yindo/rocm-automated
1
0
Fork 0
mirror of https://github.com/BillyOutlast/rocm-automated.git synced 2026-02-04 03:51:19 +01:00
Code Issues Packages Projects Releases Wiki Activity

debug build

Browse Source
This commit is contained in:
John Doe
2025-11-23 23:47:39 -05:00
parent 3ebe7d2764
commit fce375c41d
2 changed files with 25 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
comfyui-setup.sh
View File

@@ -8,9 +8,9 @@ cd comfyui-manager && pip install -r requirements.txt && cd ..
git clone -b AMD https://github.com/crystian/ComfyUI-Crystools.git
cd ComfyUI-Crystools && pip install -r requirements.txt && cd ..
# MIGraphX
git clone https://github.com/pnikolic-amd/ComfyUI_MIGraphX.git
cd ComfyUI_MIGraphX && pip install -r requirements.txt && cd ..
# MIGraphX DISABLED DUE TO CRASHES
#git clone https://github.com/pnikolic-amd/ComfyUI_MIGraphX.git
#cd ComfyUI_MIGraphX && pip install -r requirements.txt && cd ..
# Unsafe Torch
git clone https://github.com/ltdrdata/comfyui-unsafe-torch

44
docker-compose.yaml
View File

@@ -107,33 +107,33 @@ services:
- HIP_VISIBLE_DEVICES=0
- RROCR_VISIBLE_DEVICES=1
# === GPU targeting ===
- HCC_AMDGPU_TARGET="gfx1100" # Change for your GPU
- PYTORCH_ROCM_ARCH="gfx1100" # e.g., gfx1030 for RX 6800/6900
# - HCC_AMDGPU_TARGET="gfx1100" # Change for your GPU
# - PYTORCH_ROCM_ARCH="gfx1100" # e.g., gfx1030 for RX 6800/6900
# === Memory allocator tuning ===
- PYTORCH_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:6144
# - PYTORCH_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:6144
# === Precision and performance ===
- TORCH_BLAS_PREFER_HIPBLASLT=0
- TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_BACKENDS="CK,TRITON,ROCBLAS"
- TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_SEARCH_SPACE="BEST"
- TORCHINDUCTOR_FORCE_FALLBACK=0
# - TORCH_BLAS_PREFER_HIPBLASLT=0
# - TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_BACKENDS="CK,TRITON,ROCBLAS"
# - TORCHINDUCTOR_MAX_AUTOTUNE_GEMM_SEARCH_SPACE="BEST"
# - TORCHINDUCTOR_FORCE_FALLBACK=0
# === Flash Attention ===
- FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"
- FLASH_ATTENTION_BACKEND="flash_attn_triton_amd"
- FLASH_ATTENTION_TRITON_AMD_SEQ_LEN=4096
- USE_CK=ON
- TRANSFORMERS_USE_FLASH_ATTENTION=1
- TRITON_USE_ROCM=ON
- TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1
# - FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"
# - FLASH_ATTENTION_BACKEND="flash_attn_triton_amd"
# - FLASH_ATTENTION_TRITON_AMD_SEQ_LEN=4096
# - USE_CK=ON
# - TRANSFORMERS_USE_FLASH_ATTENTION=1
# - TRITON_USE_ROCM=ON
# - TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1
# === CPU threading ===
- OMP_NUM_THREADS=8
- MKL_NUM_THREADS=8
- NUMEXPR_NUM_THREADS=8
# - OMP_NUM_THREADS=8
# - MKL_NUM_THREADS=8
# - NUMEXPR_NUM_THREADS=8
# === Experimental ROCm flags ===
- HSA_ENABLE_ASYNC_COPY=1
- HSA_ENABLE_SDMA=1
- MIOPEN_FIND_MODE=2
- MIOPEN_ENABLE_CACHE=1
- MIGRAPHX_MLIR_USE_SPECIFIC_OPS="attention"
# - HSA_ENABLE_ASYNC_COPY=1
# - HSA_ENABLE_SDMA=1
# - MIOPEN_FIND_MODE=2
# - MIOPEN_ENABLE_CACHE=1
# - MIGRAPHX_MLIR_USE_SPECIFIC_OPS="attention"
ports:
- "8188:8188"
networks: