From 0b8fc625579d3dd841e0513d0329f0606635aca5 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 10:45:20 +0200
Subject: [PATCH 01/10] test module skill with careless

---
 careless/README.md         | 24 ++++++++++++++++++++++++
 careless/build             | 27 +++++++++++++++++++++++++++
 careless/files/config.yaml | 11 +++++++++++
 careless/modulefile        | 16 ++++++++++++++++
 4 files changed, 78 insertions(+)
 create mode 100644 careless/README.md
 create mode 100755 careless/build
 create mode 100644 careless/files/config.yaml
 create mode 100644 careless/modulefile

diff --git a/careless/README.md b/careless/README.md
new file mode 100644
index 0000000..9901225
--- /dev/null
+++ b/careless/README.md
@@ -0,0 +1,24 @@
+# careless
+
+Scaling and merging of X-ray diffraction data with approximate Bayesian
+inference (variational merging), built on TensorFlow.
+
+- Upstream: https://github.com/rs-station/careless
+- License: MIT
+
+## How it is built
+
+The build installs a private miniconda into `$PREFIX/miniconda`, creates a
+`careless_<version>` env with Python 3.12 (upstream supports >=3.9,<3.13),
+and runs `pip install careless[cuda]==<version>` inside it. The `[cuda]`
+extra pulls the NVIDIA libraries as pip wheels, so the same install works
+on GPU and CPU-only Ra nodes.
+
+Loading the module puts the env's `bin/` on `PATH` (the `careless` command
+is directly available, no `conda activate` needed) and sets `CARELESS_ENV`
+to the conda env path.
+
+## Adding a new version
+
+Add the version under `versions:` in `files/config.yaml`, then on a Ra
+login node: `./build <version>`.
diff --git a/careless/build b/careless/build
new file mode 100755
index 0000000..8869288
--- /dev/null
+++ b/careless/build
@@ -0,0 +1,27 @@
+#!/usr/bin/env modbuild
+
+pbuild::prep() {
+    :
+}
+
+pbuild::configure() {
+    # careless is pip-installed, but upstream recommends a fresh conda env
+    # (TensorFlow dependency conflicts), so install a private miniconda in PREFIX
+    mkdir -p "$PREFIX/miniconda"
+    wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O "$PREFIX/miniconda/miniconda.sh"
+    bash "$PREFIX/miniconda/miniconda.sh" -b -u -p "$PREFIX/miniconda/"
+    "$PREFIX/miniconda/condabin/conda" create -y -n "careless_$V" python=3.12
+}
+
+pbuild::compile() {
+    :
+}
+
+pbuild::install() {
+    source "$PREFIX/miniconda/etc/profile.d/conda.sh"
+    conda activate "careless_$V"
+    pip install --upgrade pip
+    # [cuda] extra ships the NVIDIA libs via pip wheels; works on CPU-only nodes too
+    pip install --no-cache-dir "careless[cuda]==${V_PKG}"
+    conda deactivate
+}
diff --git a/careless/files/config.yaml b/careless/files/config.yaml
new file mode 100644
index 0000000..cfd5995
--- /dev/null
+++ b/careless/files/config.yaml
@@ -0,0 +1,11 @@
+---
+format: 1
+careless:
+  defaults:
+    group: MX
+    overlay: base
+    relstage: unstable
+  versions:
+    0.5.4:
+      config:
+        relstage: unstable
diff --git a/careless/modulefile b/careless/modulefile
new file mode 100644
index 0000000..e2c2743
--- /dev/null
+++ b/careless/modulefile
@@ -0,0 +1,16 @@
+#%Module1.0
+
+module-whatis       "careless - merging crystallography data with variational inference"
+module-url          "https://github.com/rs-station/careless"
+module-license      "MIT"
+module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
+
+module-help     "
+careless: scaling and merging of X-ray diffraction data using
+approximate Bayesian inference (TensorFlow). Installed with the
+[cuda] extra, so it uses NVIDIA GPUs when available and falls
+back to CPU otherwise.
+"
+
+setenv          CARELESS_ENV    $PREFIX/miniconda/envs/careless_$V
+prepend-path    PATH            $PREFIX/miniconda/envs/careless_$V/bin

From c0410dfe2d168e91ad8fd44dd7caed87bf62e706 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 10:51:44 +0200
Subject: [PATCH 02/10] careless: build venv with uv instead of conda
 (anaconda.com blocked on Ra)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 careless/README.md  | 21 ++++++++++++---------
 careless/build      | 17 ++++++-----------
 careless/modulefile |  4 ++--
 3 files changed, 20 insertions(+), 22 deletions(-)

diff --git a/careless/README.md b/careless/README.md
index 9901225..b2df471 100644
--- a/careless/README.md
+++ b/careless/README.md
@@ -8,17 +8,20 @@ inference (variational merging), built on TensorFlow.
 
 ## How it is built
 
-The build installs a private miniconda into `$PREFIX/miniconda`, creates a
-`careless_<version>` env with Python 3.12 (upstream supports >=3.9,<3.13),
-and runs `pip install careless[cuda]==<version>` inside it. The `[cuda]`
-extra pulls the NVIDIA libraries as pip wheels, so the same install works
-on GPU and CPU-only Ra nodes.
+Upstream recommends a fresh conda env, but repo.anaconda.com is not
+reachable from Ra, so the build uses uv instead: it fetches a standalone
+Python 3.12 into `$PREFIX/python` (upstream supports >=3.9,<3.13), creates
+`$PREFIX/venv`, and runs `uv pip install careless[cuda]==<version>` into
+it. The `[cuda]` extra pulls the NVIDIA libraries as pip wheels, so the
+same install works on GPU and CPU-only Ra nodes.
 
-Loading the module puts the env's `bin/` on `PATH` (the `careless` command
-is directly available, no `conda activate` needed) and sets `CARELESS_ENV`
-to the conda env path.
+`uv` must be on PATH on the build host (e.g. `pip install --user uv`).
+
+Loading the module puts the venv's `bin/` on `PATH` (the `careless`
+command is directly available, no activation needed) and sets
+`CARELESS_ENV` to the venv path.
 
