopy(EXCLUDE_SCHEMA), deepcopy(INTERNAL_SCHEMA)]
for sub_shcema in schema_list:
sub_shcema[SchemaOptional('op_types')] = ['Conv2d', 'Linear']
schema = CompressorSchema(schema_list, model, _logger)
schema.validate(config_list)
def _validate_config_before_canonical(self, model: Module,
config_list: List[Dict]):
schema_list = [
deepcopy(NORMAL_SCHEMA),
deepcopy(EXCLUDE_SCHEMA),
deepcopy(INTERNAL_SCHEMA)
]
schema = CompressorSchema(schema_list, model, _logger)
schema.validate(config_list)
def _validate_config_before_canonical(self, model: Module, config_list: List[Dict]):
schema_list = [deepcopy(EXCLUDE_SCHEMA), deepcopy(INTERNAL_SCHEMA)]
if self.mode == 'global':
schema_list.append(deepcopy(GLOBAL_SCHEMA))
else:
schema_list.append(deepcopy(NORMAL_SCHEMA))
for sub_shcema in schema_list:
sub_shcema[SchemaOptional('op_types')] = ['BatchNorm2d']
schema = CompressorSchema(schema_list, model, _logger)
schema.validate(config_list)
def _validate_config_before_canonical(self, model: Module, config_list: List[Dict]):
schema_list = [deepcopy(EXCLUDE_SCHEMA), deepcopy(INTERNAL_SCHEMA)]
if self.mode == 'global':
schema_list.append(deepcopy(GLOBAL_SCHEMA))
else:
schema_list.append(deepcopy(NORMAL_SCHEMA))
for sub_shcema in schema_list:
sub_shcema[SchemaOptional('op_types')] = ['Conv2d', 'Linear']
schema = CompressorSchema(schema_list, model, _logger)
try:
schema.validate(config_list)
except SchemaError as e:
if "Missing key: 'total_sparsity'" in str(e):
_logger.error('`config_list` validation failed. If global mode is set in this pruner, `sparsity_per_layer` and `sparsity` are not supported, make sure `total_sparsity` is set in config_list.')
raise e