-
-
Notifications
You must be signed in to change notification settings - Fork 13.4k
Open
Description
The following minimalistic Dockerfile:
FROM debian:11
RUN apt-get update
RUN apt-get install -y \
curl \
git \
python3-distutils
RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=/usr/local/bin sh
RUN git clone --depth 1 https://github.com/deepfakes/faceswap.git
RUN uv pip install --directory faceswap --system --break-system-packages \
--requirements requirements/requirements_rocm.txt
yields:
Using Python 3.9.2 environment at: /usr
× No solution found when resolving dependencies:
╰─▶ Because tensorflow-rocm>=2.10.0.520,<=2.10.1.550 depends on
numpy>=1.20,<1.23 and only the following versions of tensorflow-rocm
are available:
tensorflow-rocm<2.10.0
tensorflow-rocm==2.10.0.520
tensorflow-rocm==2.10.0.530
tensorflow-rocm==2.10.1.540
tensorflow-rocm==2.10.1.550
tensorflow-rocm>2.11.0
we can conclude that tensorflow-rocm>=2.10.0,<2.11.0 depends on
numpy>=1.20,<1.23.
And because you require numpy>=1.26.0,<2.0.0 and
tensorflow-rocm>=2.10.0,<2.11.0, we can conclude that your requirements
are unsatisfiable.
I know that this is not the way FS is supposed to be installed. I only did it because PIP is unable to give a proper error message (instead, it aborts), but uv can analyse the requirements successfully with an expressive result message.
tensortflow-rocm is too old. If FS-2 is not supposed to be upgraded to a newer tf version – I understand that such big changes might go to FS-3 only – I recommend to tell in the README that ROCm is not supported.
Metadata
Metadata
Assignees
Labels
No labels