Upload 4 files

sc1/inference_clawbenchpro-Copy1.sh

@@ -0,0 +1,272 @@
+#!/bin/bash
+# Qwen3.5-27B half-turn checkpoint 在 ClawBenchPro 高质量子集上的训练同构推理。
+#
+# 唯一 rollout 链路：
+#   CustomRLHFDataset
+#     -> VERL LLMServerManager/vLLM
+#     -> VERL AgentLoopManager
+#     -> tool_agent / ToolAgentLoop
+#     -> NanoclawWorkspaceTool
+#
+# 该脚本不实现第二套 Agent，不调用 verifier/reward。它复用训练的 system
+# prompt、Qwen3-Coder XML 工具协议、9 个 workspace tools、完整多轮历史拼接、
+# workspace 生命周期、response mask 与 trajectory 持久化。
+
+set -x
+
+SCRIPT_DIR=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0708_new
+BUNDLE_ROOT=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu
+
+# ==============================================================================
+# 直接在这里填写要推理的多个模型。每项是一个普通 Bash 字符串：
+#   "唯一模型输出名|已合并 Hugging Face checkpoint 的绝对路径"
+#
+# 示例（删除行首 # 后改成实际路径）：
+MODEL_CHECKPOINTS=(
+    "qwen35_4b_step_36|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_36"
+    "qwen35_4b_step_38|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_38"
+    "qwen35_4b_step_40|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_40"
+    "qwen35_4b_step_42|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_42"
+    "qwen35_4b_step_44|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_44"
+    "qwen35_4b_step_46|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_46"
+    "qwen35_4b_step_48|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_48"
+    "qwen35_4b_step_50|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_50"
+    "qwen35_4b_step_52|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_52"
+    "qwen35_4b_step_54|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_54"
+    "qwen35_4b_step_56|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_56"
+    "qwen35_4b_step_58|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_58"
+    "qwen35_4b_step_60|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_60"
+    "qwen35_4b_step_62|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_62"
+    "qwen35_4b_step_64|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_64"
+    "qwen35_4b_step_66|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_66"
+    "qwen35_4b_step_68|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_68"
+)
+# ==============================================================================
+
+# 多 checkpoint 输入，按以下优先级解析：
+#   1. MODEL_PATH_LIST（换行分隔）及可选 MODEL_NAME_LIST；
+#   2. 兼容旧用法的单个 MODEL_PATH / MODEL_NAME；
+#   3. 上面的 MODEL_CHECKPOINTS 字符串数组（推荐日常使用）。
+# 所有路径都必须是 vLLM 可直接加载的、已合并 Hugging Face checkpoint；
+# 未合并的 VERL/FSDP shard 不能直接用于该推理入口。
+export MODEL_INPUT_VALIDATE_ONLY=${MODEL_INPUT_VALIDATE_ONLY:-0}
+
+declare -a INPUT_MODEL_PATHS=()
+declare -a INPUT_MODEL_NAMES=()
+
+if [ -n "${MODEL_PATH_LIST:-}" ]; then
+    while IFS= read -r model_path; do
+        model_path=${model_path%$'\r'}
+        if [ -n "${model_path}" ]; then
+            INPUT_MODEL_PATHS+=("${model_path}")
+        fi
+    done <<< "${MODEL_PATH_LIST}"
+    if [ -n "${MODEL_NAME_LIST:-}" ]; then
+        while IFS= read -r model_name; do
+            model_name=${model_name%$'\r'}
+            if [ -n "${model_name}" ]; then
+                INPUT_MODEL_NAMES+=("${model_name}")
+            fi
+        done <<< "${MODEL_NAME_LIST}"
+    fi
+elif [ -n "${MODEL_PATH:-}" ]; then
+    INPUT_MODEL_PATHS+=("${MODEL_PATH}")
+    if [ -n "${MODEL_NAME:-}" ]; then
+        INPUT_MODEL_NAMES+=("${MODEL_NAME}")
+    fi
+elif [ "${#MODEL_CHECKPOINTS[@]}" -gt 0 ]; then
+    for model_spec in "${MODEL_CHECKPOINTS[@]}"; do
+        if [[ "${model_spec}" != *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected \"model_name|/absolute/checkpoint/path\": ${model_spec}" >&2
+            exit 2
+        fi
+        model_name=${model_spec%%|*}
+        model_path=${model_spec#*|}
+        if [ -z "${model_name}" ] || [ -z "${model_path}" ] || [[ "${model_path}" == *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected exactly one | delimiter: ${model_spec}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAMES+=("${model_name}")
+        INPUT_MODEL_PATHS+=("${model_path}")
+    done
+fi
+
+if [ "${#INPUT_MODEL_PATHS[@]}" -eq 0 ]; then
+    echo "ERROR: no model checkpoints configured." >&2
+    echo "Edit MODEL_CHECKPOINTS at the top of this script, set MODEL_PATH_LIST, or set legacy MODEL_PATH." >&2
+    exit 2
+fi
+if [ "${#INPUT_MODEL_NAMES[@]}" -ne 0 ] && [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+    echo "ERROR: model name count ${#INPUT_MODEL_NAMES[@]} does not match path count ${#INPUT_MODEL_PATHS[@]}." >&2
+    exit 2
+fi
+
+declare -A INPUT_MODEL_NAME_SEEN=()
+for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+    model_path=${INPUT_MODEL_PATHS[model_index]}
+    if [[ "${model_path}" != /* ]]; then
+        echo "ERROR: model checkpoint path must be absolute: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -d "${model_path}" ]; then
+        echo "ERROR: model checkpoint directory not found: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -f "${model_path}/config.json" ]; then
+        echo "ERROR: merged Hugging Face config.json not found: ${model_path}/config.json" >&2
+        exit 2
+    fi
+    if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+        model_name=${INPUT_MODEL_NAMES[model_index]}
+        if [[ ! "${model_name}" =~ ^[a-zA-Z0-9._-]+$ ]]; then
+            echo "ERROR: model name may only contain letters, digits, dot, underscore and hyphen: ${model_name}" >&2
+            exit 2
+        fi
+        if [ -n "${INPUT_MODEL_NAME_SEEN[${model_name}]:-}" ]; then
+            echo "ERROR: duplicate configured model name: ${model_name}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAME_SEEN[${model_name}]=1
+    else
+        model_name='<auto>'
+    fi
+    echo "MODEL_INPUT[$model_index] name=${model_name} path=${model_path}"
+done
+
+printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_PATHS[@]}"
+export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED}
+if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+    printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_NAMES[@]}"
+    export MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}
+else
+    unset MODEL_NAME_LIST
+fi
+unset MODEL_PATH MODEL_NAME
+
+echo "MODEL_INPUT_COUNT=${#INPUT_MODEL_PATHS[@]}"
+if [ "${MODEL_INPUT_VALIDATE_ONLY}" = "1" ]; then
+    echo "MODEL_INPUT_VALIDATE_ONLY=1: model configuration is valid; inference not started."
