# qairt.api.common.backends.aic - *pydantic model* qairt.api.common.backends.aic.config.AicCompilerConfig - Compiler configuration for AIC Backend - *field* buffer\_dealloc\_delay*: int* *= 2* - Specifies the buffer deallocation delay. - *field* cluster\_sizes*: str* *= '1 2 0'* - Specifies the cluster configuration for single device partitioning. - *field* combine\_inputs*: bool* *= True* - Specifies whether to combine inputs into fewer buffers for transfer to device. - *field* combine\_outputs*: bool* *= True* - Specifies whether to combine outputs into a single buffer for transfer to host. - *field* compilation\_target*: str* *= 'hardware'* - Specifies the compilation target for the graph. - *field* compile\_threads*: int* *= 2* - Specifies the number of threads to use for compilation. - *field* convert\_to\_fp16*: bool* *= True* - Specifies whether to convert the graph from FP32 precision to FP16 precision inside AIC backend. - *field* crc\_stride*: int* *= 256* - Specifies the size of stride to calculate CRC in the stride section. - *field* direct\_api*: bool* *= True* - Specifies whether to use a platform-specific shared memory API. - *field* do\_ddr\_to\_multicast*: bool* *= True* - Specifies whether to reduce DDR bandwidth by loading weights used on multiple-cores only once and multicasting to other cores. - *field* do\_host\_preproc*: bool* *= True* - Specifies whether to perform host-side preprocessing. Setting compiler\_do\_host\_preproc to false disables all pre-/post-processing on the host. Otherwise, all all pre-/post-processing are performed on the host - *field* enable\_debug*: bool* *= True* - Specifies whether to enable debug mode during model compilation. - *field* enable\_depth\_first*: bool* *= True* - Specifies whether to enable depth-first compilation. - *field* exclude\_mxfp6\_matmul*: str* *= ''* - Specifies the list of operations to exclude from MXFP6 matmul. - *field* force\_vtcm\_spill*: bool* *= True* - Specifies whether to force all VTCM buffers to be spilled to DDR. - *field* gen\_crc*: bool* *= True* - Specifies whether to enable CRC check for inputs and outputs of the network. - *field* graph\_names*: List[str]* *[Optional]* - Specifies list of graph names. - *field* hardware\_version*: str* *= '2.0'* - Specifies the AI 100 hardware version to generate model binary. - *field* max\_out\_channel\_split*: str* *= '1 2 0'* - Specifies the effort level to reduce the on-chip memory usage. - *field* mdp\_dump\_partition\_config*: str* *= 'mdp\_pipeline\_template.json'* - Path to dump multi-device partition configuration template file. - *field* mdp\_load\_partition\_config*: str* *= 'mdp\_config.json'* - Path to load multi-device partition configuration file. - *field* mxfp6\_matmul\_weights*: bool* *= False* - Specifies whether to enable MXFP6 matmul weights. - *field* mxint8\_mdp\_io*: bool* *= False* - Specifies whether to enable MXINT8 MDP IO. - *field* num\_of\_cores*: int* *= 1* - Specifies the number of NSP Cores to use. - *field* overlap\_split\_factor*: int* *= 2* - Specifies the factor to increase splitting of network operations. - *field* perf\_warnings*: bool* *= True* - Specifies whether to print performance warning messages. - *field* pmu\_events*: str* *= '1 2 0'* - Specifies the PMU events to track. Up to 8 events are supported. - *field* pmu\_recipe\_opt*: Literal['AxiRd', 'AxiWr', 'AxiRdWr', 'KernelUtil']* *= 'KernelUtil'* - Specifies the PMU recipe optimization. - *field* print\_ddr\_stats*: bool* *= True* - Specifies whether to collect per core DDR traffic details. - *field* print\_perf\_metrics*: bool* *= True* - Specifies whether to print compiler performance metrics. - *field* retained\_state*: bool* *= False* - Specifies whether to enable compiler retained state. - *field* size\_split\_granularity*: int* *= 1024* - Specifies the maximum tile size. - *field* stat\_level*: int* *= 10* - Specifies the level of stats to be collected when stats collection is enabled with one of the profiling options. - *field* stats\_batch\_size*: int* *= 2* - Specifies the number of batches to be used for execution. Sets the batch dimension of input/output to the model for running inferences. - *field* time\_passes*: bool* *= True* - Specifies whether to enable the region profiler to collect compile-time statistics. - *field* user\_dma\_producer\_dma\_enabled*: bool* *= True* - Specifies whether to initiate NSP DMAs from the thread that produces data being transferred. - *field* vtcm\_working\_set\_limit\_ratio*: float* *= 1.0* - Specifies the ratio of fast memory an instruction can use. - *class* qairt.api.common.backends.aic.config.AicConfigHelper - Helper class to convert AIC config to backend extension config needed by QAIRT Tools. - *static* list\_config\_options() - - *classmethod* list\_config\_types() - - *static* shared\_library\_path() - - *static* to\_backend\_extension\_dict(*compiler\_config: Optional[[AicCompilerConfig](https://docs.qualcomm.com/doc/80-87189-2/topic/qairt-api-common-backends-aic.html#qairt.api.common.backends.aic.config.AicCompilerConfig)] = None*, *runtime\_config: Optional[[AicRuntimeConfig](https://docs.qualcomm.com/doc/80-87189-2/topic/qairt-api-common-backends-aic.html#qairt.api.common.backends.aic.config.AicRuntimeConfig)] = None*) → Dict[str, Any] - Builds the backend extension dictionary based on the provided configurations. - Parameters - - **compiler\_config** – Compiler configuration. - **runtime\_config** – Runtime configuration. Examples >>> compiler_configs = AicCompilerConfig(graph_names=["graph1"]) >>> AicConfigHelper.to_backend_extension_dict(compiler_configs) Copy to clipboard - Returns - Dictionary representing the backend extension. - *pydantic model* qairt.api.common.backends.aic.config.AicRuntimeConfig - Runtime configuration for AIC Backend - *field* device\_ids*: List[int]* *[Optional]* - Specifies the device ID(s) on which to execute the network. - *field* num\_activations*: int* *= 1* - Specifies the number of concurrent activations (instances of a model) to inference. - *field* submit\_num\_retries*: int* *= 5* - Specifies the number of retries to carry out if submission times out. - *field* submit\_timeout*: int* *= 0* - Specifies the length of time (seconds) to wait for submission to complete. - *field* threads\_per\_queue*: int* *= 4* - Specifies the number of host threads to use for pre- and post-processing. Last Published: Jul 08, 2026 [Previous Topic qairt.api.common.backends](https://docs.qualcomm.com/bundle/publicresource/80-87189-2/topics/qairt-api-common-backends.md) [Next Topic qairt.api.common.backends.htp](https://docs.qualcomm.com/bundle/publicresource/80-87189-2/topics/qairt-api-common-backends-htp.md)