Installing Domino IQ

The Domino IQ installation is separate from Domino. It involves downloading a .zip file and unzipping it to the Domino program directory. This feature is available only on Windows and Linux 64-bit Domino servers with hardware NVIDIA GPU cards installed. HCL recommends NVIDIA GPU hardware with compute capability of 8.0 or higher for faster AI inferencing response time.

About this task

The .zip file contains both llama server binaries and NVIDIA binaries.

Procedure

  • Download the llama.zip file from the My HCLSoftware Portal and copy it to a temporary space on the Domino server.
  • In the Domino server's program directory , shut down the Domino server and delete the following files:
    Windows (as admin user or user with administrative privileges)
    llama-server.exe
    llama.dll
    ggml*.dll
    cudart64_12.dll
    cublas64_12.dll
    cublasLt64_12.dll
    dominoiq-llama-buildversion.txt
    Linux (as root user)
    llama-server
    libllama.so
    libggml*.so
    libcudart64.so.12
    libcublas64.so.12
    libcublasLt64.so.12
    dominoiq-llama-buildversion.txt
  • On Domino IQ servers (both Windows and Linux), unzip the contents of llama.zip into the Domino program directory.
    On Linux (as root user), additionally perform this step to provide:
    chown root:root llama-server libllama.so libggml*.so libcudart.so.12 libcublas.so.12 libcublasLt.so.12
    chmod ugo+rx llama-server libllama.so libggml*.so libcudart.so.12 libcublas.so.12 libcublasLt.so.12
  • If the Domino IQ server is already configured in the Domino IQ database, start the Domino server and make sure the DominoIQ task has loaded the model correctly.

    If the Domino IQ server is not configured, proceed to Enabling Domino IQ servers.