+    exit 0
+fi
+
+# 必须把训练时修改过的整份 VERL v12 代码同步到此目录；不能只安装上游 VERL。
+export WORK_DIR=${WORK_DIR:-${BUNDLE_ROOT}/verl}
+
+# 已离线筛选并适配好的训练兼容数据：base 100 + hard 100。这里直接读取
+# manifest-backed data_* bundle，不在推理节点重新扫描或适配完整 1000 题数据。
+# 部署到共享存储后，可通过 BASE_TASKS 覆盖为共享目录中的副本。
+export BASE_TASKS=${BASE_TASKS:-${BUNDLE_ROOT}/data/ClawBenchPro_base100_hard100_quality}
+
+# v14/0710/inference.sh 只有在该变量非空时才会启动全量 ClawBenchPro 适配器。
+# 专用入口固定使用上面的精选子集，避免意外退回 991/1000 题路径。
+export CLAWBENCHPRO_ROOT=
+export CLAWBENCHPRO_ADAPTED_ROOT=
+
+# 输出严格为 OUTPUT_ROOT/<model_name>/step_1/<task_id>_sample_<n>/。
+export OUTPUT_ROOT=${OUTPUT_ROOT:-/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/output/qwen35_4b}
+export OVERWRITE_OUTPUT=${OVERWRITE_OUTPUT:-True}
+export CONTINUE_ON_MODEL_ERROR=${CONTINUE_ON_MODEL_ERROR:-1}
+export MODEL_SWITCH_COOLDOWN=${MODEL_SWITCH_COOLDOWN:-20}
+export MODEL_RESOURCE_RELEASE_TIMEOUT=${MODEL_RESOURCE_RELEASE_TIMEOUT:-600}
+
+# 默认单机 8 NPU、TP=4，即 2 个 vLLM rollout replica。多机时所有节点提交
+# 同一脚本，ModelArts 通过 VC_TASK_INDEX 区分 rank。
+export INFER_NNODES=${INFER_NNODES:-1}
+export NPUS_PER_NODE=${NPUS_PER_NODE:-8}
+export INFER_TP=${INFER_TP:-4}
+
+# 评测默认每题 1 条轨迹；如需复现训练时的 GRPO 采样数量可设为 8。
+export N_RESP_PER_PROMPT=${N_RESP_PER_PROMPT:-1}
+export PROMPT_BATCH_SIZE=${PROMPT_BATCH_SIZE:-16}
+export AGENT_NUM_WORKERS=${AGENT_NUM_WORKERS:-32}
+export CALCULATE_LOG_PROBS=${CALCULATE_LOG_PROBS:-False}
+
+# 与 v14/0708_new/half_turn.sh 的 actor rollout 完全对齐。
+export MAX_TURNS=${MAX_TURNS:-35}
+export MAX_PROMPT_LENGTH=${MAX_PROMPT_LENGTH:-8192}
+export MAX_RESPONSE_LENGTH=${MAX_RESPONSE_LENGTH:-22768}
+export MAX_ASSISTANT_RESPONSE_LENGTH=${MAX_ASSISTANT_RESPONSE_LENGTH:-16384}
+export MAX_TOOL_RESPONSE_LENGTH=${MAX_TOOL_RESPONSE_LENGTH:-8192}
+export ROLLOUT_MAX_NUM_BATCHED_TOKENS=${ROLLOUT_MAX_NUM_BATCHED_TOKENS:-16384}
+export ROLLOUT_GPU_MEMORY_UTILIZATION=${ROLLOUT_GPU_MEMORY_UTILIZATION:-0.70}
+export ROLLOUT_TEMPERATURE=${ROLLOUT_TEMPERATURE:-1.0}
+export ROLLOUT_TOP_P=${ROLLOUT_TOP_P:-0.95}
+export ROLLOUT_TOP_K=${ROLLOUT_TOP_K:-20}
+export ROLLOUT_MIN_P=${ROLLOUT_MIN_P:-0.0}
+export ROLLOUT_PRESENCE_PENALTY=${ROLLOUT_PRESENCE_PENALTY:-0.0}
+export ROLLOUT_FREQUENCY_PENALTY=${ROLLOUT_FREQUENCY_PENALTY:-0.0}
+export ROLLOUT_REPETITION_PENALTY=${ROLLOUT_REPETITION_PENALTY:-1.0}
+export ROLLOUT_FREE_CACHE_ENGINE=${ROLLOUT_FREE_CACHE_ENGINE:-True}
+export ROLLOUT_ENFORCE_EAGER=${ROLLOUT_ENFORCE_EAGER:-False}
+
+# 与训练相同：thinking actor、Qwen3-Coder parser、同一 tool YAML、受限 bash、
+# 不保存 workspace_before，并严格禁止输出路径静默追加 request-id 后缀。
+export TOOL_CONFIG_PATH=${TOOL_CONFIG_PATH:-recipe/nanoclaw/nanoclaw_tool_config.yaml}
+export NANOCLAW_MAX_STEPS=${NANOCLAW_MAX_STEPS:-}
+export NANOCLAW_CLEANUP_WORKSPACES=False
+export NANOCLAW_KEEP_FAILED_WORKSPACES=False
+export NANOCLAW_ENV_BUILDER_TIMEOUT=${NANOCLAW_ENV_BUILDER_TIMEOUT:-600}
+export NANOCLAW_ALLOW_BASH=True
+export NANOCLAW_STRICT_RESULT_DIR=True
+export NANOCLAW_SAVE_WORKSPACE_BEFORE=False
+
+# 默认安装与训练一致的 GCC/CANN/torch-npu/vLLM/Triton/VERL 依赖。
+export SETUP_ENVIRONMENT=${SETUP_ENVIRONMENT:-1}
+
+# 只重跑失败或残缺的 checkpoint。成功模型通过严格磁盘审计后会被跳过。
+export RESUME_SKIP_COMPLETED_MODELS=1
+if [ "${RESUME_SKIP_COMPLETED_MODELS}" = "1" ]; then
+    if [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+        echo "ERROR: RESUME_SKIP_COMPLETED_MODELS=1 requires an explicit name for every model." >&2
+        exit 2
+    fi
+    declare -a FILTERED_MODEL_PATHS=()
+    declare -a FILTERED_MODEL_NAMES=()
+    for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+        model_path=${INPUT_MODEL_PATHS[model_index]}
+        model_name=${INPUT_MODEL_NAMES[model_index]}
+        model_output_root="${OUTPUT_ROOT}/${model_name}"
+        if python3 - "${BASE_TASKS}/benchmark_manifest.json" "${model_output_root}" "${N_RESP_PER_PROMPT}" <<'PY'
+import json, sys
+from pathlib import Path
+manifest_path, model_root, rollout_n = Path(sys.argv[1]), Path(sys.argv[2]), int(sys.argv[3])
+def load_json(path):
+    try: value = json.loads(path.read_text(encoding="utf-8"))
+    except Exception: return None
+    return value if isinstance(value, dict) else None
+try: task_ids = [str(x["task_id"]) for x in json.loads(manifest_path.read_text(encoding="utf-8"))["tasks"]]
+except Exception as exc:
+    print(f"[resume_audit_error] {type(exc).__name__}: {exc}", file=sys.stderr); raise SystemExit(2)
+issues=[]; step_dir=model_root/"step_1"
+for task_id in task_ids:
+  for sample in range(rollout_n):
+    d=step_dir/f"{task_id}_sample_{sample}"
+    if not d.is_dir(): issues.append(f"missing_result:{d.name}"); continue
+    if not (d/"workspace_after").is_dir(): issues.append(f"missing_workspace:{d.name}")
+    m=load_json(d/"nanoclaw_metadata.json")
+    if m is None: issues.append(f"invalid_metadata:{d.name}")
+    elif m.get("status")!="ready": issues.append(f"metadata_status={m.get('status')}:{d.name}")
+    if load_json(d/"conversation_history.json") is None: issues.append(f"invalid_conversation:{d.name}")
+    if load_json(d/"trajectory.json") is None: issues.append(f"invalid_trajectory:{d.name}")
+if issues:
+    print(f"[resume_rerun_incomplete] model={model_root.name} issues={len(issues)} preview={', '.join(issues[:8])}", file=sys.stderr); raise SystemExit(1)
+print(f"[resume_skip_complete] model={model_root.name} tasks={len(task_ids)} rollout_n={rollout_n}", file=sys.stderr)
+PY
+        then
+            echo "SKIP_COMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        else
+            audit_rc=$?
+            [ "${audit_rc}" -ne 2 ] || { echo "ERROR: unable to audit ${model_name}" >&2; exit 2; }
+            FILTERED_MODEL_PATHS+=("${model_path}"); FILTERED_MODEL_NAMES+=("${model_name}")
+            echo "RERUN_INCOMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        fi
+    done
+    [ "${#FILTERED_MODEL_PATHS[@]}" -ne 0 ] || { echo "ALL_CONFIGURED_MODELS_ALREADY_COMPLETE: no inference will be submitted."; exit 0; }
+    printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_PATHS[@]}"
+    printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_NAMES[@]}"
+    export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED} MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}
+    export OVERWRITE_OUTPUT=True
+    echo "RESUME_RERUN_MODEL_COUNT=${#FILTERED_MODEL_PATHS[@]}"
+fi
+
+exec bash "/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0710/inference.sh"

sc1/inference_clawbenchpro-Copy2.sh

@@ -0,0 +1,266 @@
+#!/bin/bash
+# Qwen3.5-27B half-turn checkpoint 在 ClawBenchPro 高质量子集上的训练同构推理。
+#
+# 唯一 rollout 链路：
+#   CustomRLHFDataset
+#     -> VERL LLMServerManager/vLLM
+#     -> VERL AgentLoopManager
+#     -> tool_agent / ToolAgentLoop
+#     -> NanoclawWorkspaceTool
+#
+# 该脚本不实现第二套 Agent，不调用 verifier/reward。它复用训练的 system
+# prompt、Qwen3-Coder XML 工具协议、9 个 workspace tools、完整多轮历史拼接、
+# workspace 生命周期、response mask 与 trajectory 持久化。
+
+set -x
+
+SCRIPT_DIR=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0708_new
+BUNDLE_ROOT=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu
+
+# ==============================================================================
+# 直接在这里填写要推理的多个模型。每项是一个普通 Bash 字符串：
+#   "唯一模型输出名|已合并 Hugging Face checkpoint 的绝对路径"
+#
+# 示例（删除行首 # 后改成实际路径）：
+MODEL_CHECKPOINTS=(
+    "qwen35_9b_base|/opt/huawei/dataset/zyr_yuyin/models/Qwen/Qwen3___5-9B"
+    "qwen35_9b_step_2|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_2"
+    "qwen35_9b_step_4|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_4"
+    "qwen35_9b_step_6|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_6"
+    "qwen35_9b_step_8|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_8"
+    "qwen35_9b_step_10|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_10"
+    "qwen35_9b_step_12|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_12"
+    "qwen35_9b_step_14|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_14"
+    "qwen35_9b_step_16|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_16"
+    "qwen35_9b_step_18|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_18"
+    "qwen35_9b_step_20|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_20"
+    "qwen35_9b_step_22|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_22"
+    "qwen35_9b_step_24|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_24"
+    "qwen35_9b_step_26|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_26"
+    "qwen35_9b_step_28|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_28"
+    "qwen35_9b_step_30|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_30"
+    "qwen35_9b_step_32|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_32"
+    "qwen35_9b_step_34|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_34"
+)
+# ==============================================================================
+
+# 多 checkpoint 输入，按以下优先级解析：
+#   1. MODEL_PATH_LIST（换行分隔）及可选 MODEL_NAME_LIST；
+#   2. 兼容旧用法的单个 MODEL_PATH / MODEL_NAME；
+#   3. 上面的 MODEL_CHECKPOINTS 字符串数组（推荐日常使用）。
+# 所有路径都必须是 vLLM 可直接加载的、已合并 Hugging Face checkpoint；
+# 未合并的 VERL/FSDP shard 不能直接用于该推理入口。
+export MODEL_INPUT_VALIDATE_ONLY=${MODEL_INPUT_VALIDATE_ONLY:-0}
+
+declare -a INPUT_MODEL_PATHS=()
+declare -a INPUT_MODEL_NAMES=()
+
+if [ -n "${MODEL_PATH_LIST:-}" ]; then
+    while IFS= read -r model_path; do
+        model_path=${model_path%$'\r'}
+        if [ -n "${model_path}" ]; then
+            INPUT_MODEL_PATHS+=("${model_path}")
+        fi
+    done <<< "${MODEL_PATH_LIST}"
+    if [ -n "${MODEL_NAME_LIST:-}" ]; then
+        while IFS= read -r model_name; do
+            model_name=${model_name%$'\r'}
+            if [ -n "${model_name}" ]; then
+                INPUT_MODEL_NAMES+=("${model_name}")
+            fi
+        done <<< "${MODEL_NAME_LIST}"
+    fi
+elif [ -n "${MODEL_PATH:-}" ]; then
+    INPUT_MODEL_PATHS+=("${MODEL_PATH}")
+    if [ -n "${MODEL_NAME:-}" ]; then
+        INPUT_MODEL_NAMES+=("${MODEL_NAME}")
+    fi
+elif [ "${#MODEL_CHECKPOINTS[@]}" -gt 0 ]; then
+    for model_spec in "${MODEL_CHECKPOINTS[@]}"; do
+        if [[ "${model_spec}" != *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected \"model_name|/absolute/checkpoint/path\": ${model_spec}" >&2
+            exit 2
+        fi
+        model_name=${model_spec%%|*}
+        model_path=${model_spec#*|}
+        if [ -z "${model_name}" ] || [ -z "${model_path}" ] || [[ "${model_path}" == *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected exactly one | delimiter: ${model_spec}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAMES+=("${model_name}")
+        INPUT_MODEL_PATHS+=("${model_path}")
+    done
+fi
+
+if [ "${#INPUT_MODEL_PATHS[@]}" -eq 0 ]; then
+    echo "ERROR: no model checkpoints configured." >&2
+    echo "Edit MODEL_CHECKPOINTS at the top of this script, set MODEL_PATH_LIST, or set legacy MODEL_PATH." >&2
+    exit 2
+fi
+if [ "${#INPUT_MODEL_NAMES[@]}" -ne 0 ] && [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+    echo "ERROR: model name count ${#INPUT_MODEL_NAMES[@]} does not match path count ${#INPUT_MODEL_PATHS[@]}." >&2
+    exit 2
+fi
+
+declare -A INPUT_MODEL_NAME_SEEN=()
+for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+    model_path=${INPUT_MODEL_PATHS[model_index]}
+    if [[ "${model_path}" != /* ]]; then
+        echo "ERROR: model checkpoint path must be absolute: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -d "${model_path}" ]; then
+        echo "ERROR: model checkpoint directory not found: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -f "${model_path}/config.json" ]; then
+        echo "ERROR: merged Hugging Face config.json not found: ${model_path}/config.json" >&2
+        exit 2
+    fi
+    if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+        model_name=${INPUT_MODEL_NAMES[model_index]}
+        if [[ ! "${model_name}" =~ ^[a-zA-Z0-9._-]+$ ]]; then
+            echo "ERROR: model name may only contain letters, digits, dot, underscore and hyphen: ${model_name}" >&2
+            exit 2
+        fi
+        if [ -n "${INPUT_MODEL_NAME_SEEN[${model_name}]:-}" ]; then
+            echo "ERROR: duplicate configured model name: ${model_name}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAME_SEEN[${model_name}]=1
+    else
+        model_name='<auto>'
+    fi
+    echo "MODEL_INPUT[$model_index] name=${model_name} path=${model_path}"
+done
+
+printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_PATHS[@]}"
+export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED}
+if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+    printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_NAMES[@]}"
+    export MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}
+else
+    unset MODEL_NAME_LIST
+fi
+unset MODEL_PATH MODEL_NAME
+
+echo "MODEL_INPUT_COUNT=${#INPUT_MODEL_PATHS[@]}"
+if [ "${MODEL_INPUT_VALIDATE_ONLY}" = "1" ]; then
+    echo "MODEL_INPUT_VALIDATE_ONLY=1: model configuration is valid; inference not started."
+    exit 0
+fi
+
+# 必须把训练时修改过的整份 VERL v12 代码同步到此目录；不能只安装上游 VERL。
+export WORK_DIR=${WORK_DIR:-${BUNDLE_ROOT}/verl}
+
+# 已离线筛选并适配好的训练兼容数据：base 100 + hard 100。这里直接读取
+# manifest-backed data_* bundle，不在推理节点重新扫描或适配完整 1000 题数据。
+# 部署到共享存储后，可通过 BASE_TASKS 覆盖为共享目录中的副本。
+export BASE_TASKS=${BASE_TASKS:-${BUNDLE_ROOT}/data/ClawBenchPro_base100_hard100_quality}
+
+# v14/0710/inference.sh 只有在该变量非空时才会启动全量 ClawBenchPro 适配器。
+# 专用入口固定使用上面的精选子集，避免意外退回 991/1000 题路径。
+export CLAWBENCHPRO_ROOT=
+export CLAWBENCHPRO_ADAPTED_ROOT=
+
+# 输出严格为 OUTPUT_ROOT/<model_name>/step_1/<task_id>_sample_<n>/。
+export OUTPUT_ROOT=${OUTPUT_ROOT:-/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/output/qwen35_9b}
+export OVERWRITE_OUTPUT=${OVERWRITE_OUTPUT:-True}
+export CONTINUE_ON_MODEL_ERROR=${CONTINUE_ON_MODEL_ERROR:-1}
+export MODEL_SWITCH_COOLDOWN=${MODEL_SWITCH_COOLDOWN:-20}
+export MODEL_RESOURCE_RELEASE_TIMEOUT=${MODEL_RESOURCE_RELEASE_TIMEOUT:-600}
+
+# 默认单机 8 NPU、TP=4，即 2 个 vLLM rollout replica。多机时所有节点提交
+# 同一脚本，ModelArts 通过 VC_TASK_INDEX 区分 rank。
+export INFER_NNODES=${INFER_NNODES:-1}
+export NPUS_PER_NODE=${NPUS_PER_NODE:-8}
+export INFER_TP=${INFER_TP:-4}
+
+# 评测默认每题 1 条轨迹；如需复现训练时的 GRPO 采样数量可设为 8。
+export N_RESP_PER_PROMPT=${N_RESP_PER_PROMPT:-1}
+export PROMPT_BATCH_SIZE=${PROMPT_BATCH_SIZE:-16}
+export AGENT_NUM_WORKERS=${AGENT_NUM_WORKERS:-32}
+export CALCULATE_LOG_PROBS=${CALCULATE_LOG_PROBS:-False}
+
+# 与 v14/0708_new/half_turn.sh 的 actor rollout 完全对齐。
+export MAX_TURNS=${MAX_TURNS:-35}
+export MAX_PROMPT_LENGTH=${MAX_PROMPT_LENGTH:-8192}
+export MAX_RESPONSE_LENGTH=${MAX_RESPONSE_LENGTH:-22768}
+export MAX_ASSISTANT_RESPONSE_LENGTH=${MAX_ASSISTANT_RESPONSE_LENGTH:-16384}
+export MAX_TOOL_RESPONSE_LENGTH=${MAX_TOOL_RESPONSE_LENGTH:-8192}
+export ROLLOUT_MAX_NUM_BATCHED_TOKENS=${ROLLOUT_MAX_NUM_BATCHED_TOKENS:-16384}
+export ROLLOUT_GPU_MEMORY_UTILIZATION=${ROLLOUT_GPU_MEMORY_UTILIZATION:-0.70}
+export ROLLOUT_TEMPERATURE=${ROLLOUT_TEMPERATURE:-1.0}
+export ROLLOUT_TOP_P=${ROLLOUT_TOP_P:-0.95}
+export ROLLOUT_TOP_K=${ROLLOUT_TOP_K:-20}
+export ROLLOUT_MIN_P=${ROLLOUT_MIN_P:-0.0}
+export ROLLOUT_PRESENCE_PENALTY=${ROLLOUT_PRESENCE_PENALTY:-0.0}
+export ROLLOUT_FREQUENCY_PENALTY=${ROLLOUT_FREQUENCY_PENALTY:-0.0}
+export ROLLOUT_REPETITION_PENALTY=${ROLLOUT_REPETITION_PENALTY:-1.0}
+export ROLLOUT_FREE_CACHE_ENGINE=${ROLLOUT_FREE_CACHE_ENGINE:-True}
+export ROLLOUT_ENFORCE_EAGER=${ROLLOUT_ENFORCE_EAGER:-False}
+
+# 与训练相同：thinking actor、Qwen3-Coder parser、同一 tool YAML、受限 bash、
+# 不保存 workspace_before，并严格禁止输出路径静默追加 request-id 后缀。
+export TOOL_CONFIG_PATH=${TOOL_CONFIG_PATH:-recipe/nanoclaw/nanoclaw_tool_config.yaml}
+export NANOCLAW_MAX_STEPS=${NANOCLAW_MAX_STEPS:-}