 ## Adding a new version
 
 Add the version under `versions:` in `files/config.yaml`, then on a Ra
-login node: `./build <version>`.
+login node: `modbuild build <version>` (modbuild/2.1.2).
diff --git a/careless/build b/careless/build
index 8869288..95a9d6c 100755
--- a/careless/build
+++ b/careless/build
@@ -5,12 +5,11 @@ pbuild::prep() {
 }
 
 pbuild::configure() {
-    # careless is pip-installed, but upstream recommends a fresh conda env
-    # (TensorFlow dependency conflicts), so install a private miniconda in PREFIX
-    mkdir -p "$PREFIX/miniconda"
-    wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O "$PREFIX/miniconda/miniconda.sh"
-    bash "$PREFIX/miniconda/miniconda.sh" -b -u -p "$PREFIX/miniconda/"
-    "$PREFIX/miniconda/condabin/conda" create -y -n "careless_$V" python=3.12
+    # repo.anaconda.com is unreachable from Ra (GnuTLS error), so no conda.
+    # uv fetches a standalone CPython instead; keep it inside PREFIX so the
+    # venv's interpreter symlinks don't depend on the builder's home cache.
+    export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
+    uv venv --python 3.12 "$PREFIX/venv"
 }
 
 pbuild::compile() {
@@ -18,10 +17,6 @@ pbuild::compile() {
 }
 
 pbuild::install() {
-    source "$PREFIX/miniconda/etc/profile.d/conda.sh"
-    conda activate "careless_$V"
-    pip install --upgrade pip
     # [cuda] extra ships the NVIDIA libs via pip wheels; works on CPU-only nodes too
-    pip install --no-cache-dir "careless[cuda]==${V_PKG}"
-    conda deactivate
+    uv pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
 }
diff --git a/careless/modulefile b/careless/modulefile
index e2c2743..54ec7d0 100644
--- a/careless/modulefile
+++ b/careless/modulefile
@@ -12,5 +12,5 @@ approximate Bayesian inference (TensorFlow). Installed with the
 back to CPU otherwise.
 "
 
-setenv          CARELESS_ENV    $PREFIX/miniconda/envs/careless_$V
-prepend-path    PATH            $PREFIX/miniconda/envs/careless_$V/bin
+setenv          CARELESS_ENV    $PREFIX/venv
+prepend-path    PATH            $PREFIX/venv/bin

From 6f7523601b62612be078b21d42ea92c590706544 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 10:54:10 +0200
Subject: [PATCH 03/10] careless: call uv by explicit path, modbuild strips
 ~/.local/bin from PATH

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 careless/build | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/careless/build b/careless/build
index 95a9d6c..6af67fd 100755
--- a/careless/build
+++ b/careless/build
@@ -8,8 +8,10 @@ pbuild::configure() {
     # repo.anaconda.com is unreachable from Ra (GnuTLS error), so no conda.
     # uv fetches a standalone CPython instead; keep it inside PREFIX so the
     # venv's interpreter symlinks don't depend on the builder's home cache.
+    # modbuild's shell strips ~/.local/bin from PATH; override with UV=/path/to/uv if needed
+    UV="${UV:-$HOME/.local/bin/uv}"
     export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
-    uv venv --python 3.12 "$PREFIX/venv"
+    "$UV" venv --python 3.12 "$PREFIX/venv"
 }
 
 pbuild::compile() {
@@ -17,6 +19,7 @@ pbuild::compile() {
 }
 
 pbuild::install() {
+    UV="${UV:-$HOME/.local/bin/uv}"
     # [cuda] extra ships the NVIDIA libs via pip wheels; works on CPU-only nodes too
-    uv pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
+    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
 }

From f983b0a174a89b5a8fc16699bdc58de566367755 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 10:55:59 +0200
Subject: [PATCH 04/10] careless: put uv cache under PREFIX, home quota too
 small for TF wheels

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 careless/build | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/careless/build b/careless/build
index 6af67fd..189d84a 100755
--- a/careless/build
+++ b/careless/build
@@ -10,6 +10,8 @@ pbuild::configure() {
     # venv's interpreter symlinks don't depend on the builder's home cache.
     # modbuild's shell strips ~/.local/bin from PATH; override with UV=/path/to/uv if needed
     UV="${UV:-$HOME/.local/bin/uv}"
+    # ~/.cache/uv hits the home disk quota; cache under PREFIX (removed after install)
+    export UV_CACHE_DIR="$PREFIX/.uv-cache"
     export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
     "$UV" venv --python 3.12 "$PREFIX/venv"
 }
@@ -20,6 +22,8 @@ pbuild::compile() {
 
 pbuild::install() {
     UV="${UV:-$HOME/.local/bin/uv}"
+    export UV_CACHE_DIR="$PREFIX/.uv-cache"
     # [cuda] extra ships the NVIDIA libs via pip wheels; works on CPU-only nodes too
     "$UV" pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
+    rm -rf "$UV_CACHE_DIR"
 }

From b55d6c1d91536f480b604c40dd8affe3d44307d8 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 11:07:12 +0200
Subject: [PATCH 05/10] Split careless into CPU module and careless-gpu using
 cuda/12.8.1 module

careless installs plain careless (CPU). careless-gpu also installs plain
careless but declares cuda/12.8.1 as build_requires/runtime_deps
(jfjoch_viewer pattern) instead of pip-bundling CUDA via the [cuda] extra.
Both builds honor UV_CACHE_DIR to share one download cache.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 careless-gpu/README.md         | 28 ++++++++++++++++++++++++++++
 careless-gpu/build             | 26 ++++++++++++++++++++++++++
 careless-gpu/files/config.yaml | 13 +++++++++++++
 careless-gpu/modulefile        | 16 ++++++++++++++++
 careless/README.md             | 11 ++++++++---
 careless/build                 | 16 +++++++++-------
 careless/modulefile            |  5 ++---
 7 files changed, 102 insertions(+), 13 deletions(-)
 create mode 100644 careless-gpu/README.md
 create mode 100755 careless-gpu/build
 create mode 100644 careless-gpu/files/config.yaml
 create mode 100644 careless-gpu/modulefile

