Update README.md

.gitea/workflows/build-comfyui-cuda.yaml

@@ -0,0 +1,41 @@
+name: Docker Build and Release Upload
+
+on:
+  push:
+    branches:
+      - torchaudio-builder
+
+jobs:
+
+  # Build ComfyUI-Rocm
+  build-comfyui-cuda:
+    runs-on: alpine-linux
+    steps:
+      - name: checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to DockerHub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Log in to docker.mistox.net
+        uses: docker/login-action@v3
+        with:
+          registry: docker.mistox.net
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: build and push comfyui-cuda
+        run: |
+          docker buildx build \
+          --platform=linux/arm64 \
+          -t docker.mistox.net/comfyui-cuda-ampere:latest \
+          -t mistox/comfyui-cuda-ampere:latest \
+          -f comfyui-cuda-dockerfile \
+          --push \
+          .

.gitea/workflows/build-comfyui-rocm.yaml

@@ -0,0 +1,51 @@
+name: Docker Build and Release Upload
+
+on:
+  push:
+    branches:
+      - torchaudio-builder
+
+jobs:
+
+  # Build ComfyUI-Rocm
+  build-comfyui-rocm:
+    runs-on: alpine-linux
+    steps:
+      - name: checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Log in to DockerHub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Log in to docker.mistox.net
+        uses: docker/login-action@v3
+        with:
+          registry: docker.mistox.net
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: build and push rocm-aarch64-base
+        run: |
+          docker buildx build \
+          --platform=linux/arm64 \
+          -t docker.mistox.net/rocm-aarch64-base:latest \
+          -t mistox/rocm-aarch64-base:latest \
+          -f rocm-base-aarch64-dockerfile \
+          --push \
+          .
+
+      - name: build and push comfyui-rocm
+        run: |
+          docker buildx build \
+          --platform=linux/arm64 \
+          -t docker.mistox.net/comfyui-rocm-ampere:latest \
+          -t mistox/comfyui-rocm-ampere:latest \
+          -f comfyui-rocm-dockerfile \
+          --push \
+          .

.gitea/workflows/build-uvicorn.yaml

@@ -3,10 +3,12 @@ name: Docker Build and Release Upload
 on:
   push:
     branches:
-      - main
+      - torchaudio-builder
 
 jobs:
-  build:
+
+  # Build uvicorn
+  build-uvicorn:
     runs-on: alpine-linux
     steps:
       - name: checkout
@@ -21,20 +23,19 @@ jobs:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
-      - name: build and push comfyui
-        run: |
-          docker buildx build \
-          --platform=linux/arm64 \
-          -t mistox/comfyui-ampere \
-          -f comfyui-dockerfile \
-          --push \
-          .
+      - name: Log in to docker.mistox.net
+        uses: docker/login-action@v3
+        with:
+          registry: docker.mistox.net
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
 
       - name: build and push uvicorn
         run: |
           docker buildx build \
           --platform=linux/arm64 \
-          -t mistox/uvicorn-ampere \
+          -t docker.mistox.net/uvicorn-ampere:latest \
+          -t mistox/uvicorn-ampere:latest \
           -f uvicorn-dockerfile \
           --push \
-          .
+          .

README.md

@@ -4,5 +4,8 @@ In order to use this repos please just copy the docker-compose
   
 ONLY TESTED WITH AMPERE ALTRA USING RTX40 SERIES  
   
-https://hub.docker.com/r/mistox/uvicorn-ampere  
-https://hub.docker.com/r/mistox/comfyui-ampere
+https://hub.docker.com/r/mistox/uvicorn-ampere - working  
+https://hub.docker.com/r/mistox/comfyui-cuda-ampere - working  
+  
+https://hub.docker.com/r/mistox/rocm-aarch64-base - work in progress  
+https://hub.docker.com/r/mistox/comfyui-rocm-ampere - pending rocm-aarch64-base before testing  

comfyui-cuda-dockerfile

@@ -0,0 +1,53 @@
+# Get cuda version from https://hub.docker.com/r/nvidia/cuda
+FROM nvcr.io/nvidia/pytorch:26.04-py3 AS build-deps
+
+# Set ENV
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
+ENV PYTHONUNBUFFERED=1
+ENV DEBIAN_FRONTEND=noninteractive
+ENV USE_CUDA=1
+ENV BUILD_SOX=1
+
+# Install system packages to build torchaudio
+RUN apt update && \
+    apt upgrade -y && \
+    apt install -y \
+    libavformat-dev \
+    libavcodec-dev \
+    libavutil-dev \
+    libavdevice-dev \
+    libavfilter-dev \
+    libswresample-dev \
+    sox \
+    libsox-dev
+
+# 2. Clone and build the torchaudio stable
+RUN wget https://github.com/pytorch/audio/archive/refs/tags/v2.11.0.tar.gz && \
+    tar -xf v2.11.0.tar.gz && \
+    cd audio-2.11.0 && \
+    pip install --no-build-isolation --no-deps -v .
+
+# Set working directory
+WORKDIR /app
+
+# Clone ComfyUI
+RUN git clone https://github.com/comfyanonymous/ComfyUI.git .
+
+# Install depencies
+RUN grep -vE "^(torch|#)" requirements.txt | xargs pip install && \
+    pip install torchsde imageio controlnet_aux
+
+# Set custom Nodes
+WORKDIR /app/custom_nodes
+
+# Install ComfyUi Manager
+RUN git clone https://github.com/Comfy-Org/ComfyUI-Manager.git
+
+# Go back to root
+WORKDIR /app
+
+# Expose the default ComfyUI port
+EXPOSE 8188
+
+# Launch ComfyUI
+CMD ["python3", "main.py", "--listen", "0.0.0.0"]

comfyui-dockerfile

@@ -1,32 +0,0 @@
-# Get cuda version from https://hub.docker.com/r/nvidia/cuda
-FROM nvcr.io/nvidia/pytorch:25.05-py3 AS build-deps
-
-WORKDIR /app
-
-# Get release version from https://github.com/comfyanonymous/ComfyUI/releases
-ARG ComfyUI_Release=0.3.71
-
-RUN pip install imageio controlnet_aux
-
-RUN wget https://github.com/comfyanonymous/ComfyUI/archive/refs/tags/v${ComfyUI_Release}.tar.gz && \
-    tar xzf v${ComfyUI_Release}.tar.gz && \
-    rm v${ComfyUI_Release}.tar.gz && \
-    mv /app/ComfyUI-${ComfyUI_Release} /app/comfyui/
-
-WORKDIR /app/comfyui
-
-RUN ln -sf /usr/bin/python3 /usr/bin/python && \
-    ln -sf /usr/bin/pip3 /usr/bin/pip
-
-RUN grep -vE "^(torch|#)" requirements.txt | xargs pip install && \
-    pip install torchsde
-
-WORKDIR /app/comfyui/custom_nodes
-
-RUN git clone https://github.com/Comfy-Org/ComfyUI-Manager.git
-
-WORKDIR /app/comfyui
-
-EXPOSE 8188
-
-CMD ["python", "main.py", "--listen", "0.0.0.0"]

comfyui-rocm-dockerfile

@@ -0,0 +1,27 @@
+# Use the rocm-base-aarch64-dockerfile
+FROM mistox/rocm-aarch64-base:latest AS base
+
+# Set Workdir
+WORKDIR /app
+
+# Clone ComfyUI
+RUN git clone https://github.com/comfyanonymous/ComfyUI.git .
+
+# Install depencies
+RUN grep -vE "^(torch|#)" requirements.txt | xargs pip install && \
+    pip install torchsde imageio controlnet_aux
+
+# Set custom Nodes
+WORKDIR /app/custom_nodes
+
+# Install ComfyUi Manager
+RUN git clone https://github.com/Comfy-Org/ComfyUI-Manager.git
+
+# Go back to root
+WORKDIR /app
+
+# Expose the default ComfyUI port
+EXPOSE 8188
+
+# Launch ComfyUI
+CMD ["python3", "main.py", "--listen", "0.0.0.0"]

cuda-docker-compose.yml

@@ -1,17 +1,21 @@
 services:
   ollama:
     container_name: ai-ollama
-    runtime: nvidia
-    pull_policy: always
-    tty: true
-    restart: unless-stopped
     image: docker.io/ollama/ollama:latest
+    pull_policy: always
+    restart: unless-stopped
+    tty: true
     volumes:
       - ./data/ollama/ollama:/root/.ollama
-    ports:
-      - 7869:11434
     environment:
       - OLLAMA_KEEP_ALIVE=24h
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [gpu]
 
   ollama-webui:
     image: ghcr.io/open-webui/open-webui:main
@@ -23,7 +27,7 @@ services:
     ports:
       - 8009:8080
     environment: # https://docs.openwebui.com/getting-started/env-configuration#default_models
-      - OLLAMA_BASE_URLS=http://host.docker.internal:7869
+      - OLLAMA_BASE_URLS=http://ollama:11434
       - ENV=prod
       - WEBUI_AUTH=True
       - WEBUI_NAME=Mistox AI
@@ -37,8 +41,6 @@ services:
     image: docker.io/searxng/searxng:latest
     container_name: ai-searxng
     restart: always
-    ports:
-      - 8888:8080
     volumes:
       - ./data/searxng/config:/etc/searxng
       - ./data/searxng/data:/var/cache/searxng
@@ -58,14 +60,20 @@ services:
 
   comfyui:
     container_name: ai-comfyui
-    image: mistox/comfyui-ampere
-    runtime: nvidia
+    image: mistox/comfyui-cuda-ampere
     volumes:
       - ./data/comfyui/models:/app/comfyui/models
       - ./data/comfyui/output:/app/comfyui/output
     ports:
       - 8010:8188
     restart: always
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities: [gpu]
 
 volumes:
-  uvicorn-src:
+  uvicorn-src:

rocm-base-aarch64-dockerfile

@@ -0,0 +1,173 @@
+# Use the LTS Ubuntu repo
+FROM ubuntu:26.04 AS reqs
+# Set global build variables
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
+ENV PYTHONUNBUFFERED=1
+ENV DEBIAN_FRONTEND=noninteractive
+
+# My sad server with 96 Gbytes of ram keeps running out of ram ;(
+# I would love to use the whole 80 core CPU ;(
+ENV MAX_JOBS=5
+
+# Install libraries
+RUN apt update -y && \
+    apt upgrade -y && \
+    apt install -y --no-install-recommends \
+    ca-certificates \
+    wget \
+    curl \
+    git \
+    gfortran \
+    ninja-build \
+    g++ \
+    pkg-config \
+    xxd \
+    automake \
+    libtool \
+    libegl1-mesa-dev \
+    texinfo \
+    bison \
+    flex \
+    cmake \
+    build-essential \
+    libsimde-dev \
+    zlib1g-dev \
+    libncurses5-dev \
+    libopenblas-dev \
+    libjpeg-dev \
+    libpng-dev \
+    libavcodec-dev \
+    libavformat-dev \
+    libswscale-dev \
+    libpthreadpool-dev \
+    libgoogle-glog-dev \
+    libgtest-dev \
+    libbenchmark-dev \
+    libgdbm-dev \
+    libnss3-dev \
+    libssl-dev \
+    libreadline-dev \
+    libffi-dev \
+    libsqlite3-dev \
+    libelf-dev \
+    libpciaccess-dev \
+    protobuf-compiler \
+    libprotobuf-dev \
+    libbz2-dev \
+    liblzma-dev \
+    && rm -rf /var/lib/apt/lists/*
+    # Tested Working
+
+# Build python 3.11
+FROM reqs AS python
+WORKDIR /build/python
+RUN wget https://www.python.org/ftp/python/3.11.15/Python-3.11.15.tgz && \
+    tar -xf Python-3.11.15.tgz && \
+    cd Python-3.11.15 && \
+    ./configure --enable-optimizations --enable-shared LDFLAGS="-Wl,-rpath /usr/local/lib" && \
+    make -j 80 && \
+    make install && \
+    ldconfig /usr/local/lib && \
+    wget https://bootstrap.pypa.io/get-pip.py && \
+    python3 get-pip.py && \
+    ln -s /usr/local/bin/pip3 /bin/pip && \
+    ln -s /usr/local/bin/pip3 /bin/pip3 && \
+    ln -s /usr/local/bin/python3 /bin/python && \
+    ln -s /usr/local/bin/python3 /bin/python3.11
+    # Tested Working
+
+# Build ROCm latest
+FROM python AS rocm
+WORKDIR /build/therock
+COPY ./TheRock .
+ENV CFLAGS="-Wno-error=discarded-qualifiers -fcommon -w"
+ENV CXXFLAGS="-Wno-error=discarded-qualifiers -fcommon -w"
+RUN env INSTALL_PREFIX=/usr/local ./dockerfiles/install_pinned_patchelf.sh && \
+    pip install -r requirements.txt && \
+    cmake -S . -B build \
+    -G Ninja \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DTHEROCK_AMDGPU_TARGETS="gfx1100" \
+    -DTHEROCK_DIST_AMDGPU_FAMILIES="gfx1100" \
+    -DTHEROCK_AMDGPU_DIST_BUNDLE_NAME="gfx1100" \
+    -DLLVM_HOST_TRIPLE="aarch64-unknown-linux-gnu" \
+    -DLLVM_DEFAULT_TARGET_TRIPLE="aarch64-unknown-linux-gnu" \
+    -DCMAKE_C_FLAGS="$CFLAGS" \
+    -DCMAKE_CXX_FLAGS="$CXXFLAGS" \
+    -DLLVM_ENABLE_RUNTIMES="compiler-rt;libcxx;libcxxabi" \
+    -DLLVM_ENABLE_PROJECTS="clang;lld;flang" \
+    -DCOMPILER_RT_BUILD_SANITIZERS=OFF \
+    -DCMAKE_INSTALL_PREFIX=./install && \
+    cmake --build build -- -j${MAX_JOBS}
+    # Work in progress [ Breaking due to x86 intrinsics not ported to Aarch64 ]
+
+# Build pytorch
+FROM rocm AS pytorch
+# Setup Envronment
+ENV PYTORCH_ROCM_ARCH=1100
+ENV USE_ROCM=1
+ENV USE_CUDA=0
+ENV USE_AOTRITON=0
+ENV USE_SYSTEM_PTHREADPOOL=1
+ENV USE_SYSTEM_PROTOC=1
+ENV BUILD_CUSTOM_PROTOBUF=OFF
+ENV USE_PYTORCH_QNNPACK=OFF
+ENV USE_XNNPACK=OFF
+ENV USE_ONNX=OFF
+ENV USE_QNNPACK=OFF
+ENV USE_NNPACK=OFF
+ENV USE_KINETO=OFF
+ENV USE_OPENMP=OFF
+ENV USE_TENSORPIPE=OFF
+ENV ROCM_PATH=/usr
+ENV DEVICE_LIB_PATH=/usr/lib/llvm-21/lib/clang/21/amdgcn/bitcode
+ENV HIPFLAGS="--rocm-device-lib-path=$DEVICE_LIB_PATH"
+ENV CMAKE_ARGS="-DCMAKE_HIP_COMPILER_WORKS=1 \
+-DCMAKE_CXX_COMPILER_WORKS=1 \
+-DCMAKE_C_COMPILER_WORKS=1 \
+-DCMAKE_HIP_COMPILER_ROCM_LIB_PATH=$DEVICE_LIB_PATH \
+-DCMAKE_HIP_FLAGS=--rocm-device-lib-path=$DEVICE_LIB_PATH \
+-DCMAKE_C_COMPILER=/usr/lib/llvm-21/bin/clang \
+-DCMAKE_CXX_COMPILER=/usr/lib/llvm-21/bin/clang++ \
+-DCMAKE_HIP_COMPILER=/usr/lib/llvm-21/bin/clang++ \
+-DPython3_ROOT_DIR=/usr/local \
+-DPYTHON_EXECUTABLE=/usr/local/bin/python3.11 \
+-DPYTHON_LIBRARY=/usr/local/lib/libpython3.11.so \
+-DPYTHON_INCLUDE_DIR=/usr/local/include/python3.11"
+# Pull and build pytorch
+WORKDIR /build/pytorch
+RUN git clone https://github.com/pytorch/pytorch.git && \
+    cd pytorch && \
+    git submodule update --init --recursive && \
+    make triton && \
+    python tools/amd_build/build_amd.py && \
+    pip install -r requirements.txt && \
+    mkdir -p /opt/rocm/bin \
+    ln -s /usr/bin/hipcc /opt/rocm/bin/hipcc \
+    ln -s /usr/lib/llvm-21/bin/clang++ /usr/bin/clang++ \
+    python3 setup.py build && \
+    python3 setup.py install
+    # Failing
+
+# Build pytorch-audio
+FROM pytorch AS torchaudio
+WORKDIR /build/torchaudio
+RUN wget https://github.com/pytorch/audio/archive/refs/tags/v2.11.0.tar.gz && \
+    tar -xf v2.11.0.tar.gz && \
+    cd audio-2.11.0 && \
+    pip install --no-build-isolation --no-deps -v .
+    # Tested Working via comfyui-cuda-dockerfile
+
+# Build pytorch-vision
+FROM torchaudio AS torchvision
+WORKDIR /build/torchaudio
+ENV FORCE_CUDA=0
+RUN git clone https://github.com/pytorch/vision.git . && \
+    pip install numpy pillow ninja setuptools wheel && \
+    python setup.py bdist_wheel && \
+    pip install dist/torchvision-*.whl
+    # Not tested and not verified ROCM build
+
+# Build final stripped down image
+FROM ubuntu:26.04
+# Copy in rocm, pytorch, torchaudio, torchvision

rocm-docker-compose.yml

@@ -0,0 +1,78 @@
+services:
+  ollama:
+    container_name: ai-ollama
+    image: docker.io/ollama/ollama:latest # This is only X86_64 right now. Working on building an image for this on ROCM
+    pull_policy: always
+    restart: unless-stopped
+    tty: true
+    volumes:
+      - ./data/ollama/ollama:/root/.ollama
+    environment:
+      - OLLAMA_KEEP_ALIVE=24h
+      - HSA_OVERRIDE_GFX_VERSION=11.0.0 # This is for the RX 7900XTX
+    devices:
+      - /dev/kfd:/dev/kfd
+      - /dev/dri:/dev/dri
+
+  ollama-webui:
+    container_name: ai-ollama-webui
+    image: ghcr.io/open-webui/open-webui:main
+    volumes:
+      - ./data/ollama/ollama-webui:/app/backend/data
+    depends_on:
+      - ollama
+    ports:
+      - 8009:8080
+    environment: # https://docs.openwebui.com/getting-started/env-configuration#default_models
+      - OLLAMA_BASE_URLS=http://ollama:11434
+      - ENV=prod
+      - WEBUI_AUTH=True
+      - WEBUI_NAME=Mistox AI
+      - WEBUI_URL=http://ai.mistox.net:8080
+      - WEBUI_SECRET_KEY=aops8gh948g@#50dfg00f
+    extra_hosts:
+      - host.docker.internal:host-gateway
+    restart: unless-stopped
+
+  searxng:
+    container_name: ai-searxng
+    image: docker.io/searxng/searxng:latest
+    restart: always
+    ports:
+      - 8888:8080
+    volumes:
+      - ./data/searxng/config:/etc/searxng
+      - ./data/searxng/data:/var/cache/searxng
+
+  ai-uvicorn:
+    container_name: ai-uvicorn
+    image: mistox/uvicorn-ampere
+    ports:
+      - 8000:8000
+      - 5678:5678
+    volumes:
+      - uvicorn-src:/code/ollama-docker
+    restart: always
+    depends_on:
+      - ollama
+      - ollama-webui
+
+  comfyui:
+    container_name: ai-comfyui
+    image: mistox/comfyui-rocm-ampere
+    restart: unless-stopped
+    pull_policy: always
+    ports:
+      - 8010:8188
+    environment:
+      - HSA_OVERRIDE_GFX_VERSION=11.0.0 # This is for the RX 7900XTX
+    devices:
+      - /dev/kfd:/dev/kfd
+      - /dev/dri:/dev/dri
+    volumes:
+      - ./data/comfyui/models:/home/user/comfyui/models
+      - ./data/comfyui/output:/home/user/comfyui/output
+      - ./data/comfyui/custom_nodes:/home/user/comfyui/custom_nodes
+
+volumes:
+  uvicorn-src: