MOVZXandHinata
d6062059b4
Rename the default model from Scarlet to Hinata in the Dockerfile environment, the healthcheck, and the file that scripts/install-as-service.sh writes to /etc/default/vllm. Compose already defaults to Hinata. Sync README.md with the code. Section 6 paired port 8000 with the Compose model name, so that request fails. Document vllm-logs.sh, the abliteration, guard, and counter variables, and the new folders, so the patch count reaches 103. Add section 10 for benchmarks/bench.sh, with the defaults, the example runs, and what DEPTH changes. Drop the duplicate CUDA_VISIBLE_DEVICES and PLE_HOME rows. Add README-ID.md, the full Indonesian translation. README.md stays the source, and both files say so. Co-Authored-By: Hinata <hinata@idihore.id> Signed-off-by: MOVZX <movzx@yahoo.com>
94 lines
3.8 KiB
Bash
Executable File
94 lines
3.8 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
|
|
# One entry point for the bare-metal systemd service: install, start, stop, status, logs,
|
|
# update, uninstall. Run from anywhere; the script uses the repo it lives in.
|
|
#
|
|
# scripts/install-as-service.sh <command>
|
|
#
|
|
# install write /etc/systemd/system/vllm.service for THIS repo directory,
|
|
# write /etc/default/vllm (checkpoint path, model name) when missing,
|
|
# install the logrotate policy, daemon-reload, enable
|
|
# start|stop|restart|status|logs systemctl for the service
|
|
# update ./build.sh in this repo, then restart the service
|
|
# uninstall disable + remove the unit, the EnvironmentFile and the logrotate file
|
|
#
|
|
# Needs root (the log goes to /var/log, the unit goes to /etc/systemd/system).
|
|
# The service runs the repo where it lives: no copy, no /opt, no second build tree.
|
|
# Env: CKPT MODEL_NAME PORT override what the EnvironmentFile sets.
|
|
|
|
set -euo pipefail
|
|
|
|
HERE="$(cd "$(dirname "$0")/.." && pwd)"
|
|
UNIT=vllm
|
|
UNIT_PATH=/etc/systemd/system/$UNIT.service
|
|
DEFAULT=/etc/default/$UNIT
|
|
LOGROTATE=/etc/logrotate.d/$UNIT
|
|
LOG="${LOG:-/var/log/vllm.log}"
|
|
|
|
[ "$(id -u)" = 0 ] || { echo "run as root: sudo $0 $*"; exit 1; }
|
|
command -v systemctl >/dev/null || { echo "no systemd on this host"; exit 1; }
|
|
|
|
cmd="${1:-}"; shift || true
|
|
case "$cmd" in
|
|
install)
|
|
ckpt="${CKPT:-$(sed -n 's/^MODEL_DIR=//p' "$HERE/.env" 2>/dev/null | head -1)}"
|
|
ckpt="${ckpt:-/data/models/halt95/Swift1.5-Qwen3.8-Flash-Next-W4A16-Merlin}"
|
|
[ -f "$HERE/start.sh" ] || { echo "not the repo root: $HERE"; exit 1; }
|
|
# 1. the unit: the shipped template, with /opt/vllm replaced by this directory
|
|
sed "s|/opt/vllm|$HERE|g" "$HERE/systemd/$UNIT.service" > "$UNIT_PATH"
|
|
echo "==> unit: $UNIT_PATH (WorkingDirectory=$HERE)"
|
|
# 2. the EnvironmentFile: create once, never overwrite an edited one
|
|
if [ ! -f "$DEFAULT" ]; then
|
|
{ echo "# Read by $UNIT.service. Uncomment and edit, then: systemctl restart $UNIT"
|
|
echo "CKPT=$ckpt"
|
|
echo "MODEL_NAME=Hinata"
|
|
echo "# LOG=/var/log/vllm.log"
|
|
echo "# CACHE_ROOT=$HERE/.vllm-cache"
|
|
echo "# CUDA_VISIBLE_DEVICES=0,1,2,3"
|
|
echo "# VLLM_LOG_STATS_INTERVAL=1"
|
|
} > "$DEFAULT"
|
|
echo "==> wrote $DEFAULT: set CKPT to the checkpoint directory"
|
|
else
|
|
echo "==> kept $DEFAULT (it exists; CKPT=$(. "$DEFAULT" 2>/dev/null && echo "${CKPT:-unset}" || echo unset))"
|
|
fi
|
|
# 3. log rotation for $LOG (same policy as the shipped systemd/vllm.logrotate)
|
|
sed "s|/var/log/vllm.log|$LOG|g" "$HERE/systemd/$UNIT.logrotate" > "$LOGROTATE"
|
|
chmod 644 "$LOGROTATE"
|
|
echo "==> logrotate: $LOGROTATE"
|
|
systemctl daemon-reload
|
|
systemctl enable "$UNIT"
|
|
ck="$( . "$DEFAULT" 2>/dev/null && echo "${CKPT:-}" )"
|
|
[ -n "$ck" ] && [ -d "$ck" ] || echo "==> next: edit CKPT in $DEFAULT (the checkpoint directory), then sudo $0 start"
|
|
echo "==> installed and enabled. Start with: sudo $0 start"
|
|
;;
|
|
start) systemctl start "$UNIT"; echo "==> started; a cold start takes ~6 min: sudo $0 logs" ;;
|
|
stop) systemctl stop "$UNIT" ;;
|
|
restart) systemctl restart "$UNIT" ;;
|
|
status)
|
|
systemctl status "$UNIT" --no-pager -l | head -12 || :
|
|
echo "==> engine log: tail -f $LOG (healthy line: GPU KV cache size)"
|
|
;;
|
|
logs) exec tail -n 100 -f "$LOG" ;;
|
|
update)
|
|
echo "==> build, then restart"
|
|
( cd "$HERE" && ./build.sh )
|
|
systemctl restart "$UNIT"
|
|
echo "==> update done"
|
|
;;
|
|
uninstall)
|
|
printf 'remove the unit, the EnvironmentFile and the logrotate file? [y/N] '
|
|
read -r a
|
|
case "$a" in
|
|
[yY]*) systemctl disable --now "$UNIT" 2>/dev/null || :
|
|
rm -f "$UNIT_PATH" "$DEFAULT" "$LOGROTATE"
|
|
systemctl daemon-reload
|
|
echo "==> removed (the repo, .venv, .vllm-cache and $LOG are kept)" ;;
|
|
*) echo "cancelled" ;;
|
|
esac
|
|
;;
|
|
*)
|
|
sed -n '3,18p' "$0"
|
|
exit 1
|
|
;;
|
|
esac
|