diff --git a/careless-gpu/README.md b/careless-gpu/README.md
new file mode 100644
index 0000000..fab39aa
--- /dev/null
+++ b/careless-gpu/README.md
@@ -0,0 +1,28 @@
+# careless-gpu
+
+GPU build of careless (see `../careless/` for the CPU module and the full
+background). Scaling and merging of X-ray diffraction data with approximate
+Bayesian inference (variational merging), built on TensorFlow.
+
+- Upstream: https://github.com/rs-station/careless
+- License: MIT
+
+## How it differs from the careless module
+
+- CUDA is not pip-installed (no `[cuda]` extra, saves several GB): the
+  cluster `cuda/12.8.1` module provides it instead, declared as
+  `build_requires`/`runtime_deps` (same pattern as `jfjoch_viewer`).
+  `runtime_deps` matters because TensorFlow dlopens the CUDA libraries
+  when careless runs, so cuda must auto-load with this module.
+- If careless reports missing cuDNN at runtime, the cuda module does not
+  ship it; the fallback is installing `careless[cuda]` in `build` instead.
+
+The venv layout, uv usage, and cache handling are identical to `careless`.
+To share one download cache between both builds, run each with
+`UV_CACHE_DIR=<shared path> modbuild build <version>` — a cache outside
+PREFIX is kept, an in-PREFIX cache is removed after a successful install.
+
+## Adding a new version
+
+Add the version under `versions:` in `files/config.yaml`, then on a Ra
+login node: `modbuild build <version>` (modbuild/2.1.2).
diff --git a/careless-gpu/build b/careless-gpu/build
new file mode 100755
index 0000000..49cb4e9
--- /dev/null
+++ b/careless-gpu/build
@@ -0,0 +1,26 @@
+#!/usr/bin/env modbuild
+
+pbuild::prep() {
+    :
+}
+
+pbuild::configure() {
+    # same uv pattern as the careless (CPU) module; see that build for the whys
+    UV="${UV:-$HOME/.local/bin/uv}"
+    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
+    export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
+    "$UV" venv --python 3.12 "$PREFIX/venv"
+}
+
+pbuild::compile() {
+    :
+}
+
+pbuild::install() {
+    UV="${UV:-$HOME/.local/bin/uv}"
+    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
+    # no [cuda] extra: CUDA comes from the cuda/12.8.1 module
+    # (build_requires/runtime_deps in files/config.yaml), not from pip wheels
+    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
+    [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
+}
diff --git a/careless-gpu/files/config.yaml b/careless-gpu/files/config.yaml
new file mode 100644
index 0000000..1a7f199
--- /dev/null
+++ b/careless-gpu/files/config.yaml
@@ -0,0 +1,13 @@
+---
+format: 1
+careless-gpu:
+  defaults:
+    group: MX
+    overlay: base
+    relstage: unstable
+  versions:
+    0.5.4:
+      config:
+        relstage: unstable
+        build_requires: ["cuda/12.8.1"]
+        runtime_deps: ["cuda/12.8.1"]
diff --git a/careless-gpu/modulefile b/careless-gpu/modulefile
new file mode 100644
index 0000000..486cb39
--- /dev/null
+++ b/careless-gpu/modulefile
@@ -0,0 +1,16 @@
+#%Module1.0
+
+module-whatis       "careless (GPU) - merging crystallography data with variational inference"
+module-url          "https://github.com/rs-station/careless"
+module-license      "MIT"
+module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
+
+module-help     "
+careless: scaling and merging of X-ray diffraction data using
+approximate Bayesian inference (TensorFlow). GPU build: CUDA is
+provided by the cuda/12.8.1 module, auto-loaded as a runtime dep.
+Use on nodes with an NVIDIA GPU; for CPU nodes load careless instead.
+"
+
+setenv          CARELESS_ENV    $PREFIX/venv
+prepend-path    PATH            $PREFIX/venv/bin
diff --git a/careless/README.md b/careless/README.md
index b2df471..757adc9 100644
--- a/careless/README.md
+++ b/careless/README.md
@@ -11,9 +11,14 @@ inference (variational merging), built on TensorFlow.
 Upstream recommends a fresh conda env, but repo.anaconda.com is not
 reachable from Ra, so the build uses uv instead: it fetches a standalone
 Python 3.12 into `$PREFIX/python` (upstream supports >=3.9,<3.13), creates
-`$PREFIX/venv`, and runs `uv pip install careless[cuda]==<version>` into
-it. The `[cuda]` extra pulls the NVIDIA libraries as pip wheels, so the
-same install works on GPU and CPU-only Ra nodes.
+`$PREFIX/venv`, and runs `uv pip install careless==<version>` into it.
+This is the CPU build; the GPU build (with the `[cuda]` extra and the
+cluster `cuda/12.8.1` module) is the separate `careless-gpu` module.
+
+To share one uv download cache between the careless and careless-gpu
+builds, run each with `UV_CACHE_DIR=<shared path> modbuild build <version>`
+— a cache outside PREFIX is kept, an in-PREFIX cache is removed after a
+successful install.
 
