Fix ROCm support in Docker container

docker/.env.sample

@@ -11,5 +11,5 @@ INVOKEAI_ROOT=
 # HUGGING_FACE_HUB_TOKEN=
 
 ## optional variables specific to the docker setup.
-# GPU_DRIVER=cuda
-# CONTAINER_UID=1000
+# GPU_DRIVER=cuda # or rocm
+# CONTAINER_UID=1000

docker/Dockerfile

@@ -18,8 +18,8 @@ ENV INVOKEAI_SRC=/opt/invokeai
 ENV VIRTUAL_ENV=/opt/venv/invokeai
 
 ENV PATH="$VIRTUAL_ENV/bin:$PATH"
-ARG TORCH_VERSION=2.0.1
-ARG TORCHVISION_VERSION=0.15.2
+ARG TORCH_VERSION=2.1.0
+ARG TORCHVISION_VERSION=0.16
 ARG GPU_DRIVER=cuda
 ARG TARGETPLATFORM="linux/amd64"
 # unused but available
@@ -35,7 +35,7 @@ RUN --mount=type=cache,target=/root/.cache/pip \
     if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
         extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cpu"; \
     elif [ "$GPU_DRIVER" = "rocm" ]; then \
-        extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/rocm5.4.2"; \
+        extra_index_url_arg="--index-url https://download.pytorch.org/whl/rocm5.6"; \
     else \
         extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cu121"; \
     fi &&\

docker/docker-compose.yml

@@ -15,6 +15,10 @@ services:
             - driver: nvidia
               count: 1
               capabilities: [gpu]
+    # For AMD support, comment out the deploy section above and uncomment the devices section below:
+    #devices:
+    #  - /dev/kfd:/dev/kfd
+    #  - /dev/dri:/dev/dri
     build:
       context: ..
       dockerfile: docker/Dockerfile

docker/run.sh

@@ -7,5 +7,5 @@ set -e
 SCRIPTDIR=$(dirname "${BASH_SOURCE[0]}")
 cd "$SCRIPTDIR" || exit 1
 
-docker compose up --build -d
+docker compose up -d
 docker compose logs -f

pyproject.toml

@@ -80,8 +80,8 @@ dependencies = [
   "semver~=3.0.1",
   "send2trash",
   "test-tube~=0.7.5",
-  "torch~=2.0.1",
-  "torchvision~=0.15.2",
+  "torch~=2.1.0",
+  "torchvision~=0.16",
   "torchmetrics~=0.11.0",
   "torchsde~=0.2.5",
   "transformers~=4.31.0",