Skip to content

Milestones

List view

  • Implement vAccel bindings for the PyTorch framework

    Add in the PyTorch framework the necessary hooks to allow creating sessions, loading models in a session and performing vAccel inference using the vAccel API. Once this milestone is achieved, a programmer will be able to write in PyTorch an inference operation which will be handled by the vAccel runtime system, both directly on a host and inside a guest VM.

    Overdue by 4 year(s)
    •
    Due by November 30, 2021
    0% complete0 open0 closed

  • Add support for PyTorch models in vAccel inference

    Extend the API to allow the registration and loading of PyTorch models from the core runtime. Also, design/extend the back-end plugin to be able to handle the new type of model, load it in accelerator bind input tensors and perform the inference operation

    Overdue by 4 year(s)
    •
    Due by October 31, 2021
    0% complete0 open0 closed

  • vAccel inference support in hypervisors supporting vsock

    Design & Implement the offloading of a vAccel inference operation from a guest VM to the host over vsock. This includes: 1. Extending the vAccel gRPC protocol with the control commands (registering models, etc.) and the actual inference operation 2. Extending the vsock back-end plugin to "speak" the new gRPC 3. Extending the vAccel host agent to understand the new gRPC and lower it to vAccel API. Once this milestone is completed any hypervisor with vsock support will support our vAccel inference operation inside guests.

    Overdue by 4 year(s)
    •
    Due by August 31, 2021
    0% complete0 open0 closed

  • TensorFlow bindings for vAccel inference operation

    Introduce vAccel inference operation as a back-end to TensorFlow. Once this is done a TensorFlow programmer should be able to write in TensorFlow (both C++ and Python) an inference operation with a pre-trained TensorFlow model. This will require adding hooks inside the TensorFlow runtime that handle the TensorFlow session creation, model loading input tensors binding and performing the operation and translate these operation in the corresponding vAccel operations.

    Overdue by 4 year(s)
    •
    Due by June 30, 2021
    0% complete0 open0 closed

  • Host-only vAccel inference operation with TensorFlow pre-trained models

    Design and implement a vAccel inference operation supporting TensorFlow pre-trained models supported on the host. This will require: 1. Design and implementation of the front-end API for the operation 2. Design and implementation of a mechanism for "registering" ML models with the runtime system 3. Implement a back-end plugin based on TensorFlow that can handle the model and perform the operation

    Overdue by 4 year(s)
    •
    Due by May 31, 2021
    •3/3 issues closed
    100% complete0 open3 closed