Back to all discussions

  •   over 1 year ago

External tool

Are we allowed to use other tools than Arango, NetworkX, and cuGraph? (we will most likely build our custom so-called "Agent")
And also about LM, can we use GeminiAPI?

  • 7 comments

  • Manager   •   over 1 year ago

    You can use any other tools but your submission must include ArangoDB, NetworkX, and preferably NVIDIA cuGraph. You can build any custom agent you like, there is no limit on what you can use, including any other tool, LLM, or API.

    However:
    If you want to use cuGraph, you must have an NVIDIA GPU. This will make your NetworkX algorithms run faster.
    If you don’t want to use cuGraph, then you are free to use the normal NetworkX algorithms.

    Actually, using NetworkX and cuGraph as part of this Hackathon is quite simple!

    NetworkX provides a Pythonic, intuitive way to model and analyze graphs, while cuGraph harnesses GPU acceleration for massive-scale graph processing with minimal code changes.

    Why is it so simple?

    Familiar Python API – NetworkX is easy to learn, and cuGraph mirrors its API.
    Easy Transition – You can switch from NetworkX to cuGraph by just replacing imports.
    Very little Setup – No need for complex configurations as you just install the libraries and start coding.
    Massive Speedup – cuGraph runs operations on the GPU, handling millions of nodes/edges effortlessly.

    ***NetworkX Setup (CPU-based, lightweight)***

    Install NetworkX using pip install networkx.
    No additional dependencies needed as it works with standard Python.
    It can run on any system (laptop, server, cloud) without special hardware.
    It's ideal for small to medium-sized graphs and algorithm prototyping.

    ***cuGraph Setup (GPU-accelerated, high-performance)***

    Requires an NVIDIA GPU with CUDA support.
    Install cuGraph via conda install -c rapidsai -c nvidia -c conda-forge -c defaults cugraph (or use a RAPIDS Docker container).
    Uses cuDF (GPU DataFrame) instead of standard Python data structures.
    Works best on large datasets where GPU acceleration provides a significant speedup.

    And of course ArangoDB is very easy to get up and running, all the detail is in the video and in the notebook itself.

    Using any additional tools, frameworks, LLMs, APIs, is fine!

  •   •   over 1 year ago

    Oh, that is nice. thanks fro the quick reply

  • Manager   •   over 1 year ago

    Looking forward to your submission Bayangmbe!

  •   •   over 1 year ago

    I'm looking forward to playing with it. I like Cuda first of all, I like very low-level stuff. But the purpose of this hackathon is to use (networkx, cuGraph, and Arango)... let's see how It goes. I will post here the progress :). So check your notification every day hhh

  • Manager   •   over 1 year ago

    That's great! Can't wait to see what you have done, and appreciate your questions and replies.

  • Manager   •   over 1 year ago

    I have confirmed the following:

    If you want to use cuGraph, you must have an NVIDIA GPU. This will make your NetworkX algorithms run faster.
    If you don’t want to use cuGraph, then you are free to use the normal NetworkX algorithms.

  •   •   over 1 year ago

    I will have a hybrid setup with a high CPU + high GPU and measure the workload.
    I'm already running experiments in parallel (multi-GPU): multi-GPU PageRank with 2xV100.. it is so fun:)

Comments are closed.