 `uv` must be on PATH on the build host (e.g. `pip install --user uv`).
 
diff --git a/careless/build b/careless/build
index 189d84a..d13569d 100755
--- a/careless/build
+++ b/careless/build
@@ -8,10 +8,11 @@ pbuild::configure() {
     # repo.anaconda.com is unreachable from Ra (GnuTLS error), so no conda.
     # uv fetches a standalone CPython instead; keep it inside PREFIX so the
     # venv's interpreter symlinks don't depend on the builder's home cache.
-    # modbuild's shell strips ~/.local/bin from PATH; override with UV=/path/to/uv if needed
     UV="${UV:-$HOME/.local/bin/uv}"
-    # ~/.cache/uv hits the home disk quota; cache under PREFIX (removed after install)
-    export UV_CACHE_DIR="$PREFIX/.uv-cache"
+    # ~/.cache/uv hits the home disk quota; default the cache into PREFIX.
+    # Overridable so careless and careless-gpu can share one cache:
+    #   UV_CACHE_DIR=<shared path> modbuild build <version>
+    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
     export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
     "$UV" venv --python 3.12 "$PREFIX/venv"
 }
@@ -22,8 +23,9 @@ pbuild::compile() {
 
 pbuild::install() {
     UV="${UV:-$HOME/.local/bin/uv}"
-    export UV_CACHE_DIR="$PREFIX/.uv-cache"
-    # [cuda] extra ships the NVIDIA libs via pip wheels; works on CPU-only nodes too
-    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
-    rm -rf "$UV_CACHE_DIR"
+    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
+    # CPU version; the GPU build with the [cuda] extra is the careless-gpu module
+    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
+    # only auto-remove the cache if it lives inside this PREFIX (never a shared one)
+    [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
 }
diff --git a/careless/modulefile b/careless/modulefile
index 54ec7d0..852264c 100644
--- a/careless/modulefile
+++ b/careless/modulefile
@@ -7,9 +7,8 @@ module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
 
 module-help     "
 careless: scaling and merging of X-ray diffraction data using
-approximate Bayesian inference (TensorFlow). Installed with the
-[cuda] extra, so it uses NVIDIA GPUs when available and falls
-back to CPU otherwise.
+approximate Bayesian inference (TensorFlow). CPU build; for
+NVIDIA GPU nodes load careless-gpu instead.
 "
 
 setenv          CARELESS_ENV    $PREFIX/venv

From f6c2f5a65046df8cde3cf6807963e5cef996b1ed Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 11:13:19 +0200
Subject: [PATCH 06/10] Drop careless-gpu: the [cuda] pip extra install covers
 GPU and CPU nodes

The installed careless/0.5.4 venv already contains the NVIDIA wheels
([cuda] extra), which need only the node driver and fall back to CPU,
so a separate GPU module and cuda module dep are unnecessary. Revert
careless/build to careless[cuda] to match the actual install.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 careless-gpu/README.md         | 28 ----------------------------
 careless-gpu/build             | 26 --------------------------
 careless-gpu/files/config.yaml | 13 -------------
 careless-gpu/modulefile        | 16 ----------------
 careless/README.md             | 12 ++++--------
 careless/build                 | 10 +++++-----
 careless/modulefile            |  5 +++--
 7 files changed, 12 insertions(+), 98 deletions(-)
 delete mode 100644 careless-gpu/README.md
 delete mode 100755 careless-gpu/build
 delete mode 100644 careless-gpu/files/config.yaml
 delete mode 100644 careless-gpu/modulefile

diff --git a/careless-gpu/README.md b/careless-gpu/README.md
deleted file mode 100644
index fab39aa..0000000
--- a/careless-gpu/README.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# careless-gpu
-
-GPU build of careless (see `../careless/` for the CPU module and the full
-background). Scaling and merging of X-ray diffraction data with approximate
-Bayesian inference (variational merging), built on TensorFlow.
-
-- Upstream: https://github.com/rs-station/careless
-- License: MIT
-
-## How it differs from the careless module
-
-- CUDA is not pip-installed (no `[cuda]` extra, saves several GB): the
-  cluster `cuda/12.8.1` module provides it instead, declared as
-  `build_requires`/`runtime_deps` (same pattern as `jfjoch_viewer`).
-  `runtime_deps` matters because TensorFlow dlopens the CUDA libraries
-  when careless runs, so cuda must auto-load with this module.
-- If careless reports missing cuDNN at runtime, the cuda module does not
-  ship it; the fallback is installing `careless[cuda]` in `build` instead.
-
-The venv layout, uv usage, and cache handling are identical to `careless`.
-To share one download cache between both builds, run each with
-`UV_CACHE_DIR=<shared path> modbuild build <version>` — a cache outside
-PREFIX is kept, an in-PREFIX cache is removed after a successful install.
-
-## Adding a new version
-
-Add the version under `versions:` in `files/config.yaml`, then on a Ra
-login node: `modbuild build <version>` (modbuild/2.1.2).
diff --git a/careless-gpu/build b/careless-gpu/build
deleted file mode 100755
index 49cb4e9..0000000
--- a/careless-gpu/build
+++ /dev/null
@@ -1,26 +0,0 @@
-#!/usr/bin/env modbuild
-
-pbuild::prep() {
-    :
-}
-
-pbuild::configure() {
-    # same uv pattern as the careless (CPU) module; see that build for the whys
-    UV="${UV:-$HOME/.local/bin/uv}"
-    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
-    export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
-    "$UV" venv --python 3.12 "$PREFIX/venv"
-}
-
-pbuild::compile() {
-    :
-}
-
-pbuild::install() {
-    UV="${UV:-$HOME/.local/bin/uv}"
-    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
-    # no [cuda] extra: CUDA comes from the cuda/12.8.1 module
-    # (build_requires/runtime_deps in files/config.yaml), not from pip wheels
-    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
-    [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
-}
diff --git a/careless-gpu/files/config.yaml b/careless-gpu/files/config.yaml
deleted file mode 100644
index 1a7f199..0000000
--- a/careless-gpu/files/config.yaml
+++ /dev/null
@@ -1,13 +0,0 @@
----
-format: 1
-careless-gpu:
-  defaults:
-    group: MX
-    overlay: base
-    relstage: unstable
-  versions:
-    0.5.4:
-      config:
-        relstage: unstable
-        build_requires: ["cuda/12.8.1"]
-        runtime_deps: ["cuda/12.8.1"]
diff --git a/careless-gpu/modulefile b/careless-gpu/modulefile
deleted file mode 100644
index 486cb39..0000000
--- a/careless-gpu/modulefile
+++ /dev/null
@@ -1,16 +0,0 @@
-#%Module1.0
-
-module-whatis       "careless (GPU) - merging crystallography data with variational inference"
-module-url          "https://github.com/rs-station/careless"
-module-license      "MIT"
-module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
-
-module-help     "
-careless: scaling and merging of X-ray diffraction data using
-approximate Bayesian inference (TensorFlow). GPU build: CUDA is
-provided by the cuda/12.8.1 module, auto-loaded as a runtime dep.
-Use on nodes with an NVIDIA GPU; for CPU nodes load careless instead.
-"
-
-setenv          CARELESS_ENV    $PREFIX/venv
-prepend-path    PATH            $PREFIX/venv/bin
diff --git a/careless/README.md b/careless/README.md
index 757adc9..084cb8a 100644
--- a/careless/README.md
+++ b/careless/README.md
@@ -11,14 +11,10 @@ inference (variational merging), built on TensorFlow.
 Upstream recommends a fresh conda env, but repo.anaconda.com is not
 reachable from Ra, so the build uses uv instead: it fetches a standalone
 Python 3.12 into `$PREFIX/python` (upstream supports >=3.9,<3.13), creates
-`$PREFIX/venv`, and runs `uv pip install careless==<version>` into it.
-This is the CPU build; the GPU build (with the `[cuda]` extra and the
-cluster `cuda/12.8.1` module) is the separate `careless-gpu` module.
-
-To share one uv download cache between the careless and careless-gpu
-builds, run each with `UV_CACHE_DIR=<shared path> modbuild build <version>`
-— a cache outside PREFIX is kept, an in-PREFIX cache is removed after a
-successful install.
+`$PREFIX/venv`, and runs `uv pip install careless[cuda]==<version>` into
+it. The `[cuda]` extra ships the NVIDIA libraries (CUDA, cuDNN, ...) as
+pip wheels, so the one install uses the GPU on GPU nodes (only the node's
+NVIDIA driver is needed — no cuda module) and falls back to CPU elsewhere.
 
