scott
2a80555c60
All checks were successful
Build ROCm Image / build (push) Successful in 35s
PyTorch passes ptr=0 size=0 workspace to MIOpen convolutions, causing GemmFwdRest to warn and fall back to a slow path on every operation. MIOPEN_DEBUG_CONV_GEMM=0 skips GEMM entirely and uses Direct/Winograd solvers which have no workspace requirement. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
47 lines
1.5 KiB
YAML
47 lines
1.5 KiB
YAML
services:
|
|
chatterbox-whisper:
|
|
image: git.sdgarren.com/scott/rocm-chatterbox-whisper:latest
|
|
build:
|
|
context: .
|
|
dockerfile: Dockerfile.rocm
|
|
restart: unless-stopped
|
|
ports:
|
|
- "${WYOMING_PORT:-10200}:10200"
|
|
devices:
|
|
- /dev/kfd
|
|
- /dev/dri
|
|
group_add:
|
|
- video
|
|
- render
|
|
ipc: host
|
|
shm_size: 8g
|
|
security_opt:
|
|
- seccomp=unconfined
|
|
volumes:
|
|
- ./config.yaml:/app/config.yaml
|
|
- ./voices:/app/voices
|
|
- ./reference_audio:/app/reference_audio
|
|
- hf_cache:/app/hf_cache
|
|
- miopen_cache:/app/miopen_cache
|
|
environment:
|
|
- HF_HUB_ENABLE_HF_TRANSFER=1
|
|
# Disable MIOpen's GEMM convolution solver. PyTorch passes a null workspace
|
|
# pointer (ptr=0, size=0) which causes GemmFwdRest to warn and fall back on
|
|
# every conv op. Disabling it makes MIOpen go straight to Direct/Winograd
|
|
# solvers that don't require workspace allocation.
|
|
- MIOPEN_DEBUG_CONV_GEMM=0
|
|
# Persist MIOpen's kernel cache so it doesn't recompile on every start.
|
|
- MIOPEN_USER_DB_PATH=/app/miopen_cache
|
|
- MIOPEN_CUSTOM_CACHE_DIR=/app/miopen_cache
|
|
# HSA_OVERRIDE_GFX_VERSION is only needed for GPUs not officially supported
|
|
# by the ROCm version in the base image. Common values:
|
|
# 10.3.0 = RX 5000/6000 series
|
|
# 11.0.0 = RX 7000 series
|
|
# 9.0.6 = Vega
|
|
# - HSA_OVERRIDE_GFX_VERSION=10.3.0
|
|
# - HF_TOKEN=your_token_here
|
|
|
|
volumes:
|
|
hf_cache:
|
|
miopen_cache:
|