qubee.configs.compile_config.CompileConfig Class Reference#
|
Mobilint SDK qb Compiler v0.12.0.0
MCS002-KR
|
Public Member Functions | |
| __init__ (self, str|None calib_data_path, bool use_random_calib, str save_path, str model_nickname, str|None save_msgpack_name, str|None model_path, str inference_scheme, bool save_sample, bool cpu_offload, bool singlecore_compile, int optimize_option, str sample_dtype, dict|None preprocess_dict, int buffer_mode, int start_dram_offset, str version, dict input_shape_dict, ResourceManagementConfig resource_management, QuantizationConfig quantization, AdvancedQuantizationConfig advanced_quantization, LlmConfig llm_config, bool optional=optional) | |
| Unified compilation configuration object for Mobilint MXQ compilation. | |
| default_config (cls, **kwargs) | |
| Create default config with optional parameter overrides. | |
| from_json (cls, str json_path, **kwargs) | |
| Load config from JSON file and merge with additional kwargs. | |
| check_valid_str_configs (self) | |
| to_dict (self) | |
Public Attributes | |
| calib_data_path | |
| save_path = save_path | |
| model_nickname = model_nickname | |
| save_msgpack_name = save_msgpack_name | |
| model_path = model_path | |
| inference_scheme = inference_scheme | |
| save_sample = save_sample | |
| cpu_offload = cpu_offload | |
| singlecore_compile = singlecore_compile | |
| optimize_option = optimize_option | |
| sample_dtype = sample_dtype | |
| preprocess_dict = preprocess_dict | |
| buffer_mode = buffer_mode | |
| start_dram_offset = start_dram_offset | |
| version = version | |
| input_shape_dict = input_shape_dict | |
| resource_management = resource_management | |
| quantization = quantization | |
| advanced_quantization = advanced_quantization | |
| llm_config = llm_config | |
| calib_file = tempfile.NamedTemporaryFile(suffix=".txt") | |
Static Public Attributes | |
| bool | optional = False |
| dict | py_to_cpp_name |
| dict | min_max_check = {"optimize_option": [0, None]} |
| list | inference_scheme_types = ["single", "multi", "global", "global4", "global8"] |
| list | sample_dtypes = ["float", "int8"] |
Protected Member Functions | |
| _get_calibration_path (self, str calib_data_path, bool use_random_calib) | |
Detailed Description
Definition at line 61 of file compile_config.py.
Constructor & Destructor Documentation
◆ __init__()
| qubee.configs.compile_config.CompileConfig.__init__ | ( | self, | |
| str | None | calib_data_path, | ||
| bool | use_random_calib, | ||
| str | save_path, | ||
| str | model_nickname, | ||
| str | None | save_msgpack_name, | ||
| str | None | model_path, | ||
| str | inference_scheme, | ||
| bool | save_sample, | ||
| bool | cpu_offload, | ||
| bool | singlecore_compile, | ||
| int | optimize_option, | ||
| str | sample_dtype, | ||
| dict | None | preprocess_dict, | ||
| int | buffer_mode, | ||
| int | start_dram_offset, | ||
| str | version, | ||
| dict | input_shape_dict, | ||
| ResourceManagementConfig | resource_management, | ||
| QuantizationConfig | quantization, | ||
| AdvancedQuantizationConfig | advanced_quantization, | ||
| LlmConfig | llm_config, | ||
| bool | optional = optional ) |
Unified compilation configuration object for Mobilint MXQ compilation.
- Parameters
-
calib_data_path str | None. Path to calibration dataset. Accepts either:
- a directory containing pre-processed NumPy samples, or
- a .txt/.json meta file listing NumPy sample file paths.
Ifuse_random_calibis True, this may be None. Otherwise, a valid path is expected for quantized compilation workflows.
use_random_calib bool. When True, uses randomly generated calibration samples (typically for smoke testing and compilability validation). When False, calib_data_pathis used as the calibration source.save_path str. Output MXQ filename/path to generate. model_nickname str. Model nickname used as a fallback identifier when intermediate results need to be saved. It is recommended to set this to the filename component of save_path.save_msgpack_name str | None. filename/path to save an intermediate serialized representation (msgpack) of the high-level compiled model/graph. Primarily intended for debugging and is now rarely used. model_path str | None. (deprecated) Outdated argument. Please refer to the model usage in the mxq_compilefunction.inference_scheme str. NPU inference scheme string. Must be one of:
"single","multi","global","global4","global8"("global"and"global8"behave identically).save_sample bool. When True, saves representative input/output tensors for debugging. cpu_offload bool. Enables CPU offload for unsupported operator groups/segments during NPU inference. This is a beta feature and is recommended only for special cases. singlecore_compile bool. Forces single-core compilation mode. Commonly used to compile large language models. optimize_option int. Compiler optimization selector. sample_dtype string. Data type for saved sample inference outputs: "float" or "int8". Primarily intended for debugging and is now rarely used. preprocess_dict dict | None. Additional preprocessing metadata. buffer_mode int. Buffer serialization mode for saving .mblt file. Common conventions:
- 0: naïve buffer serialization
- 1: mmap-backed buffer serialization
start_dram_offset int. Start Dram Address Offset (default = 0). version str. qubee version. input_shape_dict dict. Optional HWC input shape dictionary used for multi-shape compilation (for example {"input0": [[224, 224, 3], [256, 256, 3]]}). resource_management ResourceManagementConfig. Resource management configuration class. quantization QuantizationConfig. Quantization configuration class (e.g., calibration strategy, bit-widths, per-tensor/per-channel settings, quantization scheme/policies). advanced_quantization AdvancedQuantizationConfig. Advanced quantization configuration block for fine-grained or experimental controls beyond the basic quantization settings. llm_config LlmConfig. LLM-specific configuration class. optional bool. Indicates whether this config is optional. Use the default value defined as a class variable. - a directory containing pre-processed NumPy samples, or
Definition at line 89 of file compile_config.py.
