Offline Flow: Generating Binaries

  • qnn-context-binary-generator is extended to support Applying of LoRA weights :

    • Receive a new “–adapter_weight_config” parameter to receive Adapters YAML config file (produced by qairt-lora-importer tool)

    • Generate a “binary section” file for each LoRA adapter

  • The produced “binary section” file is used on-target with new QNN API to apply the LoRA adapter

    • Notice: The QAIRT SDK version that generates the base graph context binary and the adapter binary file MUST be the same.

../../_static/resources/lora/qnn_tutorial_lora_offline_generate_binaries.png