Skip to content
@Machine-Learning-Headquarters

Machine Learning Headquarters

Machine Learning Headquarters

An exploration in democratizing machine learning technology and the infrastructures that support it.


Goals and Ideals:

  • Creating scaleable, affordable hardware for those interested in learning and contributing to machine learning.
  • Open Source education and designs.
  • Lower the barrier to entry for AI, ML, and supercomputing.
  • Document everything, release everything free of charge.
  • This is not a startup, this is a lifestyle. A mission.

  • What is the first hurdle that needs to be overcome? What's the smallest goal we can meet while making some sort of difference?
    All of the goals below can be done independently in ways that will have an impact (I think), the easiest is by creating a compendium of resources for individuals to follow the same path. The next would be to make an open source design for a specific goal (this would be the super expandable super clusters). From previous conversation, it sounds like the current bottleneck is affordable high speed network technology, so we can start by understanding that and then tackling other avenues.

    What does it take to run popular machine learning models (huggingface) on risc-v hardware.

    What makes a super computer? Memory Bandwidth? CPU Clock? Instructions per clock? 200,000 nodes?

    Can we use the same hardware to scale a super computer from $100 to $100 million?

    Can we produce a risc-v processor that uses DDR5, PCIE5, 10Gbe network, and runs C/python without paying license fees? Do we need NVME?

    What is the fastest way to connect nodes in a cluster that is "plug and play" (just buy more nodes, not more connectors, or bigger nodes)

    Is it possible to run nvidia CUDA software on risc-v? If not what is a realistic alternative to CUDA that is actually competative. Is there something close to a drop in replacement? What does that look like? Does VULKAN+SPIR-V fit in here?

    Can Kubernetes run on a risc-v cluster?

    Can FreeNAS run on a risc-v cluster?

    Minimum Viable Product:

  • 150 Intel NUC kubernetes cluster benchmark training and inference for statistical learning and ML Transformers for NLP
  • A high speed, low cost network interface.
  • Stretch goals:

  • Open source blade for expanding commonly available systems like the RPi.
  • Open Source rack motherboard for connecting those blades to an ethernet link (the blade is probably useless without this)
  • Open source CPUs that can slot into the blade. This requires new designs all the way down.
  • Rank on the super computer leaderboard with a risc-v based supercomputer.
  • Current Configuration:

    A plurality of CPUs are mounted to a "blade" that allows an interface into a rack mounted motherboard, more than likely through a PCIe bus. These motherboards can be stacked and expanded through high speed ethernet interfaces that link to high speed network switches in order to expand a supercomputer cluster further.

    Popular repositories Loading

    1. mlhq.io mlhq.io Public

      A group of tasks for completing my blog MLHQ.io

      Jupyter Notebook 1

    2. .github .github Public

    3. BladeArchitectures BladeArchitectures Public

      Super computing cluster architecture (cluster block design)

    4. NetworkingArchitectures NetworkingArchitectures Public

      High speed networks connecting clusters

    5. CPUDesign CPUDesign Public

      What CPUs are powering the cluster

    6. HighFrequencyCircuitDesign HighFrequencyCircuitDesign Public

      Details for how high speed communication is handled on a physical level

    Repositories

    Showing 10 of 14 repositories
    • spark-on-k8s-operator Public Forked from kubeflow/spark-operator

      Kubernetes operator for managing the lifecycle of Apache Spark applications on Kubernetes.

      Machine-Learning-Headquarters/spark-on-k8s-operator’s past year of commit activity
      Go 0 Apache-2.0 1,396 0 0 Updated Oct 26, 2024
    • docker-containers Public

      Docker Containers

      Machine-Learning-Headquarters/docker-containers’s past year of commit activity
      Shell 0 0 0 0 Updated May 18, 2023
    • stable-dreamfusion Public Forked from ashawkey/stable-dreamfusion

      A pytorch implementation of text-to-3D dreamfusion, powered by stable diffusion.

      Machine-Learning-Headquarters/stable-dreamfusion’s past year of commit activity
      Python 0 Apache-2.0 743 0 0 Updated May 17, 2023
    • Anything-3D Public Forked from Anything-of-anything/Anything-3D

      Segment-Anything + 3D. Let's lift anything to 3D.

      Machine-Learning-Headquarters/Anything-3D’s past year of commit activity
      0 79 0 0 Updated Apr 21, 2023
    • 3DFuse Public Forked from cvlab-kaist/3DFuse

      Official implementation of "Let 2D Diffusion Model Know 3D-Consistency for Robust Text-to-3D Generation"

      Machine-Learning-Headquarters/3DFuse’s past year of commit activity
      Python 0 43 0 0 Updated Apr 10, 2023
    • .github Public
      Machine-Learning-Headquarters/.github’s past year of commit activity
      0 0 0 0 Updated Mar 25, 2023
    • NetworkingArchitectures Public

      High speed networks connecting clusters

      Machine-Learning-Headquarters/NetworkingArchitectures’s past year of commit activity
      0 0 0 0 Updated Mar 25, 2023
    • FPGADesign Public

      Repo for information on FPGAs. These tend to be "glue logic" and interfaces that are highly specialized or parallelized

      Machine-Learning-Headquarters/FPGADesign’s past year of commit activity
      0 0 0 0 Updated Mar 22, 2023
    • AIModels Public

      What AI architectures are required/supported

      Machine-Learning-Headquarters/AIModels’s past year of commit activity
      0 0 0 0 Updated Mar 21, 2023
    • BladeArchitectures Public

      Super computing cluster architecture (cluster block design)

      Machine-Learning-Headquarters/BladeArchitectures’s past year of commit activity
      0 0 0 0 Updated Mar 21, 2023

    People

    This organization has no public members. You must be a member to see who’s a part of this organization.

    Top languages

    Loading…

    Most used topics

    Loading…