hupenglong1
57f5b60a3e
- 新增 scripts/resolve_run_ids.sh：从 workflow5/ 解析出 SFT_RUNDIC / EVAL_RUNDIC，禁止 agent 心算 +1 - submit_sft_via_cml.sh 拆成 <SFT_RUNDIC> <EVAL_RUNDIC> [PREV_RUNDIC] 三参，yaml 用 SFT_RUNDIC，cml workflow run 用 EVAL_RUNDIC - program.md §746 / §5.2 / §1.2 同步约束，并加 R2 落盘多 +1 的踩坑案例 - §4.0.1 Step A.5 / §4.5 label-master 复核改逐条调用（≤8 路并发），禁止批量塞多条 query Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
125 lines
4.3 KiB
Bash
Executable File
125 lines
4.3 KiB
Bash
Executable File
#!/bin/bash
|
||
# 用 cml custom_train submit 提交 SFT 训练，训练完成后自动起 cml workflow run 评测
|
||
#
|
||
# 用法：
|
||
# ./submit_sft_via_cml.sh <SFT_RUNDIC> <EVAL_RUNDIC> [PREV_RUNDIC]
|
||
#
|
||
# 推荐调用方式（runDic 由 resolve_run_ids.sh 决定，不要手敲）：
|
||
# eval "$(./scripts/resolve_run_ids.sh)"
|
||
# ./scripts/submit_sft_via_cml.sh "$SFT_RUNDIC" "$EVAL_RUNDIC"
|
||
#
|
||
# 语义（与 program.md §746 + §5.2 对齐）：
|
||
# SFT_RUNDIC = R{n-1}.runDic（augment / yaml / 归档全用这个，不 +1）
|
||
# EVAL_RUNDIC = R{n-1}.runDic + 1（仅本步起 eval 时使用，仅在这一处 +1）
|
||
# PREV_RUNDIC = SFT_RUNDIC - 1（默认；只用作上一轮 sft_output 改名）
|
||
|
||
set -euo pipefail
|
||
|
||
SFT_RUNDIC=${1:?usage: $0 <SFT_RUNDIC> <EVAL_RUNDIC> [PREV_RUNDIC]}
|
||
EVAL_RUNDIC=${2:?usage: $0 <SFT_RUNDIC> <EVAL_RUNDIC> [PREV_RUNDIC]}
|
||
PREV_RUNDIC=${3:-$((SFT_RUNDIC-1))}
|
||
|
||
# 防御：EVAL_RUNDIC 必须严格大于 SFT_RUNDIC，否则一定是调用方算错
|
||
if [ "$EVAL_RUNDIC" -le "$SFT_RUNDIC" ]; then
|
||
echo "[cml-sft] ❌ EVAL_RUNDIC ($EVAL_RUNDIC) 必须 > SFT_RUNDIC ($SFT_RUNDIC)" >&2
|
||
exit 1
|
||
fi
|
||
|
||
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||
ROOT=${AUTORESEARCH_ROOT:-/mnt/wangsenhao/autoresearch-zk}
|
||
MODEL_OLD=${MODEL_OLD:-/mnt/wangsenhao/verl_zk/qwen4b_cispo_wokl_add_bvt_2/global_step_5/actor/huggingface}
|
||
TPL=${SFT_TRAIN_JOB_TEMPLATE:-$SCRIPT_DIR/sft_train_job.yaml.tpl}
|
||
YAML=/tmp/sft_train_job_r${SFT_RUNDIC}.yaml
|
||
|
||
if [ -f ~/.cloudml-cli/.profile ]; then
|
||
source ~/.cloudml-cli/.profile
|
||
else
|
||
echo "[cml-sft] ❌ 未找到 ~/.cloudml-cli/.profile，请先安装并初始化 cml" >&2
|
||
exit 1
|
||
fi
|
||
|
||
# 1. 渲染 yaml 模板（用 SFT_RUNDIC，不是 EVAL_RUNDIC）
|
||
sed \
|
||
-e "s|{RUNDIC}|${SFT_RUNDIC}|g" \
|
||
-e "s|{PREV_RUNDIC}|${PREV_RUNDIC}|g" \
|
||
-e "s|{AUTORESEARCH_ROOT}|${ROOT}|g" \
|
||
"$TPL" > "$YAML"
|
||
echo "[cml-sft] yaml: $YAML (SFT_RUNDIC=$SFT_RUNDIC EVAL_RUNDIC=$EVAL_RUNDIC PREV_RUNDIC=$PREV_RUNDIC)"
|
||
|
||
# 2. 提交训练任务
|
||
