add support for kohyass and bugfixes

Dockerfile

@@ -13,18 +13,34 @@ RUN apt-get update &&\
     git \
     python3.10 \
     python3-dev \
+    python3.10-tk \
     python-is-python3 \
     python3.10-venv \
     rsync
 RUN curl -sS https://bootstrap.pypa.io/get-pip.py | python3.10
 RUN python3.10 -m pip install --upgrade pip wheel setuptools
 
+# Install requirements for ROCM build-stuff
+RUN apt-get install -y hipblas hipblaslt hipsparse hipcub hip-runtime-amd rocthrust rocthrust-dev rocrand
+
 # Install PyTorch
 RUN python3.10 -m pip install torch torchvision --index-url https://download.pytorch.org/whl/nightly/rocm5.7
 
 # Create ai folder for saving projects
 RUN mkdir /ai/
 
+# Install bitsandbytes, required for many AI tools, with ROCM fork/PR
+WORKDIR /ai/git/bitsandbytes
+RUN git clone https://github.com/TimDettmers/bitsandbytes.git /ai/git/bitsandbytes
+# Clone the ROCM PR branch cause it isn't merged yet
+RUN git fetch origin refs/pull/756/head:rocmport && git checkout rocmport
+# Set the env variables to the container ROCM setup
+ENV ROCM_HOME=/opt/rocm-5.7.0
+ENV ROCM_TARGET=gfx1030
+# Build and install globally
+RUN make hip && pip install .
+
+
 # Install StableDiffusion in /ai/
 WORKDIR /ai/stablediffusion-webui
 RUN git clone -b dev https://github.com/AUTOMATIC1111/stable-diffusion-webui /ai/git/stablediffusion-webui
@@ -33,28 +49,37 @@ RUN cp -R /ai/git/stablediffusion-webui/* /ai/stablediffusion-webui/
 RUN python3.10 -m venv /ai/venv/stablediffusion/ --system-site-packages
 RUN /ai/venv/stablediffusion/bin/python launch.py --skip-torch-cuda-test --exit 
 RUN /ai/venv/stablediffusion/bin/python -m pip install opencv-python-headless tomesd protobuf --upgrade
-# RUN /ai/venv/stablediffusion/bin/python -m pip install -r requirements.txt # should not be needed
+
 
 # Install Kobold AI in /ai/
 WORKDIR /ai/koboldai
 RUN git clone https://github.com/YellowRoseCx/koboldcpp-rocm.git -b main --depth 1 /ai/git/koboldai
 RUN cp -R /ai/git/koboldai/* /ai/koboldai/
-# Create VENV for KoboldAI with inherit pytorch
-# RUN python3.10 -m venv /ai/venv/koboldai/ --system-site-packages # not needed actually
-# Install python requirements
-# RUN /ai/venv/koboldai/bin/python -m pip install -r requirements.txt # should not be needed
 # Build KoboldAI for ROCM
 RUN make LLAMA_HIPBLAS=1 -j4
 
+
 # Install LlamaCPP in /ai/
 WORKDIR /ai/llamacpp
 RUN git clone https://github.com/ggerganov/llama.cpp.git -b master /ai/git/llamacpp
 RUN cp -R /ai/git/llamacpp/* /ai/llamacpp/
-# Install requirements for LlamaCPP build
-RUN apt-get install -y hipblas hipblaslt hipsparse hipcub hip-runtime-amd rocthrust rocthrust-dev rocrand
 # Build LlamaCPP for ROCM
 RUN make LLAMA_HIPBLAS=1 -j4
+RUN pip install --ignore-installed flask requests
 
+# Install KoyhaSS in /ai/
+WORKDIR /ai/kohya_ss
+# RUN apt-get install -y libgl1 libglib2.0-0 libgoogle-perftools-dev  python3-html5lib python3-apt python3.10-distutils
+RUN git clone https://github.com/bmaltais/kohya_ss.git -b master /ai/git/kohya_ss
+# For some reason, the .release file is required and isn't copied automatically
+RUN cp -R /ai/git/kohya_ss/* /ai/kohya_ss/ && cp /ai/git/kohya_ss/.release /ai/kohya_ss/
+# Create VENV for KoyhaSS with inherit pytorch
+RUN python3.10 -m venv /ai/venv/kohya_ss/ --system-site-packages
+# Install python requirements
+RUN /ai/venv/kohya_ss/bin/python -m pip install --upgrade pip wheel
+RUN /ai/venv/kohya_ss/bin/python -m pip install -r requirements.txt && /ai/venv/kohya_ss/bin/python -m pip uninstall tensorflow
+RUN /ai/venv/kohya_ss/bin/python -m pip install accelerate tensorboard tensorflow-rocm lion_pytorch
+RUN /ai/venv/kohya_ss/bin/python -m pip install typing_extensions --upgrade
 
 # Set safe directory for extensions and stuff
 RUN git config --global --add safe.directory "*"

docker-compose.yml

@@ -45,10 +45,6 @@ services:
     container_name: koboldai-rocm
     environment:
       TZ: "Europe/Zurich"
-      ROC_ENABLE_PRE_VEGA: 1
-      COMMANDLINE_ARGS: ""
-      # HSA_OVERRIDE_GFX_VERSION: 10.3.0
-      # PYTORCH_HIP_ALLOC_CONF: garbage_collection_threshold:0.8,max_split_size_mb:128
     entrypoint: ["/bin/sh", "-c"]
     working_dir: /ai/koboldai/
     command: ["./koboldcpp.py --config config.kcpps"]
@@ -67,19 +63,24 @@ services:
     volumes:
       - ./koboldai/config.kcpps:/ai/koboldai/config.kcpps
       - ./koboldai/models:/ai/koboldai/localmodels
+
   llamacpp-rocm:
     image: ai-suite-rocm:5.7
     container_name: llamacpp-rocm
     environment:
       TZ: "Europe/Zurich"
       ROC_ENABLE_PRE_VEGA: 1
-      COMMANDLINE_ARGS: "-m /ai/llamacpp/models/llama2-13b-tiefighter.Q6_K.gguf -c 512 -b 1024 -n 256 --keep 48 --repeat_penalty 1.0 --color -i -r \"User:\" -f prompts/chat-with-ellie.txt"
+      # COMMANDLINE_ARGS: "-m /ai/llamacpp/models/llama2-13b-tiefighter.Q6_K.gguf -c 512 -b 1024 -n 256 --keep 48 --repeat_penalty 1.0 --color -i -r \"User:\" -f prompts/chat-with-ellie.txt"
+      COMMANDLINE_ARGS: "-m /ai/llamacpp/models/llama2-13b-tiefighter.Q6_K.gguf -c 2048 --n-gpu-layers 40 --port 5002 --host 0.0.0.0"
       # HSA_OVERRIDE_GFX_VERSION: 10.3.0
       # PYTORCH_HIP_ALLOC_CONF: garbage_collection_threshold:0.8,max_split_size_mb:128
     entrypoint: ["/bin/sh", "-c"]
     working_dir: /ai/llamacpp/
-    command: ["./main $$COMMANDLINE_ARGS"]
+    command: ["./server $$COMMANDLINE_ARGS & python /ai/llamacpp/examples/server/api_like_OAI.py --port 5003 --host 0.0.0.0 --llama-api http://127.0.0.1:5002"]
     tty: true
+    ports:
+      - "5002:5002"
+      - "5003:5003"
     devices:
       - "/dev/kfd:/dev/kfd"
       - "/dev/dri:/dev/dri"
@@ -93,3 +94,25 @@ services:
     volumes:
       - ./llamacpp/models:/ai/llamacpp/models
       - ./llamacpp/extra:/ai/llamacpp/extra
+
+  koyhass-rocm:
+    image: ai-suite-rocm:5.7
+    container_name: koyhass-rocm
+    environment:
+      TZ: "Europe/Zurich"
+      CLI_ARGS: ""
+    entrypoint: ["/bin/sh", "-c"]
+    working_dir: /ai/kohya_ss/
+    command: ["/ai/venv/kohya_ss/bin/python \"./kohya_gui.py\" ${CLI_ARGS} --listen 0.0.0.0 --server_port 5004"]
+    ports:
+      - "5004:5004"
+    devices:
+      - "/dev/kfd:/dev/kfd"
+      - "/dev/dri:/dev/dri"
+    group_add:
+      - video
+    ipc: host
+    cap_add:
+      - SYS_PTRACE
+    security_opt:
+      - seccomp:unconfined