Release 0.4.0

Changelogs

The major improvements and bug fixes since 0.3.0 release are as followings:

• Add image segmentation example using deeplabv3 resnet50 (#22)
• Add image segmentation example using quantized deeplabv3 resnet50 (#23)
• Ensure the contiguousness of ndarray before invoking LIBNUX.tensor_fill_buffer (#26)

Installation

pip install --upgrade 'furiosa-sdk[full]==0.4.0'

Release 0.2.1

Changelogs

• Handle zero-scale issues in quantizer

Installation

pip install --upgrade furiosa-sdk[full]==0.2.1

Release 0.1.0

Changes:

• Add model validator which validates if a Model can run on FuriosaAI NPU
• Add quantizer to transform an onnx model to a quantized model
• Add cli tools, including compiler toolchain, profiler
• Reorganize the source tree for pypi

Installation

pip install --upgrade furiosa-sdk[full]==0.1.0

full will install all features, but you can selectively choose following extra packages:

• cli - cloud-based toolchain and profier
• quantizer - a quantizer to transform an onnx model to a quantized model
• runtime - Python library of FuriosaAI Runtime
• validator - validator which validates if a Model can run on FuriosaAI NPU

For example, if you want runtime and cli, you can install as following:

pip install --upgrade furiosa-sdk[runtime,cli]