 `uv` must be on PATH on the build host (e.g. `pip install --user uv`).
 
diff --git a/careless/build b/careless/build
index d13569d..f5ae5b0 100755
--- a/careless/build
+++ b/careless/build
@@ -9,9 +9,8 @@ pbuild::configure() {
     # uv fetches a standalone CPython instead; keep it inside PREFIX so the
     # venv's interpreter symlinks don't depend on the builder's home cache.
     UV="${UV:-$HOME/.local/bin/uv}"
-    # ~/.cache/uv hits the home disk quota; default the cache into PREFIX.
-    # Overridable so careless and careless-gpu can share one cache:
-    #   UV_CACHE_DIR=<shared path> modbuild build <version>
+    # ~/.cache/uv hits the home disk quota; default the cache into PREFIX
+    # (overridable: UV_CACHE_DIR=<path> modbuild build <version>)
     export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
     export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
     "$UV" venv --python 3.12 "$PREFIX/venv"
@@ -24,8 +23,9 @@ pbuild::compile() {
 pbuild::install() {
     UV="${UV:-$HOME/.local/bin/uv}"
     export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
-    # CPU version; the GPU build with the [cuda] extra is the careless-gpu module
-    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
+    # [cuda] extra ships the NVIDIA stack (CUDA, cuDNN, ...) as pip wheels:
+    # uses the GPU when present (needs only the node driver), else falls back to CPU
+    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
     # only auto-remove the cache if it lives inside this PREFIX (never a shared one)
     [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
 }
diff --git a/careless/modulefile b/careless/modulefile
index 852264c..50cf9be 100644
--- a/careless/modulefile
+++ b/careless/modulefile
@@ -7,8 +7,9 @@ module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
 
 module-help     "
 careless: scaling and merging of X-ray diffraction data using
-approximate Bayesian inference (TensorFlow). CPU build; for
-NVIDIA GPU nodes load careless-gpu instead.
+approximate Bayesian inference (TensorFlow). Installed with the
+[cuda] extra: uses NVIDIA GPUs when present (no cuda module
+needed), falls back to CPU otherwise.
 "
 
 setenv          CARELESS_ENV    $PREFIX/venv

From cc44565e353a4ed59dc2ba455cc69a6a16752e50 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 11:14:13 +0200
Subject: [PATCH 07/10] Revert "Drop careless-gpu: the [cuda] pip extra install
 covers GPU and CPU nodes"

This reverts commit f6c2f5a65046df8cde3cf6807963e5cef996b1ed.
---
 careless-gpu/README.md         | 28 ++++++++++++++++++++++++++++
 careless-gpu/build             | 26 ++++++++++++++++++++++++++
 careless-gpu/files/config.yaml | 13 +++++++++++++
 careless-gpu/modulefile        | 16 ++++++++++++++++
 careless/README.md             | 12 ++++++++----
 careless/build                 | 10 +++++-----
 careless/modulefile            |  5 ++---
 7 files changed, 98 insertions(+), 12 deletions(-)
 create mode 100644 careless-gpu/README.md
 create mode 100755 careless-gpu/build
 create mode 100644 careless-gpu/files/config.yaml
 create mode 100644 careless-gpu/modulefile

diff --git a/careless-gpu/README.md b/careless-gpu/README.md
new file mode 100644
index 0000000..fab39aa
--- /dev/null
+++ b/careless-gpu/README.md
@@ -0,0 +1,28 @@
+# careless-gpu
+
+GPU build of careless (see `../careless/` for the CPU module and the full
+background). Scaling and merging of X-ray diffraction data with approximate
+Bayesian inference (variational merging), built on TensorFlow.
+
+- Upstream: https://github.com/rs-station/careless
+- License: MIT
+
+## How it differs from the careless module
+
+- CUDA is not pip-installed (no `[cuda]` extra, saves several GB): the
+  cluster `cuda/12.8.1` module provides it instead, declared as
+  `build_requires`/`runtime_deps` (same pattern as `jfjoch_viewer`).
+  `runtime_deps` matters because TensorFlow dlopens the CUDA libraries
+  when careless runs, so cuda must auto-load with this module.
+- If careless reports missing cuDNN at runtime, the cuda module does not
+  ship it; the fallback is installing `careless[cuda]` in `build` instead.
+
+The venv layout, uv usage, and cache handling are identical to `careless`.
+To share one download cache between both builds, run each with
+`UV_CACHE_DIR=<shared path> modbuild build <version>` — a cache outside
+PREFIX is kept, an in-PREFIX cache is removed after a successful install.
+
+## Adding a new version
+
+Add the version under `versions:` in `files/config.yaml`, then on a Ra
+login node: `modbuild build <version>` (modbuild/2.1.2).
diff --git a/careless-gpu/build b/careless-gpu/build
new file mode 100755
index 0000000..49cb4e9
--- /dev/null
+++ b/careless-gpu/build
@@ -0,0 +1,26 @@
+#!/usr/bin/env modbuild
+
+pbuild::prep() {
+    :
+}
+
+pbuild::configure() {
+    # same uv pattern as the careless (CPU) module; see that build for the whys
+    UV="${UV:-$HOME/.local/bin/uv}"
+    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
+    export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
+    "$UV" venv --python 3.12 "$PREFIX/venv"
+}
+
+pbuild::compile() {
+    :
+}
+
+pbuild::install() {
+    UV="${UV:-$HOME/.local/bin/uv}"
+    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
+    # no [cuda] extra: CUDA comes from the cuda/12.8.1 module
+    # (build_requires/runtime_deps in files/config.yaml), not from pip wheels
+    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
+    [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
+}
diff --git a/careless-gpu/files/config.yaml b/careless-gpu/files/config.yaml
new file mode 100644
index 0000000..1a7f199
--- /dev/null
+++ b/careless-gpu/files/config.yaml
@@ -0,0 +1,13 @@
+---
+format: 1
+careless-gpu:
+  defaults:
+    group: MX
+    overlay: base
+    relstage: unstable
+  versions:
+    0.5.4:
+      config:
+        relstage: unstable
+        build_requires: ["cuda/12.8.1"]
+        runtime_deps: ["cuda/12.8.1"]
diff --git a/careless-gpu/modulefile b/careless-gpu/modulefile
new file mode 100644
index 0000000..486cb39
--- /dev/null
+++ b/careless-gpu/modulefile
@@ -0,0 +1,16 @@
+#%Module1.0
+
+module-whatis       "careless (GPU) - merging crystallography data with variational inference"
+module-url          "https://github.com/rs-station/careless"
+module-license      "MIT"
+module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
+
+module-help     "
+careless: scaling and merging of X-ray diffraction data using
+approximate Bayesian inference (TensorFlow). GPU build: CUDA is
+provided by the cuda/12.8.1 module, auto-loaded as a runtime dep.
+Use on nodes with an NVIDIA GPU; for CPU nodes load careless instead.
+"
+
+setenv          CARELESS_ENV    $PREFIX/venv
+prepend-path    PATH            $PREFIX/venv/bin
diff --git a/careless/README.md b/careless/README.md
index 084cb8a..757adc9 100644
--- a/careless/README.md
+++ b/careless/README.md
@@ -11,10 +11,14 @@ inference (variational merging), built on TensorFlow.
 Upstream recommends a fresh conda env, but repo.anaconda.com is not
 reachable from Ra, so the build uses uv instead: it fetches a standalone
 Python 3.12 into `$PREFIX/python` (upstream supports >=3.9,<3.13), creates
-`$PREFIX/venv`, and runs `uv pip install careless[cuda]==<version>` into
-it. The `[cuda]` extra ships the NVIDIA libraries (CUDA, cuDNN, ...) as
-pip wheels, so the one install uses the GPU on GPU nodes (only the node's
-NVIDIA driver is needed — no cuda module) and falls back to CPU elsewhere.
+`$PREFIX/venv`, and runs `uv pip install careless==<version>` into it.
+This is the CPU build; the GPU build (with the `[cuda]` extra and the
+cluster `cuda/12.8.1` module) is the separate `careless-gpu` module.
+
+To share one uv download cache between the careless and careless-gpu
+builds, run each with `UV_CACHE_DIR=<shared path> modbuild build <version>`
+— a cache outside PREFIX is kept, an in-PREFIX cache is removed after a
+successful install.
 
 `uv` must be on PATH on the build host (e.g. `pip install --user uv`).
 
diff --git a/careless/build b/careless/build
index f5ae5b0..d13569d 100755
--- a/careless/build
+++ b/careless/build
@@ -9,8 +9,9 @@ pbuild::configure() {
     # uv fetches a standalone CPython instead; keep it inside PREFIX so the
     # venv's interpreter symlinks don't depend on the builder's home cache.
     UV="${UV:-$HOME/.local/bin/uv}"
-    # ~/.cache/uv hits the home disk quota; default the cache into PREFIX
-    # (overridable: UV_CACHE_DIR=<path> modbuild build <version>)
+    # ~/.cache/uv hits the home disk quota; default the cache into PREFIX.
+    # Overridable so careless and careless-gpu can share one cache:
+    #   UV_CACHE_DIR=<shared path> modbuild build <version>
     export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
     export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
     "$UV" venv --python 3.12 "$PREFIX/venv"
@@ -23,9 +24,8 @@ pbuild::compile() {
 pbuild::install() {
     UV="${UV:-$HOME/.local/bin/uv}"
     export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
-    # [cuda] extra ships the NVIDIA stack (CUDA, cuDNN, ...) as pip wheels:
-    # uses the GPU when present (needs only the node driver), else falls back to CPU
-    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
+    # CPU version; the GPU build with the [cuda] extra is the careless-gpu module
+    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
     # only auto-remove the cache if it lives inside this PREFIX (never a shared one)
     [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
 }
diff --git a/careless/modulefile b/careless/modulefile
index 50cf9be..852264c 100644
--- a/careless/modulefile
+++ b/careless/modulefile
@@ -7,9 +7,8 @@ module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
 
 module-help     "
 careless: scaling and merging of X-ray diffraction data using
-approximate Bayesian inference (TensorFlow). Installed with the
-[cuda] extra: uses NVIDIA GPUs when present (no cuda module
-needed), falls back to CPU otherwise.
+approximate Bayesian inference (TensorFlow). CPU build; for
+NVIDIA GPU nodes load careless-gpu instead.
 "
 
 setenv          CARELESS_ENV    $PREFIX/venv

From 7544d0d952f8c6d6e2099663c8e7366a6a73eea4 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 11:14:19 +0200
Subject: [PATCH 08/10] Revert "Split careless into CPU module and careless-gpu
 using cuda/12.8.1 module"

This reverts commit b55d6c1d91536f480b604c40dd8affe3d44307d8.
---
 careless-gpu/README.md         | 28 ----------------------------
 careless-gpu/build             | 26 --------------------------
 careless-gpu/files/config.yaml | 13 -------------
 careless-gpu/modulefile        | 16 ----------------
 careless/README.md             | 11 +++--------
 careless/build                 | 16 +++++++---------
 careless/modulefile            |  5 +++--
 7 files changed, 13 insertions(+), 102 deletions(-)
 delete mode 100644 careless-gpu/README.md
 delete mode 100755 careless-gpu/build
 delete mode 100644 careless-gpu/files/config.yaml
 delete mode 100644 careless-gpu/modulefile

diff --git a/careless-gpu/README.md b/careless-gpu/README.md
deleted file mode 100644
index fab39aa..0000000
--- a/careless-gpu/README.md
+++ /dev/null
@@ -1,28 +0,0 @@
-# careless-gpu
-
-GPU build of careless (see `../careless/` for the CPU module and the full
-background). Scaling and merging of X-ray diffraction data with approximate
-Bayesian inference (variational merging), built on TensorFlow.
-
-- Upstream: https://github.com/rs-station/careless
-- License: MIT
-
-## How it differs from the careless module
-
-- CUDA is not pip-installed (no `[cuda]` extra, saves several GB): the
-  cluster `cuda/12.8.1` module provides it instead, declared as
-  `build_requires`/`runtime_deps` (same pattern as `jfjoch_viewer`).
-  `runtime_deps` matters because TensorFlow dlopens the CUDA libraries
-  when careless runs, so cuda must auto-load with this module.
-- If careless reports missing cuDNN at runtime, the cuda module does not
-  ship it; the fallback is installing `careless[cuda]` in `build` instead.
-
-The venv layout, uv usage, and cache handling are identical to `careless`.
-To share one download cache between both builds, run each with
-`UV_CACHE_DIR=<shared path> modbuild build <version>` — a cache outside
-PREFIX is kept, an in-PREFIX cache is removed after a successful install.
-
-## Adding a new version
-
-Add the version under `versions:` in `files/config.yaml`, then on a Ra
-login node: `modbuild build <version>` (modbuild/2.1.2).
diff --git a/careless-gpu/build b/careless-gpu/build
deleted file mode 100755
index 49cb4e9..0000000
--- a/careless-gpu/build
+++ /dev/null
@@ -1,26 +0,0 @@
-#!/usr/bin/env modbuild
-
-pbuild::prep() {
-    :
-}
-
-pbuild::configure() {
-    # same uv pattern as the careless (CPU) module; see that build for the whys
-    UV="${UV:-$HOME/.local/bin/uv}"
-    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
-    export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
-    "$UV" venv --python 3.12 "$PREFIX/venv"
-}
-
-pbuild::compile() {
-    :
-}
-
-pbuild::install() {
-    UV="${UV:-$HOME/.local/bin/uv}"
-    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
-    # no [cuda] extra: CUDA comes from the cuda/12.8.1 module
-    # (build_requires/runtime_deps in files/config.yaml), not from pip wheels
-    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
-    [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
-}
diff --git a/careless-gpu/files/config.yaml b/careless-gpu/files/config.yaml
deleted file mode 100644
index 1a7f199..0000000
--- a/careless-gpu/files/config.yaml
+++ /dev/null
@@ -1,13 +0,0 @@
----
-format: 1
-careless-gpu:
-  defaults:
-    group: MX
-    overlay: base
-    relstage: unstable
-  versions:
-    0.5.4:
-      config:
-        relstage: unstable
-        build_requires: ["cuda/12.8.1"]
-        runtime_deps: ["cuda/12.8.1"]
diff --git a/careless-gpu/modulefile b/careless-gpu/modulefile
deleted file mode 100644
index 486cb39..0000000
--- a/careless-gpu/modulefile
+++ /dev/null
@@ -1,16 +0,0 @@
-#%Module1.0
-
-module-whatis       "careless (GPU) - merging crystallography data with variational inference"
-module-url          "https://github.com/rs-station/careless"
-module-license      "MIT"
-module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
-
-module-help     "
-careless: scaling and merging of X-ray diffraction data using
-approximate Bayesian inference (TensorFlow). GPU build: CUDA is
-provided by the cuda/12.8.1 module, auto-loaded as a runtime dep.
-Use on nodes with an NVIDIA GPU; for CPU nodes load careless instead.
-"
-
-setenv          CARELESS_ENV    $PREFIX/venv
-prepend-path    PATH            $PREFIX/venv/bin
diff --git a/careless/README.md b/careless/README.md
index 757adc9..b2df471 100644
--- a/careless/README.md
+++ b/careless/README.md
@@ -11,14 +11,9 @@ inference (variational merging), built on TensorFlow.
 Upstream recommends a fresh conda env, but repo.anaconda.com is not
 reachable from Ra, so the build uses uv instead: it fetches a standalone
 Python 3.12 into `$PREFIX/python` (upstream supports >=3.9,<3.13), creates
-`$PREFIX/venv`, and runs `uv pip install careless==<version>` into it.
-This is the CPU build; the GPU build (with the `[cuda]` extra and the
-cluster `cuda/12.8.1` module) is the separate `careless-gpu` module.
-
-To share one uv download cache between the careless and careless-gpu
-builds, run each with `UV_CACHE_DIR=<shared path> modbuild build <version>`
-— a cache outside PREFIX is kept, an in-PREFIX cache is removed after a
-successful install.
+`$PREFIX/venv`, and runs `uv pip install careless[cuda]==<version>` into
+it. The `[cuda]` extra pulls the NVIDIA libraries as pip wheels, so the
+same install works on GPU and CPU-only Ra nodes.
 
