Class MkldnnQuantizerConfig¶

Defined in File paddle_mkldnn_quantizer_config.h

Class Documentation¶

class paddle::MkldnnQuantizerConfig¶

Config for mkldnn quantize.

The MkldnnQuantizerConfig is used to configure Mkldnn’s quantization parameters, including scale algorithm, warmup data, warmup batch size, quantized op list, etc.

It is not recommended to use this config directly, please refer to AnalysisConfig::mkldnn_quantizer_config()

Public Functions

MkldnnQuantizerConfig()¶: Construct a new Mkldnn Quantizer Config object.

inline void SetScaleAlgo(std::string op_type_name, std::string conn_name, ScaleAlgo algo)¶

Set the scale algo.

Specify a quantization algorithm for a connection (input/output) of the operator type.

Parameters

op_type_name – [in] the operator’s name.
conn_name – [in] name of the connection (input/output) of the operator.
algo – [in] the algorithm for computing scale.

ScaleAlgo scale_algo(const std::string &op_type_name, const std::string &conn_name) const¶

Get the scale algo.

Get the quantization algorithm for a connection (input/output) of the operator type.

Parameters

op_type_name – [in] the operator’s name.
conn_name – [in] name of the connection (input/output) of the operator.

Returns

the scale algo.

inline void SetWarmupData(std::shared_ptr<std::vector<PaddleTensor>> data)¶

Set the warmup data.

Set the batch of data to be used for warm-up iteration.

Parameters: data – [in] batch of data.

inline std::shared_ptr<std::vector<PaddleTensor>> warmup_data() const¶

Get the warmup data.

Get the batch of data used for warm-up iteration.

Returns: the warm up data

inline void SetWarmupBatchSize(int batch_size)¶

Set the warmup batch size.

Set the batch size for warm-up iteration.

Parameters: batch_size – [in] warm-up batch size

inline int warmup_batch_size() const¶

Get the warmup batch size.

Get the batch size for warm-up iteration.

Returns: the warm up batch size

inline void SetEnabledOpTypes(std::unordered_set<std::string> op_list)¶

Set quantized op list.

In the quantization process, set the op list that supports quantization

Parameters: op_list – [in] List of quantized ops

inline const std::unordered_set<std::string> &enabled_op_types() const¶

Get quantized op list.

Returns: list of quantized ops

inline void SetExcludedOpIds(std::unordered_set<int> op_ids_list)¶

Set the excluded op ids.

Parameters: op_ids_list – [in] excluded op ids

inline const std::unordered_set<int> &excluded_op_ids() const¶

Get the excluded op ids.

Returns: exclude op ids

inline void SetDefaultScaleAlgo(ScaleAlgo algo)¶

Set default scale algorithm.

Parameters: algo – [in] Method for calculating scale in quantization process

inline ScaleAlgo default_scale_algo() const¶

Get default scale algorithm.

Returns: Method for calculating scale in quantization process

Protected Attributes

std::map<std::string, std::map<std::string, ScaleAlgo>> rules_¶

std::unordered_set<std::string> enabled_op_types_¶

std::unordered_set<int> excluded_op_ids_¶

std::shared_ptr<std::vector<PaddleTensor>> warmup_data_¶

int warmup_bs_ = {1}¶

ScaleAlgo default_scale_algo_ = {ScaleAlgo::MAX}¶