Installation Error

[captain1fs]

When I installed fewbit, I got the following error：

pip install -U fewbit
ERROR: Could not find a version that satisfies the requirement fewbit (from versions: none)
ERROR: No matching distribution found for fewbit

pip install -U https://github.com/SkoltechAI/fewbit.git
Collecting https://github.com/SkoltechAI/fewbit.git
Downloading https://github.com/SkoltechAI/fewbit.git
\ 223.1 kB 2.1 MB/s 0:00:00
ERROR: Cannot unpack file /tmp/pip-unpack-a3ukc7tp/fewbit.git (downloaded from /tmp/pip-req-build-aipk8zcb, content-type: text/html; charset=utf-8); cannot detect archive format
ERROR: Cannot determine archive format of /tmp/pip-req-build-aipk8zcb

So how should I do?Thanks.

[daskol]

Thanks! This is a typo. Actual command follows.

pip install -U git+https://github.com/SkoltechAI/fewbit.git

Also, I'd recommend to switch to wheels branch (see #3) where we are trying to build manylinux wheels for different CUDAs and PyTorch versions.

[captain1fs]

Thanks for your reply! I have installed fewbit successfully.But I got errors when using Piece-wise Activation Functions, as follows:

• RuntimeError: fewbit::relu() expected at most 1 argument(s) but received 2 argument(s). Declaration: fewbit::relu(Tensor(a!) self) -> (Tensor(a!))
• RuntimeError: fewbit::hardsigmoid() expected at most 1 argument(s) but received 2 argument(s). Declaration: fewbit::hardsigmoid(Tensor(a!) self) -> (Tensor(a!))

It is right when using Continous Activation Functions. So what should I do? Thanks!

[daskol]

Yes. It is known issues and should already be fixed which appears for specific PyTorch version. It works well for the newer PyTorch version.

[daskol]

@captain1fs New release published: multiple fixes and pre-built wheels.