Open Source Development

Power Open Source Development

Explore the open source tools and capabilities for building and deploying modern applications on IBM Power platforms including AIX, IBM i, and Linux.


#Power


#Power

 View Only
  • 1.  AI models on AIX

    Posted 7 days ago
    Edited by Niklas Vosskoetter 7 days ago

    Hey all,

    I want to share my experience with AI on AIX and would like to know if you have already run some tests of AI on AIX. Which models you have been working with and what use cases do you have identified?

    Our test case was triggered by Db2 and the following Github project: https://github.com/shaikhq/db2-langchain-rag-local

    Downloads

    • llama: https://github.com/ggml-org/llama.cpp
    • qwen model: https://huggingface.co/Qwen/Qwen2.5-3B-Instruct-GGUF/resolve/main/qwen2.5-3b-instruct-q4_k_m.gguf

    Prepare AIX

    dnf install gcc gcc-c++ cmake python3-numpy-2.3.2 python3.12-pyyaml-6.0.2

    # tqdm is not available yet, but I was able to copy the files from MacOS into /opt/freeware/lib64/python3.12/site-packages/tqdm

    Build llama (copy the downloaded llama from Github to your AIX test system and extract the zip file if needed)

    # Set all the required vars
    export PATH=/opt/freeware/bin:$PATH
    export LESS=eFRX
    ulimit -m unlimited
    ulimit -d unlimited
    ulimit -f unlimited
    ulimit -c unlimited
    export OBJECT_MODE=64
    export CC="gcc"
    export CXX="g++"
    export CFLAGS="-pthread -maix64 -lm"
    export CXXFLAGS="-pthread -maix64 -lm"
    export LDFLAGS="-lm -lbsd -lcrypt -DLLAMA_NATIVE=ON -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON"
    # Move into the llama directory and run cmake
    cd llama.cpp
    cmake -B build_llama -DLLAMA_CURL=OFF
    cmake --build build_llama/

    Prepare the model from huggingface for big endian (copy the downloaded Model from Huggingface to your AIX test system)

    python3.12 gguf-py/gguf/scripts/gguf_convert_endian.py /your/path/qwen2.5-3b-instruct.Q4_K_M.gguf native
    Run the model
    #  ./build_llama/bin/llama-simple-chat -m /your/path/qwen2.5-3b-instruct.Q4_K_M.gguf
    ........................................................................................
    > Hello world!
    Hello! It's nice to meet you. Is there something you would like to talk about or ask? I'm here to help with any information you need.
    Many thanks to our contacts at IBM for helping us with this!
    Best regards



    ------------------------------
    Niklas Vosskoetter
    Deputy Team Lead & System Engineer Unix and Cloud Service Power
    ------------------------------



  • 2.  RE: AI models on AIX

    Posted 4 days ago

    Hi @Niklas Vosskoetter,

    Which server is it? Power10 or Power11? Scale out?

    What have the results been?

    Are you uploading the internal documents to RAG for questions and answers?



    ------------------------------
    Rafael Cezario
    Technical Director
    BLUE TRUST
    Brasilia
    ------------------------------