+export NANOCLAW_CLEANUP_WORKSPACES=False
+export NANOCLAW_KEEP_FAILED_WORKSPACES=False
+export NANOCLAW_ENV_BUILDER_TIMEOUT=${NANOCLAW_ENV_BUILDER_TIMEOUT:-600}
+export NANOCLAW_ALLOW_BASH=True
+export NANOCLAW_STRICT_RESULT_DIR=True
+export NANOCLAW_SAVE_WORKSPACE_BEFORE=False
+
+# 默认安装与训练一致的 GCC/CANN/torch-npu/vLLM/Triton/VERL 依赖。
+export SETUP_ENVIRONMENT=${SETUP_ENVIRONMENT:-1}
+
+# 只重跑失败或残缺的 checkpoint。成功模型通过严格磁盘审计后会被跳过。
+export RESUME_SKIP_COMPLETED_MODELS=1
+if [ "${RESUME_SKIP_COMPLETED_MODELS}" = "1" ]; then
+    if [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+        echo "ERROR: RESUME_SKIP_COMPLETED_MODELS=1 requires an explicit name for every model." >&2
+        exit 2
+    fi
+    declare -a FILTERED_MODEL_PATHS=()
+    declare -a FILTERED_MODEL_NAMES=()
+    for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+        model_path=${INPUT_MODEL_PATHS[model_index]}; model_name=${INPUT_MODEL_NAMES[model_index]}
+        model_output_root="${OUTPUT_ROOT}/${model_name}"
+        if python3 - "${BASE_TASKS}/benchmark_manifest.json" "${model_output_root}" "${N_RESP_PER_PROMPT}" <<'PY'
+import json, sys
+from pathlib import Path
+manifest_path, model_root, rollout_n = Path(sys.argv[1]), Path(sys.argv[2]), int(sys.argv[3])
+def load_json(path):
+    try: value=json.loads(path.read_text(encoding="utf-8"))
+    except Exception: return None
+    return value if isinstance(value,dict) else None
+try: task_ids=[str(x["task_id"]) for x in json.loads(manifest_path.read_text(encoding="utf-8"))["tasks"]]
+except Exception as exc: print(f"[resume_audit_error] {type(exc).__name__}: {exc}",file=sys.stderr); raise SystemExit(2)
+issues=[]; step_dir=model_root/"step_1"
+for task_id in task_ids:
+  for sample in range(rollout_n):
+    d=step_dir/f"{task_id}_sample_{sample}"
+    if not d.is_dir(): issues.append(f"missing_result:{d.name}"); continue
+    if not (d/"workspace_after").is_dir(): issues.append(f"missing_workspace:{d.name}")
+    m=load_json(d/"nanoclaw_metadata.json")
+    if m is None: issues.append(f"invalid_metadata:{d.name}")
+    elif m.get("status")!="ready": issues.append(f"metadata_status={m.get('status')}:{d.name}")
+    if load_json(d/"conversation_history.json") is None: issues.append(f"invalid_conversation:{d.name}")
+    if load_json(d/"trajectory.json") is None: issues.append(f"invalid_trajectory:{d.name}")
+if issues: print(f"[resume_rerun_incomplete] model={model_root.name} issues={len(issues)} preview={', '.join(issues[:8])}",file=sys.stderr); raise SystemExit(1)
+print(f"[resume_skip_complete] model={model_root.name} tasks={len(task_ids)} rollout_n={rollout_n}",file=sys.stderr)
+PY
+        then echo "SKIP_COMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        else
+            audit_rc=$?; [ "${audit_rc}" -ne 2 ] || { echo "ERROR: unable to audit ${model_name}" >&2; exit 2; }
+            FILTERED_MODEL_PATHS+=("${model_path}"); FILTERED_MODEL_NAMES+=("${model_name}")
+            echo "RERUN_INCOMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        fi
+    done
+    [ "${#FILTERED_MODEL_PATHS[@]}" -ne 0 ] || { echo "ALL_CONFIGURED_MODELS_ALREADY_COMPLETE: no inference will be submitted."; exit 0; }
+    printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_PATHS[@]}"; printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_NAMES[@]}"
+    export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED} MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}; export OVERWRITE_OUTPUT=True
+    echo "RESUME_RERUN_MODEL_COUNT=${#FILTERED_MODEL_PATHS[@]}"
+fi
+
+exec bash "/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0710/inference.sh"

sc1/inference_clawbenchpro-Copy3.sh

@@ -0,0 +1,262 @@
+#!/bin/bash
+# Qwen3.5-27B half-turn checkpoint 在 ClawBenchPro 高质量子集上的训练同构推理。
+#
+# 唯一 rollout 链路：
+#   CustomRLHFDataset
+#     -> VERL LLMServerManager/vLLM
+#     -> VERL AgentLoopManager
+#     -> tool_agent / ToolAgentLoop
+#     -> NanoclawWorkspaceTool
+#
+# 该脚本不实现第二套 Agent，不调用 verifier/reward。它复用训练的 system
+# prompt、Qwen3-Coder XML 工具协议、9 个 workspace tools、完整多轮历史拼接、
+# workspace 生命周期、response mask 与 trajectory 持久化。
+
+set -x
+
+SCRIPT_DIR=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0708_new
+BUNDLE_ROOT=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu
+
+# ==============================================================================
+# 直接在这里填写要推理的多个模型。每项是一个普通 Bash 字符串：
+#   "唯一模型输出名|已合并 Hugging Face checkpoint 的绝对路径"
+#
+# 示例（删除行首 # 后改成实际路径）：
+MODEL_CHECKPOINTS=(
+    "qwen35_9b_step_36|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_36"
+    "qwen35_9b_step_38|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_38"
+    "qwen35_9b_step_40|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_40"
+    "qwen35_9b_step_42|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_42"
+    "qwen35_9b_step_44|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_44"
+    "qwen35_9b_step_46|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_46"
+    "qwen35_9b_step_48|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_48"
