diff options
| -rw-r--r-- | var/spack/repos/builtin/packages/py-clip-anytorch/package.py | 33 |
1 files changed, 33 insertions, 0 deletions
diff --git a/var/spack/repos/builtin/packages/py-clip-anytorch/package.py b/var/spack/repos/builtin/packages/py-clip-anytorch/package.py new file mode 100644 index 0000000000..283adbfeea --- /dev/null +++ b/var/spack/repos/builtin/packages/py-clip-anytorch/package.py @@ -0,0 +1,33 @@ +# Copyright 2013-2024 Lawrence Livermore National Security, LLC and other +# Spack Project Developers. See the top-level COPYRIGHT file for details. +# +# SPDX-License-Identifier: (Apache-2.0 OR MIT) + + +from spack.package import * + + +class PyClipAnytorch(PythonPackage): + """CLIP (Contrastive Language-Image Pre-Training) is a neural network + trained on a variety of (image, text) pairs. It can be instructed in + natural language to predict the most relevant text snippet, given an image, + without directly optimizing for the task, similarly to the zero-shot + capabilities of GPT-2 and 3. We found CLIP matches the performance of the + original ResNet50 on ImageNet "zero-shot" without using any of the original + 1.28M labeled examples, overcoming several major challenges in computer + vision.""" + + homepage = "https://github.com/rom1504/CLIP" + # PyPI source is missing requirements.txt + url = "https://github.com/rom1504/CLIP/archive/refs/tags/2.6.0.tar.gz" + + license("MIT", checked_by="qwertos") + + version("2.6.0", sha256="1ac1f6ca47dfb5d4e55be8f45cc2f3bdf6415b91973a04b4529e812a8ae29bea") + + depends_on("py-setuptools", type="build") + depends_on("py-ftfy", type=("build", "run")) + depends_on("py-regex", type=("build", "run")) + depends_on("py-tqdm", type=("build", "run")) + depends_on("py-torch", type=("build", "run")) + depends_on("py-torchvision", type=("build", "run")) |