Intel Gaudi Software for Open Source
Welcome to the Intel Gaudi Software Open Source Project
The Intel Gaudi Software Open Source Project is shared by the Intel Gaudi team with the broader AI community. By providing this project to the open-source community, The Intel Gaudi team encourages the development of cutting-edge AI technologies and solutions that benefit researchers, developers, and enthusiasts worldwide.
-
Download Package: Obtain the package containing all project files from [source location].
-
Extract Files: Extract the contents of the package to your desired location on your system.
- Run habana_set_env_val.sh:
- Open a terminal window.
- Navigate to the package directory.
- Execute the following command:
source trees/habana_set_env_val.sh ./trees ./builds
-
Build hl_logger:
- After setting up the environment, run the following command:
build_hl_logger [-r]
This command will build the
hl_loggerproject. Ensure that it is built successfully before proceeding. - After setting up the environment, run the following command:
-
Build hl_gcfg:
- To build the
hl_gcfgproject, use the following command:build_hl_gcfg [-r]
- To build the
-
Build scal:
- To build the
scalproject, use the following command:build_scal [-r]
- To build the
-
Build mme:
- To build the
mmeproject, use the following command:build_mme [-r]
- To build the
-
Build synapse:
- To build the
synapseproject, use the following command:build_synapse [-r]
- To build the
Note: It's essential to build with this order.
Note: Before running the tests, if the engines_fw or rdma-core package is not installed, ensure that the following environment variables point to the correct locations:
- For eng_XXX.bin: export HABANA_SCAL_BIN_PATH=PATH/IN/YOUR/SYSTEM/builds/engines_fw_release_build
- For libhlib.so: export RDMA_CORE_LIB=PATH/IN/YOUR/SYSTEM/trees/npu-stack/rdma-core/build/lib
Information about the project's license. Specify the license type and provide a link to the full license text if applicable.