 `uv` must be on PATH on the build host (e.g. `pip install --user uv`).
 
diff --git a/careless/build b/careless/build
index d13569d..189d84a 100755
--- a/careless/build
+++ b/careless/build
@@ -8,11 +8,10 @@ pbuild::configure() {
     # repo.anaconda.com is unreachable from Ra (GnuTLS error), so no conda.
     # uv fetches a standalone CPython instead; keep it inside PREFIX so the
     # venv's interpreter symlinks don't depend on the builder's home cache.
+    # modbuild's shell strips ~/.local/bin from PATH; override with UV=/path/to/uv if needed
     UV="${UV:-$HOME/.local/bin/uv}"
-    # ~/.cache/uv hits the home disk quota; default the cache into PREFIX.
-    # Overridable so careless and careless-gpu can share one cache:
-    #   UV_CACHE_DIR=<shared path> modbuild build <version>
-    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
+    # ~/.cache/uv hits the home disk quota; cache under PREFIX (removed after install)
+    export UV_CACHE_DIR="$PREFIX/.uv-cache"
     export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
     "$UV" venv --python 3.12 "$PREFIX/venv"
 }
@@ -23,9 +22,8 @@ pbuild::compile() {
 
 pbuild::install() {
     UV="${UV:-$HOME/.local/bin/uv}"
-    export UV_CACHE_DIR="${UV_CACHE_DIR:-$PREFIX/.uv-cache}"
-    # CPU version; the GPU build with the [cuda] extra is the careless-gpu module
-    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless==${V_PKG}"
-    # only auto-remove the cache if it lives inside this PREFIX (never a shared one)
-    [[ "$UV_CACHE_DIR" == "$PREFIX"* ]] && rm -rf "$UV_CACHE_DIR" || :
+    export UV_CACHE_DIR="$PREFIX/.uv-cache"
+    # [cuda] extra ships the NVIDIA libs via pip wheels; works on CPU-only nodes too
+    "$UV" pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
+    rm -rf "$UV_CACHE_DIR"
 }
diff --git a/careless/modulefile b/careless/modulefile
index 852264c..54ec7d0 100644
--- a/careless/modulefile
+++ b/careless/modulefile
@@ -7,8 +7,9 @@ module-maintainer   "MX Data <jiaxin.duan@psi.ch>"
 
 module-help     "
 careless: scaling and merging of X-ray diffraction data using
-approximate Bayesian inference (TensorFlow). CPU build; for
-NVIDIA GPU nodes load careless-gpu instead.
+approximate Bayesian inference (TensorFlow). Installed with the
+[cuda] extra, so it uses NVIDIA GPUs when available and falls
+back to CPU otherwise.
 "
 
 setenv          CARELESS_ENV    $PREFIX/venv

From a1e15f765db809578270d3989f7afb3f8266209b Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 11:17:38 +0200
Subject: [PATCH 09/10] use common cache dir to save disk space and time

---
 careless/build | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/careless/build b/careless/build
index 189d84a..99f5894 100755
--- a/careless/build
+++ b/careless/build
@@ -11,7 +11,7 @@ pbuild::configure() {
     # modbuild's shell strips ~/.local/bin from PATH; override with UV=/path/to/uv if needed
     UV="${UV:-$HOME/.local/bin/uv}"
     # ~/.cache/uv hits the home disk quota; cache under PREFIX (removed after install)
-    export UV_CACHE_DIR="$PREFIX/.uv-cache"
+    export UV_CACHE_DIR="/opt/psi/MX/.uv-cache"
     export UV_PYTHON_INSTALL_DIR="$PREFIX/python"
     "$UV" venv --python 3.12 "$PREFIX/venv"
 }
@@ -22,8 +22,7 @@ pbuild::compile() {
 
 pbuild::install() {
     UV="${UV:-$HOME/.local/bin/uv}"
-    export UV_CACHE_DIR="$PREFIX/.uv-cache"
+    export UV_CACHE_DIR="/opt/psi/MX/.uv-cache"
     # [cuda] extra ships the NVIDIA libs via pip wheels; works on CPU-only nodes too
     "$UV" pip install --python "$PREFIX/venv/bin/python" "careless[cuda]==${V_PKG}"
-    rm -rf "$UV_CACHE_DIR"
 }

From b02b3e36f3de7ea8f0c4531eb01ba7e2e509f2c9 Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Tue, 18 Aug 2026 11:52:48 +0200
Subject: [PATCH 10/10] Document install verification checklist in careless
 README

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 careless/README.md | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)

diff --git a/careless/README.md b/careless/README.md
index b2df471..ff05205 100644
--- a/careless/README.md
+++ b/careless/README.md
@@ -21,6 +21,40 @@ Loading the module puts the venv's `bin/` on `PATH` (the `careless`
 command is directly available, no activation needed) and sets
 `CARELESS_ENV` to the venv path.
 
+## Verifying the install
+
+Run on a Ra login node after building (in order, cheap to expensive):
+
+```sh
+# 1. module is visible and loads
+module search careless
+module load careless/<version>
+
+# 2. modulefile wiring: PATH + CARELESS_ENV
+which careless          # -> .../careless/<version>/venv/bin/careless
+echo $CARELESS_ENV      # -> .../venv
+
+# 3. venv is self-contained: must resolve inside the module PREFIX,
+#    not the builder's home cache (which may be cleaned later)
+readlink -f "$CARELESS_ENV/bin/python"
+
+# 4. CLI entry point; imports the full TensorFlow stack
+careless --help          # usage text, no traceback
+
+# 5. installed version matches the one requested
+"$CARELESS_ENV/bin/python" -c "import careless; print(careless.__version__)"
+
+# 6. GPU pickup (GPU node only; [] on a CPU node is fine)
+"$CARELESS_ENV/bin/python" -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"
+```
+
+Optional end-to-end smoke test with a small unmerged mtz (a few seconds;
+proves the TF graph builds and the optimizer steps):
+
+```sh
+careless mono dHKL <file.mtz> /tmp/careless_test --iterations 10
+```
+
 ## Adding a new version
 
 Add the version under `versions:` in `files/config.yaml`, then on a Ra