Update converter build instructions

docs/compute-engine/build/converter.md

@@ -2,25 +2,18 @@
 
 To build the pip package with the converter, Bazel needs to know for which
 Python version to build the package.
-If you are using a Python virtual environment, please run bazel from that
-environment, as Bazel will automatically detect the Python version of the
-virtual environment.
-
-Bazel will detect the Python version the first time you try to build anything
-that requires Python. If you have run Bazel outside of your virtual
-environment and now want to run it in the environment, you have to make Bazel
-re-detect the Python version. This requires a full clean, a normal `clean` is
-not enough to trigger the redetection.:
+You can set this through the environment variable `TF_PYTHON_VERSION`, for example
 
 ```bash
-bazel clean --expunge
+export TF_PYTHON_VERSION="3.12"
 ```
 
-Install the TensorFlow pip package to ensure that all required dependencies
-are available:
+If you have run Bazel without setting the python version, or you want to change
+the version, you have to make Bazel re-detect the Python version. This requires
+a full clean, a normal `clean` is not enough to trigger the redetection:
 
 ```bash
-pip install tensorflow
+bazel clean --expunge
 ```
 
 To build the pip package, now simply run
@@ -30,8 +23,15 @@ bazel build :build_pip_pkg
 bazel-bin/build_pip_pkg artifacts
 ```
 
-The script stores the wheel file in the `artifacts/` directory located in the
-LCE root directory. To install the PIP package:
+If you get an error about a `@pypi_lce_XXXXX` package, then it can help to
+re-generate the `requirements.txt` file that Bazel uses. To do so, run
+
+```bash
+pip-compile --allow-unsafe --no-emit-index-url --strip-extras larq_compute_engine/requirements.in
+```
+
+After the bazel build is finished, the script stores the wheel file in the
+`artifacts/` directory located in the LCE root directory. To install the PIP package:
 
 ```bash
 pip install artifacts/*.whl

docs/compute-engine/build/docker.md

@@ -7,6 +7,7 @@ container, and we recommend first following
 to try and build LCE outside of a Docker container.
 
 There are scenarios in which building in the Docker container is preferred:
+
 - To build a [`manylinux2010`](https://www.python.org/dev/peps/pep-0571/)-compatible pip package
 - To build a benchmark binary for Android
 - When the build outside of the container fails and can't be fixed.
@@ -17,13 +18,13 @@ to migrate the build targets in-between the host machine and the container.
 To be able to build the LCE converter's
 [`manylinux2010`](https://www.python.org/dev/peps/pep-0571/) compatible PIP
 package, we need to use the
-[`tensorflow:custom-op-ubuntu16`](https://hub.docker.com/r/tensorflow/tensorflow)
-image.
+[`tensorflow/build:2.16-python3.12`](https://hub.docker.com/r/tensorflow/tensorflow)
+image (replace the Python version number with the desired version).
 
 First, download the Docker image:
 
 ```bash
-docker pull tensorflow/tensorflow:custom-op-ubuntu16
+docker pull tensorflow/build:2.16-python3.12
 ```
 
 Clone the LCE repository on the host machine:
@@ -38,7 +39,7 @@ directory inside the container:
 
 ```bash
 docker run -it -v $PWD/lce-volume:/tmp/lce-volume \
-    -w /tmp/lce-volume tensorflow/tensorflow:custom-op-ubuntu16 /bin/bash
+    -w /tmp/lce-volume tensorflow/build:2.16-python3.12 /bin/bash
 ```
 
 Now, you will be able to build bazel targets inside the container, by following
@@ -58,7 +59,7 @@ change the Python version to the desired version:
 
 ```bash
 cd /tmp/lce-volume
-export PYTHON_VERSION="3.8"
+export TF_PYTHON_VERSION="3.12"
 .github/tools/release_linux.sh
 ```