ASCEND_CACHE_PATH

Description

Specifies the storage path of shared files. Set this environment variable if you require all shared files generated during component build and runtime to be flushed to a unified directory.

The path can contain letters, digits, underscores (_), hyphens (-), and periods (.).
Ensure that the specified path exists and is valid. The execution user must have the read, write, and execute permissions on the path.

The following table lists the files in ${ASCEND_CACHE_PATH}.

**Table 1** Paths of shared files flushed to disks
Flushed File	File Content	Priority Specification
${ASCEND_CACHE_PATH}/aoe_data	AOE tuning repository file, which can be shared by multiple servers.	TUNE_BANK_PATH > ASCEND_CACHE_PATH > Default knowledge base path For details about the default knowledge base path, see "Note" in TUNE_BANK_PATH.
${ASCEND_CACHE_PATH}/kernel_cache	Operator build cache file, which can be shared by multiple devices on a server.	The following scenarios provide parameters or APIs for setting the storage path of operator build cache files. Their priority is higher than that of the ASCEND_CACHE_PATH environment variable. The details are as follows: Offline model compilation using ATC: Parameter --op_compiler_cache_dir > ASCEND_CACHE_PATH > Default path ($HOME/atc_data) Model build or compilation using AscendCL APIs: Parameter OP_COMPILER_CACHE_DIR in the graph construction API aclgrphBuildInitialize > ASCEND_CACHE_PATH > Default path ($HOME/atc_data) Parameter OP_COMPILER_CACHE_DIR in the graph construction API aclgrphBuildModel > ASCEND_CACHE_PATH > Default path ($HOME/atc_data) Parameter ACL_OP_COMPILER_CACHE_DIR in the application build API aclCompileOpt > ASCEND_CACHE_PATH > Default path ($HOME/atc_data) TensorFlow network training or online inference: TF Adapter configuration parameter op_compiler_cache_dir > ASCEND_CACHE_PATH > Default path ($HOME/atc_data) For details about op_compiler_cache_dir, see: Session Configuration > Session Configuration Parameters in TF Adapter API (1.x). npu.global_options > Configuration Parameters in TF Adapter API (2.x).

Example

export ASCEND_CACHE_PATH=/repo/task001/cache

Constraints

File locking in Linux needs to be supported by the shared storage (such as NFSv4, OceanStor Pacific 9950, and OceanStor Pacific 9550). But if the storage, such as NFSv3, does not support file locking, you are advised to store files to local directories.
In the multi-server scenario, the AI processor model, firmware, driver, and CANN software version on each server must be the same.

Applicability

Atlas training product

Atlas inference product

Atlas A2 training product / Atlas A2 inference product

Atlas A3 training product / Atlas A3 inference product

Atlas 350 Accelerator Card

Atlas 200I/500 A2 inference product

Parent topic: Installation