Torch Plugin🔗

The Torch plugin for vAccel implements basic PyTorch C++ API (LibTorch) support for vAccel operations.

Supported operations🔗

Installing PyTorch C/C++ API files (LibTorch)🔗

You can find instructions on how to install the required files on the Build and Install PyTorch page. The rest of this guide assumes PyTorch is installed at /opt/pytorch and its' libraries are at /opt/pytorch/lib.

Installing the plugin🔗

You can get the latest Torch plugin binary release from the Releases page of the vAccel repository. Releases include binaries for x86_64/aarch64 Ubuntu-based systems.

Requirements🔗

The prebuilt Torch plugin binaries depend on libstb. You can install it with:

sudo apt install libstb0

TAR🔗

To install the TAR binary package of the latest Torch plugin release:

x86ARM (64-bit)

wget https://github.com/nubificus/vaccel/releases/download/v0.8.0/vaccel-torch_0.3.0+vaccel.0.8.0_x86_64.tar.gz
# Replace '/usr/local' below with the desired installation prefix
tar xfv vaccel-torch_0.3.0+vaccel.0.8.0_x86_64.tar.gz --strip-components=2 -C /usr/local
# Update pkg-config files with the correct prefix
find /usr/local -name "vaccel-torch.pc" -exec sed -i 's:^\(prefix=\).*:\1/usr/local:g' {} \;

wget https://github.com/nubificus/vaccel/releases/download/v0.8.0/vaccel-torch_0.3.0+vaccel.0.8.0_aarch64.tar.gz
# Replace '/usr/local' below with the desired installation prefix
tar xfv vaccel-torch_0.3.0+vaccel.0.8.0_aarch64.tar.gz --strip-components=2 -C /usr/local
# Update pkg-config files with the correct prefix
find /usr/local -name "vaccel-torch.pc" -exec sed -i 's:^\(prefix=\).*:\1/usr/local:g' {} \;

Latest artifacts🔗

You can also find prebuilt artifacts of the latest Torch plugin revision at:

x86ARM (64-bit)

# TAR
https://s3.nbfc.io/nbfc-assets/github/vaccel/plugins/torch/rev/main/x86_64/release/vaccel-torch_latest_x86_64.tar.gz

# TAR
https://s3.nbfc.io/nbfc-assets/github/vaccel/plugins/torch/rev/main/aarch64/release/vaccel-torch_latest_aarch64.tar.gz

Usage🔗

To specify Torch plugin as the selected plugin for vAccel execution:

export VACCEL_PLUGINS=libvaccel-torch.so

Ensure LibTorch, vAccel and the Torch plugin libraries are in the library search paths before trying to use the plugin.

To run image classification you will also need to specify a file with labels using VACCEL_TORCH_LABELS. vAccel includes an file imagenet labels file. Assuming vAccel is installed at /usr/local, you can specify the file with:

export VACCEL_TORCH_LABELS=/usr/local/share/vaccel/labels/imagenet.txt

If vAccel is installed using the DEB package (at /usr) you can omit the variable.

Running an example🔗

Export the necessary variables:

export VACCEL_PLUGINS=libvaccel-torch.so
export VACCEL_TORCH_LABELS=/usr/local/share/vaccel/labels/imagenet.txt
# Optionally, for verbose output
export VACCEL_LOG_LEVEL=4

Assuming vAccel is installed at /usr/local, you can run an image classification with a ResNet model from https://s3.nbfc.io/torch/resnet18.pt with:

$ classify /usr/local/share/vaccel/images/example.jpg 1 \
      https://s3.nbfc.io/torch/resnet18.pt
2026.04.29-15:19:39.63 - <debug> Initializing vAccel
2026.04.29-15:19:39.63 - <info> vAccel 0.7.1-93-ebc23b1f
2026.04.29-15:19:39.63 - <debug> Config:
2026.04.29-15:19:39.63 - <debug>   plugins = libvaccel-torch.so
2026.04.29-15:19:39.63 - <debug>   log_level = debug
2026.04.29-15:19:39.63 - <debug>   log_file = (null)
2026.04.29-15:19:39.63 - <debug>   profiling_enabled = false
2026.04.29-15:19:39.63 - <debug>   version_ignore = false
2026.04.29-15:19:39.74 - <debug> Created top-level rundir: /run/user/0/vaccel/uscxgC
2026.04.29-15:19:39.95 - <info> Registered plugin torch 0.2.1-27-abc7d840
2026.04.29-15:19:39.95 - <debug> Registered op torch_model_load from plugin torch
2026.04.29-15:19:39.95 - <debug> Registered op torch_model_run from plugin torch
2026.04.29-15:19:39.95 - <debug> Registered op torch_sgemm from plugin torch
2026.04.29-15:19:39.95 - <debug> Registered op image_classify from plugin torch
2026.04.29-15:19:39.95 - <debug> Loaded plugin torch from libvaccel-torch.so
2026.04.29-15:19:39.95 - <debug> New rundir for session 1: /run/user/0/vaccel/uscxgC/session.1
2026.04.29-15:19:39.95 - <debug> Initialized session 1 with plugin torch
Initialized session with id: 1
2026.04.29-15:19:39.95 - <debug> Initialized resource 1
2026.04.29-15:19:39.95 - <debug> New rundir for resource 1: /run/user/0/vaccel/uscxgC/resource.1
2026.04.29-15:19:39.95 - <debug> Downloading https://s3.nbfc.io/torch/resnet18.pt
2026.04.29-15:19:40.99 - <debug> Downloaded: 44.7 MB of 44.7 MB (100.0%) | Speed: 43.19 MB/sec
2026.04.29-15:19:40.99 - <debug> Download completed successfully
2026.04.29-15:19:40.99 - <debug> session:1 Registered resource 1
2026.04.29-15:19:40.99 - <debug> session:1 Looking for func implementing op image_classify
2026.04.29-15:19:40.99 - <debug> Returning func for op image_classify from plugin torch
2026.04.29-15:19:40.99 - <warn> [torch] Registered model is not loaded; loading...
2026.04.29-15:19:40.99 - <debug> [torch] Running in CPU mode
2026.04.29-15:19:40.99 - <debug> [torch] Loading model from /run/user/0/vaccel/uscxgC/resource.1/resnet18.pt
2026.04.29-15:19:41.05 - <debug> [torch] Loaded registered model
2026.04.29-15:19:41.06 - <debug> [torch] Disabling graph executor optimization
2026.04.29-15:19:41.09 - <debug> [torch] Prediction: banana 87.34%
classification tags: banana
classification imagename: PLACEHOLDER
2026.04.29-15:19:41.09 - <debug> session:1 Unregistered resource 1
2026.04.29-15:19:41.09 - <debug> Removing file /run/user/0/vaccel/uscxgC/resource.1/resnet18.pt
2026.04.29-15:19:41.10 - <debug> Released resource 1
2026.04.29-15:19:41.10 - <debug> Released session 1
2026.04.29-15:19:41.16 - <debug> Cleaning up vAccel
2026.04.29-15:19:41.16 - <debug> Cleaning up sessions
2026.04.29-15:19:41.16 - <debug> Cleaning up resources
2026.04.29-15:19:41.16 - <debug> Cleaning up plugins
2026.04.29-15:19:41.16 - <debug> Unregistered plugin torch

To run a torch inference example with the generic jitload_forward operation:

$ torch_inference /usr/local/share/vaccel/images/example.jpg \
      https://s3.nbfc.io/torch/resnet18.pt \
      "${VACCEL_TORCH_LABELS}"
2026.04.29-15:19:57.82 - <debug> Initializing vAccel
2026.04.29-15:19:57.82 - <info> vAccel 0.7.1-93-ebc23b1f
2026.04.29-15:19:57.82 - <debug> Config:
2026.04.29-15:19:57.82 - <debug>   plugins = libvaccel-torch.so
2026.04.29-15:19:57.82 - <debug>   log_level = debug
2026.04.29-15:19:57.82 - <debug>   log_file = (null)
2026.04.29-15:19:57.82 - <debug>   profiling_enabled = false
2026.04.29-15:19:57.82 - <debug>   version_ignore = false
2026.04.29-15:19:57.82 - <debug> Created top-level rundir: /run/user/0/vaccel/d2iSTG
2026.04.29-15:19:58.01 - <info> Registered plugin torch 0.2.1-27-abc7d840
2026.04.29-15:19:58.01 - <debug> Registered op torch_model_load from plugin torch
2026.04.29-15:19:58.01 - <debug> Registered op torch_model_run from plugin torch
2026.04.29-15:19:58.01 - <debug> Registered op torch_sgemm from plugin torch
2026.04.29-15:19:58.01 - <debug> Registered op image_classify from plugin torch
2026.04.29-15:19:58.01 - <debug> Loaded plugin torch from libvaccel-torch.so
2026.04.29-15:19:58.01 - <debug> Initialized resource 1
Initialized model resource 1
2026.04.29-15:19:58.01 - <debug> New rundir for session 1: /run/user/0/vaccel/d2iSTG/session.1
2026.04.29-15:19:58.01 - <debug> Initialized session 1 with plugin torch
Initialized vAccel session 1
2026.04.29-15:19:58.01 - <debug> New rundir for resource 1: /run/user/0/vaccel/d2iSTG/resource.1
2026.04.29-15:19:58.01 - <debug> Downloading https://s3.nbfc.io/torch/resnet18.pt
2026.04.29-15:19:59.04 - <debug> Downloaded: 44.7 MB of 44.7 MB (100.0%) | Speed: 43.19 MB/sec
2026.04.29-15:19:59.04 - <debug> Download completed successfully
2026.04.29-15:19:59.04 - <debug> session:1 Registered resource 1
2026.04.29-15:19:59.04 - <debug> session:1 Looking for func implementing op torch_model_load
2026.04.29-15:19:59.04 - <debug> Returning func for op torch_model_load from plugin torch
2026.04.29-15:19:59.04 - <debug> [torch] Running in CPU mode
2026.04.29-15:19:59.05 - <debug> [torch] Loading model from /run/user/0/vaccel/d2iSTG/resource.1/resnet18.pt
2026.04.29-15:19:59.12 - <debug> session:1 Looking for func implementing op torch_model_run
2026.04.29-15:19:59.12 - <debug> Returning func for op torch_model_run from plugin torch
2026.04.29-15:19:59.12 - <debug> [torch] session:1 Jitload & Forward Process
2026.04.29-15:19:59.12 - <debug> [torch] Model: /run/user/0/vaccel/d2iSTG/resource.1/resnet18.pt
2026.04.29-15:19:59.12 - <debug> [torch] Disabling graph executor optimization
Success!
Result Tensor :
Output tensor => type:7 nr_dims:2 size:4000B
Prediction: banana
2026.04.29-15:19:59.16 - <debug> session:1 Unregistered resource 1
2026.04.29-15:19:59.16 - <debug> Released session 1
2026.04.29-15:19:59.16 - <debug> Removing file /run/user/0/vaccel/d2iSTG/resource.1/resnet18.pt
2026.04.29-15:19:59.17 - <debug> Released resource 1
2026.04.29-15:19:59.23 - <debug> Cleaning up vAccel
2026.04.29-15:19:59.23 - <debug> Cleaning up sessions
2026.04.29-15:19:59.23 - <debug> Cleaning up resources
2026.04.29-15:19:59.23 - <debug> Cleaning up plugins
2026.04.29-15:19:59.23 - <debug> Unregistered plugin torch