Claude/resync logi pyramid upstream

Dockerfile.server

@@ -0,0 +1,30 @@
+FROM busybox:1.37
+
+LABEL name=fir-server
+LABEL version=0.0.1
+
+RUN adduser -D static
+USER static
+WORKDIR /home/static/fir
+
+COPY --chown=static:static ./foxhole /home/static/fir/foxhole
+COPY --chown=static:static ./index.html /home/static/fir
+COPY --chown=static:static ./includes /home/static/fir/includes
+COPY --chown=static:static ./favicon.ico /home/static/fir
+COPY --chown=static:static LICENSE /home/static/fir
+
+ARG PORT=8000
+ENV PORT=${PORT}
+
+EXPOSE ${PORT}
+
+# Busybox's httpd doesn't support a mime type for .mjs
+# files by default, so we'll need to add it to httpd.conf
+# manually. The http daemon also treats .js files as
+# application/javascript by default. However, in
+# RFC 9239 application/javascript has been marked as
+# obselete in favor of text/javascript. So we're overriding
+# that mime type as well.
+RUN printf ".mjs:text/javascript\n.js:text/javascript\n" > ../httpd.conf
+
+CMD busybox httpd -f -v -p "$PORT" -h /home/static/fir -c /home/static/httpd.conf

Dockerfile.trainer

@@ -0,0 +1,25 @@
+# GPU-Enabled TensorFlow Image
+FROM tensorflow/tensorflow:2.16.2-gpu
+
+# Path to extracted Foxhole files
+ARG WAR_LOCATION
+ENV WAR_LOCATION=${WAR_LOCATION}
+
+# Update apt to ensure all packages are available
+RUN apt-get update --allow-insecure-repositories
+
+# Setup node and npm with nvm to run all javascript files
+SHELL ["/bin/bash", "--login", "-c", "-i"]
+RUN curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.37.2/install.sh | bash
+RUN nvm install 20.10
+RUN nvm use 20.10
+
+# Install pipenv for python
+RUN pip3 install pipenv
+
+# Install dependencies for image manipulation
+RUN apt-get install -y --allow-unauthenticated imagemagick
+RUN apt-get install -y --allow-unauthenticated optipng
+
+# Run the build script
+CMD sh build.sh ${WAR_LOCATION}

README.md

@@ -13,13 +13,60 @@ Currently a bit of a hack around fir as a proof of concept but thanks to their s
 
 Feel free to open an issue if you notice anything not working or have a suggestion on how it can be improved.
 
+## Deployment
+### Local
+To deploy a non-containerized server run:
+```
+cd fir
+python3 -m http.server
+```
+### Docker
+#### Building the Docker Container
+To build the docker container run:
+```
+docker build -f Dockerfile.server --tag 'fir_server' .
+```
+
+##### Overriding the listen port
+If you'd like to override the override the port the server listens on run:
+```
+docker build -f Dockerfile.server --build-arg PORT=<override port> --tag 'fir_server' .
+```
+#### Running the Docker Container
+To run the FIR server in the built docker continer:
+```
+docker run -p <host port>:<fir port> fir_server
+```
+The `-p` argument maps the host port to the fir server port inside the container. FIR defaults to listening on port
+8000. To override the port please see [this section](#overriding-the-listen-port).
+
 ## Development
 Standalone website:
 ```
 cd fir
 python3 -m http.server
 ```
 
+To build the google spreadsheet sidebar, run `./sundial/gs-build.sh` and find the files to be added to Google Apps Script in `./sundial/gs-build`.
+
+## Training
+
+### Standalone
+
+The standalone method will require you to manually install all the necessary dependencies, such as Node, NPM, TensorFlow, etc.
+To begin training, simply run `build.sh <FModel-Data-Directory>`.
+
+### Docker
+
+Training can also be performed using a Docker container instead. [https://docs.docker.com/desktop/features/wsl/]
+If you plan to use your GPU(s) for training, you will need to install NVIDIA drivers and the NVIDIA Container Toolkit. [https://docs.nvidia.com/ai-enterprise/deployment/vmware/latest/docker.html]
+
+Build the docker container by running docker `docker build -f Dockerfile.trainer --tag 'fir_trainer' .`
+
+If you only want to utilize your CPU for training, run `docker run -f Dockerfile.trainer -it --rm -v $PWD:/tmp -w /tmp -e WAR_LOCATION=<FModel-Data-Directory> fir_trainer`
+
+If you want to utilize both your CPU and GPU(s) for training, run `docker run -f Dockerfile.trainer --gpus all -it --rm -v $PWD:/tmp -w /tmp -e WAR_LOCATION=<FModel-Data-Directory> fir_trainer`
+
 ## License
 All original source code and contributions available under MIT License.
 

build.sh

@@ -9,7 +9,7 @@ then
 fi
 
 warLocation=$(cd "${1}"; pwd)
-version='naval-57'
+version='infantry-61'
 
 parseCatalog() {
   echo "Parsing catalog. (downloading / updating npm packages)"
@@ -30,11 +30,6 @@ generateIconTraining() {
   rangeMax=$(expr ${cpus} - 1)
   seq 0 $rangeMax | xargs -I@ -n1 -P$cpus node generate_training.js "${warLocation}" ../foxhole/${version}/catalog.json training @ $cpus
 
-  # Textured Icons mod uses the same icon for both FieldMGAmmo and MGAmmo. This
-  # confuses the model, and the icon looks more like MGAmmo, so ignore the
-  # FieldMGAmmo icon.
-  rm training/FieldMGAmmo*/textured-icons-*.png || true
-
   ./find-duplicates.sh $cpus
 
   echo "Copying each training png to jpg."
@@ -76,9 +71,10 @@ buildClassifier() {
 
   CUDNN_PATH=$(dirname $(pipenv run python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"))
   export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDNN_PATH/lib
+  export TF_FORCE_GPU_ALLOW_GROWTH=true
 
   rm -r model-tf || true
-  pipenv run python train.py 50 rgb 0.10 0.005 ../catalog/training/
+  pipenv run python train.py 100 rgb 0.20 0.005 ../catalog/training/
 
   echo "Training complete, assembling results."
   rm -r ../foxhole/${version}/classifier || true
@@ -87,7 +83,11 @@ buildClassifier() {
 
   #pipenv run python train.py 16 grayscale 0.05 0.05 quantity_training
 
-  pipenv run tensorflowjs_converter --input_format tf_saved_model --output_format=tfjs_graph_model model-tf ../foxhole/${version}/classifier
+  cd convert
+  pipenv clean
+  pipenv install
+  pipenv run tensorflowjs_converter --input_format tf_saved_model --output_format=tfjs_graph_model ../model-tf ../../foxhole/${version}/classifier
+  cd ..
 
   pipenv run python sort_json.py ../foxhole/${version}/classifier/model.json