launcher/docker-images/glm53-kindling/glm53

#!/bin/bash
# docker compose with the model and every overlay in experimental/compose, in
# the order the README's numbers were measured with (later files win).
#
#   ./glm53 up -d                       start
#   ./glm53 down --timeout 60           stop
#   ./glm53 logs -f | ps | config       anything else docker compose takes
#   ./glm53 -f compose/ext.yaml up -d   your own files, after the overlays
#   ./glm53 --stock up -d               compose/glm53.yaml alone (stock kernels;
#                                       the entrypoint also wants ALLOW_STOCK=1)
#
# On a host running dispramd (kindling-spark-os) it also adds
# experimental/compose/dispram.yaml, which puts the tail of the KV pool in the
# GB10's display carveout. DISPRAM=0 leaves it out.
set -e
cd "$(dirname "$0")"
overlays=(arx snapshot adaptive-k fp8 megamoe fixes sp recoverssm)
files=(-f compose/glm53.yaml)
extra=()
stock=0
while [ $# -gt 0 ]; do
  case $1 in
    --stock) stock=1; shift ;;
    -f) extra+=(-f "$2"); shift 2 ;;
    *) break ;;
  esac
done
if [ $stock = 0 ]; then
  for o in "${overlays[@]}"; do files+=(-f "experimental/compose/$o.yaml"); done
  if [ "${DISPRAM:-1}" != 0 ] && [ -S /run/dispram/dispram.sock ] && [ -d /opt/kindling/dispram/python ]; then
    files+=(-f experimental/compose/dispram.yaml)
  fi
fi
exec docker compose "${files[@]}" "${extra[@]}" "$@"