orpheus-fastapi/docker-compose-gpu-rocm.yml
wizardeur 0391696552
Added support for ROCm 6.4 (#60)
* Added support for ROCm 6.4

* Update README.md with information about ROCm support

---------

Co-authored-by: Kamil Andrusz <kamil@woospeedup.com>
2025-07-05 10:59:23 +10:00

86 lines
2 KiB
YAML

services:
orpheus-fastapi:
container_name: orpheus-fastapi
image: orpheus-tts-fastapi-server-orpheus-fastapi:latest
build:
context: .
dockerfile: Dockerfile.gpu-rocm
ports:
- "5005:5005"
env_file:
- .env
environment:
- ORPHEUS_API_URL=http://llama-cpp-server:5006/v1/completions
ipc: host
privileged: true
security_opt:
- seccomp=unconfined
cap_add:
- SYS_PTRACE
- CAP_SYS_ADMIN
devices:
- /dev/kfd
- /dev/dri
- /dev/mem
group_add:
- video
- render
# - 993 # Add numeric render/video group id(s) if your system has different group id(s) than the image - 44(video),109(render)
shm_size: 8g
restart: unless-stopped
depends_on:
llama-cpp-server:
condition: service_started
llama-cpp-server:
image: ghcr.io/ggml-org/llama.cpp:server-vulkan
ports:
- "5006:5006"
volumes:
- ./models:/models
env_file:
- .env
depends_on:
model-init:
condition: service_completed_successfully
cap_add:
- SYS_PTRACE
- CAP_SYS_ADMIN
security_opt:
- seccomp=unconfined
privileged: true
devices:
- /dev/kfd
- /dev/dri
- /dev/mem
group_add:
- video
- 993
ipc: host
shm_size: 8g
restart: unless-stopped
command: >
-m /models/${ORPHEUS_MODEL_NAME}
--port 5006
--host 0.0.0.0
--n-gpu-layers 29
--ctx-size ${ORPHEUS_MAX_TOKENS}
--n-predict ${ORPHEUS_MAX_TOKENS}
--rope-scaling linear
model-init:
image: curlimages/curl:latest
user: ${UID}:${GID}
volumes:
- ./models:/app/models
working_dir: /app
command: >
sh -c '
if [ ! -f /app/models/${ORPHEUS_MODEL_NAME} ]; then
echo "Downloading model file..."
wget -P /app/models https://huggingface.co/lex-au/${ORPHEUS_MODEL_NAME}/resolve/main/${ORPHEUS_MODEL_NAME}
else
echo "Model file already exists"
fi'
restart: "no"