I was recently bumping versions of common libraries in one of my old repositories, that used EvalAI CLI v1.3.7, PyTorch v1.1.0, AllenNLP v0.8.4
I decided to bump PyTorch to v1.7.0 and AllenNLP to v1.0+ (the most recent versions right now). But I am facing a dependency deadlock between EvalAI and AllenNLP:
I wonder if you could make boto3 version requirement a bit softer. This may hinder a bunch of old PyTorch codebases to use newer versions of PyTorch, because AllenNLP versions have a strict dependency on PyTorch.
In general, could you review all your requirements and see if they can be made softer (e.g. tqdm)? For example, the major version could remain a hard requirement, but minor versions could be soft.
I was recently bumping versions of common libraries in one of my old repositories, that used EvalAI CLI v1.3.7, PyTorch v1.1.0, AllenNLP v0.8.4
I decided to bump PyTorch to v1.7.0 and AllenNLP to v1.0+ (the most recent versions right now). But I am facing a dependency deadlock between EvalAI and AllenNLP:
I wonder if you could make boto3 version requirement a bit softer. This may hinder a bunch of old PyTorch codebases to use newer versions of PyTorch, because AllenNLP versions have a strict dependency on PyTorch.
In general, could you review all your requirements and see if they can be made softer (e.g. tqdm)? For example, the major version could remain a hard requirement, but minor versions could be soft.