Files
pretrain_kaiyuan2b/scripts/kaiyuan2b-training/start_training.sh

93 lines
2.3 KiB
Bash
Executable File

#!/usr/bin/env bash
set -euo pipefail
SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
ARTIFACT_ROOT=${ARTIFACT_ROOT:-/apps/yi/model_training/artifacts}
RUN_STATE_DIR="${ARTIFACT_ROOT}/run_state"
LOG_DIR="${ARTIFACT_ROOT}/logs"
usage() {
cat <<'EOF'
Usage:
bash start_training.sh <model> [mode] [train_name]
Models:
gpt_smoke
qwen3_1p7b
Examples:
bash start_training.sh gpt_smoke smoke smoke_gpt
bash start_training.sh qwen3_1p7b qwen3_1p7b_smoke_yi qwen3_1p7b_smoke_yi
Environment overrides:
CHECKPOINT_KEEP_RECENT=3
CHECKPOINT_CLEANUP_INTERVAL_SECONDS=300
EXTRA_ARGS="--exit-duration-in-mins 120"
EOF
}
model=${1:-}
mode=${2:-}
train_name=${3:-}
if [ -z "$model" ] || [ "$model" = "-h" ] || [ "$model" = "--help" ]; then
usage
exit 0
fi
case "$model" in
gpt_smoke)
train_script="${SCRIPT_DIR}/training_smoke_gpt2.sh"
mode=${mode:-smoke}
train_name=${train_name:-smoke_gpt}
;;
qwen3_1p7b)
train_script="${SCRIPT_DIR}/training_smoke_qwen3_1p7b.sh"
mode=${mode:-qwen3_1p7b_smoke_yi}
train_name=${train_name:-qwen3_1p7b_smoke_yi}
;;
*)
echo "Unknown model: $model" >&2
usage >&2
exit 1
;;
esac
mkdir -p "$RUN_STATE_DIR" "$LOG_DIR"
pid_file="${RUN_STATE_DIR}/${train_name}.pid"
meta_file="${RUN_STATE_DIR}/${train_name}.env"
log_file="${LOG_DIR}/${train_name}.log"
if [ -f "$pid_file" ]; then
old_pid=$(cat "$pid_file")
if [ -n "$old_pid" ] && kill -0 "$old_pid" 2>/dev/null; then
echo "Training already appears to be running: train_name=${train_name}, pid=${old_pid}" >&2
exit 1
fi
fi
combined_extra_args="--exit-signal-handler ${EXTRA_ARGS:-}"
cd "$SCRIPT_DIR"
EXTRA_ARGS="$combined_extra_args" setsid bash "$train_script" "$mode" "$train_name" > "$log_file" 2>&1 &
pid=$!
pgid=$(ps -o pgid= -p "$pid" | tr -d ' ' || true)
printf '%s\n' "$pid" > "$pid_file"
cat > "$meta_file" <<EOF
MODEL=${model}
MODE=${mode}
TRAIN_NAME=${train_name}
PID=${pid}
PGID=${pgid}
LOG_FILE=${log_file}
TRAIN_SCRIPT=${train_script}
CHECKPOINT_KEEP_RECENT=${CHECKPOINT_KEEP_RECENT:-3}
CHECKPOINT_CLEANUP_INTERVAL_SECONDS=${CHECKPOINT_CLEANUP_INTERVAL_SECONDS:-300}
EOF
echo "Started training: model=${model}, mode=${mode}, train_name=${train_name}, pid=${pid}, pgid=${pgid:-unknown}"
echo "Log: ${log_file}"
echo "Stop: bash ${SCRIPT_DIR}/stop_training.sh ${train_name}"