SUBMIT_OUT=$(cml custom_train submit --filename "$YAML" 2>&1)
|
||
echo "$SUBMIT_OUT"
|
||
JOB_ID=$(echo "$SUBMIT_OUT" | grep -oE 't-[0-9]+-[a-z0-9]+' | head -1)
|
||
if [ -z "$JOB_ID" ]; then
|
||
echo "[cml-sft] ❌ 提交失败" >&2
|
||
exit 1
|
||
fi
|
||
echo "[cml-sft] ✅ JobID: $JOB_ID"
|
||
|
||
# 3. 后台 watcher：等训练成功 → 自动起评测（用 EVAL_RUNDIC）
|
||
WATCHER_LOG=/tmp/r${SFT_RUNDIC}_logs/cml_watcher.log
|
||
mkdir -p "$(dirname "$WATCHER_LOG")"
|
||
|
||
nohup bash -c '
|
||
JOB_ID='"$JOB_ID"'
|
||
SFT_RUNDIC='"$SFT_RUNDIC"'
|
||
EVAL_RUNDIC='"$EVAL_RUNDIC"'
|
||
LOG='"$WATCHER_LOG"'
|
||
ROOT='"$ROOT"'
|
||
MODEL_NEW="$ROOT/sft_output"
|
||
MODEL_OLD='"$MODEL_OLD"'
|
||
EVAL_WORKFLOW_ID=f-20260408161444-wu3pz
|
||
EVAL_VERSION=v28
|
||
|
||
source ~/.cloudml-cli/.profile
|
||
|
||
echo "[$(date)] 等待 cml job $JOB_ID 完成..." >> "$LOG"
|
||
while true; do
|
||
STATE=$(cml custom_train describe "$JOB_ID" 2>/dev/null | grep -o "\"state\": \"[a-z]*\"" | head -1 | sed "s/.*: \"//;s/\"//")
|
||
case "$STATE" in
|
||
succeed)
|
||
echo "[$(date)] 训练成功" >> "$LOG"
|
||
break ;;
|
||
failed|killed)
|
||
echo "[$(date)] ❌ 训练 $STATE" >> "$LOG"
|
||
exit 1 ;;
|
||
*)
|
||
sleep 60 ;;
|
||
esac
|
||
done
|
||
|
||
# 验证产出
|
||
[ ! -f "$MODEL_NEW/_SUCCESS" ] && [ ! -f "$MODEL_NEW/config.json" ] && {
|
||
echo "[$(date)] ❌ 产出缺失" >> "$LOG"; exit 1
|
||
}
|
||
|
||
# 起评测（这里且仅这里用 EVAL_RUNDIC）
|
||
echo "[$(date)] 启动 CML 评测 EVAL_RUNDIC=$EVAL_RUNDIC" >> "$LOG"
|
||
cml workflow run \
|
||
--workflow_id $EVAL_WORKFLOW_ID --version $EVAL_VERSION \
|
||
--global_inputs runDic=$EVAL_RUNDIC \
|
||
--global_inputs model_path_new=$MODEL_NEW \
|
||
--global_inputs model_path_old=$MODEL_OLD >> "$LOG" 2>&1
|
||
echo "[$(date)] cml workflow run 提交完成 (workflow$EVAL_RUNDIC)" >> "$LOG"
|
||
' > /tmp/r${SFT_RUNDIC}_logs/watcher_runner.log 2>&1 &
|
||
|
||
WATCHER_PID=$!
|
||
echo "[cml-sft] watcher PID: $WATCHER_PID（log: $WATCHER_LOG）"
|
||
|
||
# 4. 立即输出可查看命令
|
||
cat <<EOF
|
||
|
||
[cml-sft] 任务提交完成，关键命令：
|
||
查看任务状态: cml custom_train describe $JOB_ID
|
||
查看实时日志: cml custom_train logs $JOB_ID --follow
|
||
停止任务: cml custom_train kill $JOB_ID
|
||
watcher log: tail -f $WATCHER_LOG
|
||
|
||
训练完成后，evaluation workflow 会自动启动；评测产物在
|
||
/mnt/xiaoai-zk-model-train-tj5/workflow5/workflow${EVAL_RUNDIC}/
|
||
|
||
JobID: $JOB_ID
|
||
SFT_RUNDIC: $SFT_RUNDIC (yaml / sft_output 命名)
|
||
EVAL_RUNDIC: $EVAL_RUNDIC (workflow 评测目录)
|
||
EOF
|