Update installation instructions to reflect currently reality, and use uv's --torch-backend

Contributing.md

@@ -27,7 +27,7 @@ See [Pull Request Recommendations](#pull-request-recommendations) for hints on m
 Run these commands in each of your sandboxes to enable our pre-commit hooks and gitlint:
 
 ```sh
-pip install -r requirements.dev.txt
+pip install -e '.[dev]'
 pre-commit install
 gitlint install-hook
 git submodule foreach 'pre-commit install'
@@ -61,11 +61,11 @@ All the pre-commit hooks are executed using a tool called
 [pre-commit](https://pre-commit.com/). Once you enable pre-commit, it will run all the
 hooks each time you try to commit anything in this repo.
 
-We've listed all the developper dependencies for the project in
-[requirements.dev.txt](requirements.dev.txt) to make them easy to install:
+The developer dependencies are listed in the dev extra in [pyproject.toml](pyproject.toml)
+to make them easy to install:
 
 ```sh
-pip install -r requirements.dev.txt
+pip install -e '.[dev]'
 pre-commit install
 ```
 
@@ -132,12 +132,12 @@ commit-msg hook in Git.
 Run this command in your sandbox to install and enable the commit-msg hook:
 
 ```sh
-pip install -r requirements/requirements.dev.txt
+pip install -e '.[dev]'
 gitlint install-hook
 ```
 
-- Now, next time you make a change and commit it, your commit log will be checked:
-  - `git commit -m'non-compliant commit log text'` outputs an error
+- Now, your commit log will be checked:
+  - `git commit -m'non-compliant commit message'` outputs an error
   - `git commit -m'fix(g2p): fixing a bug in g2p integration'` works
 
 ### Initializing submodules too

docs/install.md

@@ -89,7 +89,7 @@ we maintain a script to automate the process and keep it reliable.
     conda create --name EveryVoice python=3.12 ffmpeg
     conda activate EveryVoice
     conda install sox -c conda-forge
-    CUDA_TAG=cu121 pip install -r requirements.torch.txt --find-links https://download.pytorch.org/whl/torch_stable.html
+    CUDA_TAG=cu126 pip install -r requirements.torch.txt --find-links https://download.pytorch.org/whl/torch_stable.html
     pip install -e '.[dev]'
     ```
 
@@ -99,8 +99,7 @@ we maintain a script to automate the process and keep it reliable.
     ```
     uv venv -p 3.12 .venv
     source .venv/bin/activate
-    uv pip install torch==2.3.1+cu121 torchaudio==2.3.1+cu121 --find-links https://download.pytorch.org/whl/torch_stable.html
-    uv pip install -e '.[dev]'
+    uv pip install -e '.[dev]' --torch-backend=cu126
     ```
 
 ### Option 2-b — Manual installation — Detailed
@@ -133,24 +132,25 @@ Create a new virtual environment and activate it:
     Install our pytorch requirements from `requirements.torch.txt`:
 
     ```sh
-    CUDA_TAG=cu121 pip install -r requirements.torch.txt --find-links https://download.pytorch.org/whl/torch_stable.html
+    CUDA_TAG=cu126 pip install -r requirements.torch.txt --find-links https://download.pytorch.org/whl/torch_stable.html
     ```
 
 === "Using uv"
 
-    Install our pytorch requirements specified in `requirements.torch.txt`, but manually.
-    (Unfortunately, `uv` does not support the environment variable we use with pip and conda.)
+    Since version 0.6.9, `uv` has a `--torch-backend` option for automatically
+    fetching the right pre-compiled wheels given your CUDA driver, so add it when
+    you install everyvoice itself:
 
     ```sh
-    uv pip install torch==2.3.1+cu121 torchaudio==2.3.1+cu121 --find-links https://download.pytorch.org/whl/torch_stable.html
+    uv pip install . --torch-backend=cu126
     ```
 
-Replace `cu121` above (for CUDA 12.1) by your actual CUDA version tag (cu118 or
-cu121), or by `cpu` for a CPU-only installation.
+Replace `cu126` above (for CUDA 12.1) by your actual CUDA version tag (cu118,
+cu126 or cu128), or by `cpu` for a CPU-only installation.
 
 Alternatively, you can follow the [PyTorch installation instructions](https://pytorch.org/get-started/locally/) relevant to your hardware.
-Make sure you specify the version declared in `requirements.torch.txt`, which is 2.3.1 at the moment,
-if you install EveryVoice from GitHub, but 2.1.0 if you install it from PyPI.
+Make sure you specify the version declared in `requirements.torch.txt`, which was 2.7.1 when this
+document was last updated.
 
 #### Non-Python dependencies
 
@@ -196,6 +196,8 @@ Install EveryVoice locally from your cloned sandbox:
     uv pip install -e .
     ```
 
+    (possibly with the optional `--torch-backend=cu126` option discussed above)
+
 #### Dev dependencies
 
 Before you can run the test suites, you'll also need to install the dev dependencies:

make-everyvoice-env

@@ -185,7 +185,7 @@ else
         else
             echo "Warning: Mismatched CUDA version found. Specified: CUDA_VERSION=$CUDA_VERSION. Found:"
             nvidia-smi | grep CUDA
-            echo "Please make sure the CUDA version available at runtime will match $CUDA_VERSION."
+            echo "Please make sure the CUDA version available at runtime will be compatible with $CUDA_VERSION."
         fi
     else
         echo "Please make sure the CUDA version installed on your system matches CUDA_VERSION=$CUDA_VERSION."
@@ -217,6 +217,8 @@ if [[ $USE_UV ]]; then
     r uv venv --python-preference=only-managed --python="$PYTHON_VERSION" "$ENV_PREFIX"
     r source "$ENV_PREFIX"/*/activate
 
+    # Note, adding `--torch-backend=$CUDA_TAG` to `uv pip install` would replace this block
+    # for uv>=0.6.9, but let's maintain backward compatibility in this automated script!
     if [[ $CUDA_TAG ]]; then
         TEMP_REQS_DIR=$(mktemp -d ./torch-reqs.XXXXXX)
         sed 's/\${CUDA_TAG}/'"$CUDA_TAG/" requirements.torch.txt > "$TEMP_REQS_DIR/requirements.torch.txt"