Member Function Documentation
◆ default_config()
| qubee.configs.compile_config.CompileConfig.default_config | ( | cls, | |
| ** | kwargs ) |
Create default config with optional parameter overrides.
Args: **kwargs: Parameters to override defaults
Definition at line 206 of file compile_config.py.
◆ from_json()
| qubee.configs.compile_config.CompileConfig.from_json | ( | cls, | |
| str | json_path, | ||
| ** | kwargs ) |
Load config from JSON file and merge with additional kwargs.
Args: json_path: Path to JSON configuration file **kwargs: Additional compile-time parameters (calib_data_path, save_path, etc.)
Definition at line 244 of file compile_config.py.
◆ check_valid_str_configs()
| qubee.configs.compile_config.CompileConfig.check_valid_str_configs | ( | self | ) |
Definition at line 276 of file compile_config.py.
◆ to_dict()
| qubee.configs.compile_config.CompileConfig.to_dict | ( | self | ) |
Definition at line 292 of file compile_config.py.
◆ _get_calibration_path()
|
protected |
Definition at line 313 of file compile_config.py.
Member Data Documentation
◆ optional
|
static |
Definition at line 62 of file compile_config.py.
◆ py_to_cpp_name
|
static |
Definition at line 63 of file compile_config.py.
◆ min_max_check
|
static |
Definition at line 79 of file compile_config.py.
◆ inference_scheme_types
|
static |
Definition at line 80 of file compile_config.py.
◆ sample_dtypes
|
static |
Definition at line 81 of file compile_config.py.
◆ calib_data_path
| qubee.configs.compile_config.CompileConfig.calib_data_path |
Definition at line 167 of file compile_config.py.
◆ save_path
| qubee.configs.compile_config.CompileConfig.save_path = save_path |
Definition at line 170 of file compile_config.py.
◆ model_nickname
| qubee.configs.compile_config.CompileConfig.model_nickname = model_nickname |
Definition at line 171 of file compile_config.py.
◆ save_msgpack_name
| qubee.configs.compile_config.CompileConfig.save_msgpack_name = save_msgpack_name |
Definition at line 172 of file compile_config.py.
◆ model_path
| qubee.configs.compile_config.CompileConfig.model_path = model_path |
Definition at line 173 of file compile_config.py.
◆ inference_scheme
| qubee.configs.compile_config.CompileConfig.inference_scheme = inference_scheme |
Definition at line 174 of file compile_config.py.
◆ save_sample
| qubee.configs.compile_config.CompileConfig.save_sample = save_sample |
Definition at line 175 of file compile_config.py.
◆ cpu_offload
| qubee.configs.compile_config.CompileConfig.cpu_offload = cpu_offload |
Definition at line 176 of file compile_config.py.
◆ singlecore_compile
| qubee.configs.compile_config.CompileConfig.singlecore_compile = singlecore_compile |
Definition at line 177 of file compile_config.py.
◆ optimize_option
| qubee.configs.compile_config.CompileConfig.optimize_option = optimize_option |
Definition at line 178 of file compile_config.py.
◆ sample_dtype
| qubee.configs.compile_config.CompileConfig.sample_dtype = sample_dtype |
Definition at line 179 of file compile_config.py.
◆ preprocess_dict
| qubee.configs.compile_config.CompileConfig.preprocess_dict = preprocess_dict |
Definition at line 180 of file compile_config.py.
◆ buffer_mode
| qubee.configs.compile_config.CompileConfig.buffer_mode = buffer_mode |
Definition at line 181 of file compile_config.py.
◆ start_dram_offset
| qubee.configs.compile_config.CompileConfig.start_dram_offset = start_dram_offset |
Definition at line 182 of file compile_config.py.
◆ version
| qubee.configs.compile_config.CompileConfig.version = version |
Definition at line 183 of file compile_config.py.
◆ input_shape_dict
| qubee.configs.compile_config.CompileConfig.input_shape_dict = input_shape_dict |
Definition at line 184 of file compile_config.py.
◆ resource_management
| qubee.configs.compile_config.CompileConfig.resource_management = resource_management |
Definition at line 187 of file compile_config.py.
◆ quantization
| qubee.configs.compile_config.CompileConfig.quantization = quantization |
Definition at line 188 of file compile_config.py.
◆ advanced_quantization
| qubee.configs.compile_config.CompileConfig.advanced_quantization = advanced_quantization |
Definition at line 189 of file compile_config.py.
◆ llm_config
| qubee.configs.compile_config.CompileConfig.llm_config = llm_config |
Definition at line 190 of file compile_config.py.
◆ calib_file
| qubee.configs.compile_config.CompileConfig.calib_file = tempfile.NamedTemporaryFile(suffix=".txt") |
Definition at line 324 of file compile_config.py.
The documentation for this class was generated from the following file:
Generated by