+    "qwen35_9b_step_50|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_50"
+    "qwen35_9b_step_52|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_52"
+    "qwen35_9b_step_54|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_54"
+    "qwen35_9b_step_56|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_56"
+    "qwen35_9b_step_58|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_58"
+    "qwen35_9b_step_60|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_60"
+    "qwen35_9b_step_62|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_62"
+    "qwen35_9b_step_64|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_64"
+    "qwen35_9b_step_66|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_66"
+    "qwen35_9b_step_68|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-9b_31k/qwen3.5-9b_31k_step_68"
+)
+# ==============================================================================
+
+# 多 checkpoint 输入，按以下优先级解析：
+#   1. MODEL_PATH_LIST（换行分隔）及可选 MODEL_NAME_LIST；
+#   2. 兼容旧用法的单个 MODEL_PATH / MODEL_NAME；
+#   3. 上面的 MODEL_CHECKPOINTS 字符串数组（推荐日常使用）。
+# 所有路径都必须是 vLLM 可直接加载的、已合并 Hugging Face checkpoint；
+# 未合并的 VERL/FSDP shard 不能直接用于该推理入口。
+export MODEL_INPUT_VALIDATE_ONLY=${MODEL_INPUT_VALIDATE_ONLY:-0}
+
+declare -a INPUT_MODEL_PATHS=()
+declare -a INPUT_MODEL_NAMES=()
+
+if [ -n "${MODEL_PATH_LIST:-}" ]; then
+    while IFS= read -r model_path; do
+        model_path=${model_path%$'\r'}
+        if [ -n "${model_path}" ]; then
+            INPUT_MODEL_PATHS+=("${model_path}")
+        fi
+    done <<< "${MODEL_PATH_LIST}"
+    if [ -n "${MODEL_NAME_LIST:-}" ]; then
+        while IFS= read -r model_name; do
+            model_name=${model_name%$'\r'}
+            if [ -n "${model_name}" ]; then
+                INPUT_MODEL_NAMES+=("${model_name}")
+            fi
+        done <<< "${MODEL_NAME_LIST}"
+    fi
+elif [ -n "${MODEL_PATH:-}" ]; then
+    INPUT_MODEL_PATHS+=("${MODEL_PATH}")
+    if [ -n "${MODEL_NAME:-}" ]; then
+        INPUT_MODEL_NAMES+=("${MODEL_NAME}")
+    fi
+elif [ "${#MODEL_CHECKPOINTS[@]}" -gt 0 ]; then
+    for model_spec in "${MODEL_CHECKPOINTS[@]}"; do
+        if [[ "${model_spec}" != *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected \"model_name|/absolute/checkpoint/path\": ${model_spec}" >&2
+            exit 2
+        fi
+        model_name=${model_spec%%|*}
+        model_path=${model_spec#*|}
+        if [ -z "${model_name}" ] || [ -z "${model_path}" ] || [[ "${model_path}" == *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected exactly one | delimiter: ${model_spec}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAMES+=("${model_name}")
+        INPUT_MODEL_PATHS+=("${model_path}")
+    done
+fi
+
+if [ "${#INPUT_MODEL_PATHS[@]}" -eq 0 ]; then
+    echo "ERROR: no model checkpoints configured." >&2
+    echo "Edit MODEL_CHECKPOINTS at the top of this script, set MODEL_PATH_LIST, or set legacy MODEL_PATH." >&2
+    exit 2
+fi
+if [ "${#INPUT_MODEL_NAMES[@]}" -ne 0 ] && [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+    echo "ERROR: model name count ${#INPUT_MODEL_NAMES[@]} does not match path count ${#INPUT_MODEL_PATHS[@]}." >&2
+    exit 2
+fi
+
+declare -A INPUT_MODEL_NAME_SEEN=()
+for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+    model_path=${INPUT_MODEL_PATHS[model_index]}
+    if [[ "${model_path}" != /* ]]; then
+        echo "ERROR: model checkpoint path must be absolute: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -d "${model_path}" ]; then
+        echo "ERROR: model checkpoint directory not found: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -f "${model_path}/config.json" ]; then
+        echo "ERROR: merged Hugging Face config.json not found: ${model_path}/config.json" >&2
+        exit 2
+    fi
+    if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+        model_name=${INPUT_MODEL_NAMES[model_index]}
+        if [[ ! "${model_name}" =~ ^[a-zA-Z0-9._-]+$ ]]; then
+            echo "ERROR: model name may only contain letters, digits, dot, underscore and hyphen: ${model_name}" >&2
+            exit 2
+        fi
+        if [ -n "${INPUT_MODEL_NAME_SEEN[${model_name}]:-}" ]; then
+            echo "ERROR: duplicate configured model name: ${model_name}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAME_SEEN[${model_name}]=1
+    else
+        model_name='<auto>'
+    fi
+    echo "MODEL_INPUT[$model_index] name=${model_name} path=${model_path}"
+done
+
+printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_PATHS[@]}"
+export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED}
+if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+    printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_NAMES[@]}"
+    export MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}
+else
+    unset MODEL_NAME_LIST
+fi
+unset MODEL_PATH MODEL_NAME
+
+echo "MODEL_INPUT_COUNT=${#INPUT_MODEL_PATHS[@]}"
+if [ "${MODEL_INPUT_VALIDATE_ONLY}" = "1" ]; then
+    echo "MODEL_INPUT_VALIDATE_ONLY=1: model configuration is valid; inference not started."
+    exit 0
+fi
+
+# 必须把训练时修改过的整份 VERL v12 代码同步到此目录；不能只安装上游 VERL。
+export WORK_DIR=${WORK_DIR:-${BUNDLE_ROOT}/verl}
+
+# 已离线筛选并适配好的训练兼容数据：base 100 + hard 100。这里直接读取
+# manifest-backed data_* bundle，不在推理节点重新扫描或适配完整 1000 题数据。
+# 部署到共享存储后，可通过 BASE_TASKS 覆盖为共享目录中的副本。
+export BASE_TASKS=${BASE_TASKS:-${BUNDLE_ROOT}/data/ClawBenchPro_base100_hard100_quality}
+
+# v14/0710/inference.sh 只有在该变量非空时才会启动全量 ClawBenchPro 适配器。
+# 专用入口固定使用上面的精选子集，避免意外退回 991/1000 题路径。
+export CLAWBENCHPRO_ROOT=
+export CLAWBENCHPRO_ADAPTED_ROOT=
+
+# 输出严格为 OUTPUT_ROOT/<model_name>/step_1/<task_id>_sample_<n>/。
+export OUTPUT_ROOT=${OUTPUT_ROOT:-/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/output/qwen35_9b}
+export OVERWRITE_OUTPUT=${OVERWRITE_OUTPUT:-True}
+export CONTINUE_ON_MODEL_ERROR=${CONTINUE_ON_MODEL_ERROR:-1}
+export MODEL_SWITCH_COOLDOWN=${MODEL_SWITCH_COOLDOWN:-20}
+export MODEL_RESOURCE_RELEASE_TIMEOUT=${MODEL_RESOURCE_RELEASE_TIMEOUT:-600}
+
+# 默认单机 8 NPU、TP=4，即 2 个 vLLM rollout replica。多机时所有节点提交
+# 同一脚本，ModelArts 通过 VC_TASK_INDEX 区分 rank。
+export INFER_NNODES=${INFER_NNODES:-1}
+export NPUS_PER_NODE=${NPUS_PER_NODE:-8}
+export INFER_TP=${INFER_TP:-4}
+
+# 评测默认每题 1 条轨迹；如需复现训练时的 GRPO 采样数量可设为 8。
+export N_RESP_PER_PROMPT=${N_RESP_PER_PROMPT:-1}
+export PROMPT_BATCH_SIZE=${PROMPT_BATCH_SIZE:-16}
+export AGENT_NUM_WORKERS=${AGENT_NUM_WORKERS:-32}
+export CALCULATE_LOG_PROBS=${CALCULATE_LOG_PROBS:-False}
+
+# 与 v14/0708_new/half_turn.sh 的 actor rollout 完全对齐。
+export MAX_TURNS=${MAX_TURNS:-35}
+export MAX_PROMPT_LENGTH=${MAX_PROMPT_LENGTH:-8192}
+export MAX_RESPONSE_LENGTH=${MAX_RESPONSE_LENGTH:-22768}
+export MAX_ASSISTANT_RESPONSE_LENGTH=${MAX_ASSISTANT_RESPONSE_LENGTH:-16384}
+export MAX_TOOL_RESPONSE_LENGTH=${MAX_TOOL_RESPONSE_LENGTH:-8192}
+export ROLLOUT_MAX_NUM_BATCHED_TOKENS=${ROLLOUT_MAX_NUM_BATCHED_TOKENS:-16384}
+export ROLLOUT_GPU_MEMORY_UTILIZATION=${ROLLOUT_GPU_MEMORY_UTILIZATION:-0.70}
+export ROLLOUT_TEMPERATURE=${ROLLOUT_TEMPERATURE:-1.0}
+export ROLLOUT_TOP_P=${ROLLOUT_TOP_P:-0.95}
+export ROLLOUT_TOP_K=${ROLLOUT_TOP_K:-20}
+export ROLLOUT_MIN_P=${ROLLOUT_MIN_P:-0.0}
+export ROLLOUT_PRESENCE_PENALTY=${ROLLOUT_PRESENCE_PENALTY:-0.0}
+export ROLLOUT_FREQUENCY_PENALTY=${ROLLOUT_FREQUENCY_PENALTY:-0.0}
+export ROLLOUT_REPETITION_PENALTY=${ROLLOUT_REPETITION_PENALTY:-1.0}
+export ROLLOUT_FREE_CACHE_ENGINE=${ROLLOUT_FREE_CACHE_ENGINE:-True}
+export ROLLOUT_ENFORCE_EAGER=${ROLLOUT_ENFORCE_EAGER:-False}
+
+# 与训练相同：thinking actor、Qwen3-Coder parser、同一 tool YAML、受限 bash、
+# 不保存 workspace_before，并严格禁止输出路径静默追加 request-id 后缀。
+export TOOL_CONFIG_PATH=${TOOL_CONFIG_PATH:-recipe/nanoclaw/nanoclaw_tool_config.yaml}
+export NANOCLAW_MAX_STEPS=${NANOCLAW_MAX_STEPS:-}
+export NANOCLAW_CLEANUP_WORKSPACES=False
+export NANOCLAW_KEEP_FAILED_WORKSPACES=False
+export NANOCLAW_ENV_BUILDER_TIMEOUT=${NANOCLAW_ENV_BUILDER_TIMEOUT:-600}
+export NANOCLAW_ALLOW_BASH=True
+export NANOCLAW_STRICT_RESULT_DIR=True
+export NANOCLAW_SAVE_WORKSPACE_BEFORE=False
+
+# 默认安装与训练一致的 GCC/CANN/torch-npu/vLLM/Triton/VERL 依赖。
+export SETUP_ENVIRONMENT=${SETUP_ENVIRONMENT:-1}
+
+# 只重跑失败或残缺的 checkpoint。成功模型通过严格磁盘审计后会被跳过。
+export RESUME_SKIP_COMPLETED_MODELS=1
+if [ "${RESUME_SKIP_COMPLETED_MODELS}" = "1" ]; then
+    if [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+        echo "ERROR: RESUME_SKIP_COMPLETED_MODELS=1 requires an explicit name for every model." >&2
+        exit 2
+    fi
+    declare -a FILTERED_MODEL_PATHS=(); declare -a FILTERED_MODEL_NAMES=()
+    for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+        model_path=${INPUT_MODEL_PATHS[model_index]}; model_name=${INPUT_MODEL_NAMES[model_index]}; model_output_root="${OUTPUT_ROOT}/${model_name}"
+        if python3 - "${BASE_TASKS}/benchmark_manifest.json" "${model_output_root}" "${N_RESP_PER_PROMPT}" <<'PY'
+import json,sys
+from pathlib import Path
+manifest_path,model_root,rollout_n=Path(sys.argv[1]),Path(sys.argv[2]),int(sys.argv[3])
+def load_json(p):
+    try:v=json.loads(p.read_text(encoding="utf-8"))
+    except Exception:return None
+    return v if isinstance(v,dict) else None
+try:task_ids=[str(x["task_id"]) for x in json.loads(manifest_path.read_text(encoding="utf-8"))["tasks"]]
+except Exception as exc:print(f"[resume_audit_error] {type(exc).__name__}: {exc}",file=sys.stderr);raise SystemExit(2)
+issues=[];step_dir=model_root/"step_1"
+for task_id in task_ids:
+  for sample in range(rollout_n):
+    d=step_dir/f"{task_id}_sample_{sample}"
+    if not d.is_dir():issues.append(f"missing_result:{d.name}");continue
+    if not (d/"workspace_after").is_dir():issues.append(f"missing_workspace:{d.name}")
+    m=load_json(d/"nanoclaw_metadata.json")
+    if m is None:issues.append(f"invalid_metadata:{d.name}")
+    elif m.get("status")!="ready":issues.append(f"metadata_status={m.get('status')}:{d.name}")
+    if load_json(d/"conversation_history.json") is None:issues.append(f"invalid_conversation:{d.name}")
+    if load_json(d/"trajectory.json") is None:issues.append(f"invalid_trajectory:{d.name}")
+if issues:print(f"[resume_rerun_incomplete] model={model_root.name} issues={len(issues)} preview={', '.join(issues[:8])}",file=sys.stderr);raise SystemExit(1)
+print(f"[resume_skip_complete] model={model_root.name} tasks={len(task_ids)} rollout_n={rollout_n}",file=sys.stderr)
+PY
+        then echo "SKIP_COMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        else
+            audit_rc=$?; [ "${audit_rc}" -ne 2 ] || { echo "ERROR: unable to audit ${model_name}" >&2; exit 2; }
+            FILTERED_MODEL_PATHS+=("${model_path}"); FILTERED_MODEL_NAMES+=("${model_name}"); echo "RERUN_INCOMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        fi
+    done
+    [ "${#FILTERED_MODEL_PATHS[@]}" -ne 0 ] || { echo "ALL_CONFIGURED_MODELS_ALREADY_COMPLETE: no inference will be submitted."; exit 0; }
+    printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_PATHS[@]}"; printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_NAMES[@]}"
+    export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED} MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}; export OVERWRITE_OUTPUT=True
+    echo "RESUME_RERUN_MODEL_COUNT=${#FILTERED_MODEL_PATHS[@]}"
+fi
+
+exec bash "/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0710/inference.sh"

sc1/inference_clawbenchpro.sh

@@ -0,0 +1,317 @@
+#!/bin/bash
+# Qwen3.5-27B half-turn checkpoint 在 ClawBenchPro 高质量子集上的训练同构推理。
+#
+# 唯一 rollout 链路：
+#   CustomRLHFDataset
+#     -> VERL LLMServerManager/vLLM
+#     -> VERL AgentLoopManager
+#     -> tool_agent / ToolAgentLoop
+#     -> NanoclawWorkspaceTool
+#
+# 该脚本不实现第二套 Agent，不调用 verifier/reward。它复用训练的 system
+# prompt、Qwen3-Coder XML 工具协议、9 个 workspace tools、完整多轮历史拼接、
+# workspace 生命周期、response mask 与 trajectory 持久化。
+
+set -x
+
+SCRIPT_DIR=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0708_new
+BUNDLE_ROOT=/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu
+
+# ==============================================================================
+# 直接在这里填写要推理的多个模型。每项是一个普通 Bash 字符串：
+#   "唯一模型输出名|已合并 Hugging Face checkpoint 的绝对路径"
+#
+# 示例（删除行首 # 后改成实际路径）：
+MODEL_CHECKPOINTS=(
+    "qwen35_4b_base|/opt/huawei/dataset/zyr_yuyin/models/Qwen/Qwen3___5-4B"
+    "qwen35_4b_step_2|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_2"
+    "qwen35_4b_step_4|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_4"
+    "qwen35_4b_step_6|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_6"
+    "qwen35_4b_step_8|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_8"
+    "qwen35_4b_step_10|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_10"
+    "qwen35_4b_step_12|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_12"
+    "qwen35_4b_step_14|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_14"
+    "qwen35_4b_step_16|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_16"
+    "qwen35_4b_step_18|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_18"
+    "qwen35_4b_step_20|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_20"
+    "qwen35_4b_step_22|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_22"
+    "qwen35_4b_step_24|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_24"
+    "qwen35_4b_step_26|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_26"
+    "qwen35_4b_step_28|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_28"
+    "qwen35_4b_step_30|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_30"
+    "qwen35_4b_step_32|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_32"
+    "qwen35_4b_step_34|/opt/huawei/dataset/zyr_yuyin/lyf/verl-nanoclaw-rl/nanoclawRL_temp_ckpt_hugging_face/qwen3.5-4b_31k/qwen3.5-4b_31k_step_34"
+)
+# ==============================================================================
+
+# 多 checkpoint 输入，按以下优先级解析：
+#   1. MODEL_PATH_LIST（换行分隔）及可选 MODEL_NAME_LIST；
+#   2. 兼容旧用法的单个 MODEL_PATH / MODEL_NAME；
+#   3. 上面的 MODEL_CHECKPOINTS 字符串数组（推荐日常使用）。
+# 所有路径都必须是 vLLM 可直接加载的、已合并 Hugging Face checkpoint；
+# 未合并的 VERL/FSDP shard 不能直接用于该推理入口。
+export MODEL_INPUT_VALIDATE_ONLY=${MODEL_INPUT_VALIDATE_ONLY:-0}
+
+declare -a INPUT_MODEL_PATHS=()
+declare -a INPUT_MODEL_NAMES=()
+
+if [ -n "${MODEL_PATH_LIST:-}" ]; then
+    while IFS= read -r model_path; do
+        model_path=${model_path%$'\r'}
+        if [ -n "${model_path}" ]; then
+            INPUT_MODEL_PATHS+=("${model_path}")
+        fi
+    done <<< "${MODEL_PATH_LIST}"
+    if [ -n "${MODEL_NAME_LIST:-}" ]; then
+        while IFS= read -r model_name; do
+            model_name=${model_name%$'\r'}
+            if [ -n "${model_name}" ]; then
+                INPUT_MODEL_NAMES+=("${model_name}")
+            fi
+        done <<< "${MODEL_NAME_LIST}"
+    fi
+elif [ -n "${MODEL_PATH:-}" ]; then
+    INPUT_MODEL_PATHS+=("${MODEL_PATH}")
+    if [ -n "${MODEL_NAME:-}" ]; then
+        INPUT_MODEL_NAMES+=("${MODEL_NAME}")
+    fi
+elif [ "${#MODEL_CHECKPOINTS[@]}" -gt 0 ]; then
+    for model_spec in "${MODEL_CHECKPOINTS[@]}"; do
+        if [[ "${model_spec}" != *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected \"model_name|/absolute/checkpoint/path\": ${model_spec}" >&2
+            exit 2
+        fi
+        model_name=${model_spec%%|*}
+        model_path=${model_spec#*|}
+        if [ -z "${model_name}" ] || [ -z "${model_path}" ] || [[ "${model_path}" == *"|"* ]]; then
+            echo "ERROR: invalid MODEL_CHECKPOINTS item; expected exactly one | delimiter: ${model_spec}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAMES+=("${model_name}")
+        INPUT_MODEL_PATHS+=("${model_path}")
+    done
+fi
+
+if [ "${#INPUT_MODEL_PATHS[@]}" -eq 0 ]; then
+    echo "ERROR: no model checkpoints configured." >&2
+    echo "Edit MODEL_CHECKPOINTS at the top of this script, set MODEL_PATH_LIST, or set legacy MODEL_PATH." >&2
+    exit 2
+fi
+if [ "${#INPUT_MODEL_NAMES[@]}" -ne 0 ] && [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+    echo "ERROR: model name count ${#INPUT_MODEL_NAMES[@]} does not match path count ${#INPUT_MODEL_PATHS[@]}." >&2
+    exit 2
+fi
+
+declare -A INPUT_MODEL_NAME_SEEN=()
+for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+    model_path=${INPUT_MODEL_PATHS[model_index]}
+    if [[ "${model_path}" != /* ]]; then
+        echo "ERROR: model checkpoint path must be absolute: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -d "${model_path}" ]; then
+        echo "ERROR: model checkpoint directory not found: ${model_path}" >&2
+        exit 2
+    fi
+    if [ ! -f "${model_path}/config.json" ]; then
+        echo "ERROR: merged Hugging Face config.json not found: ${model_path}/config.json" >&2
+        exit 2
+    fi
+    if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+        model_name=${INPUT_MODEL_NAMES[model_index]}
+        if [[ ! "${model_name}" =~ ^[a-zA-Z0-9._-]+$ ]]; then
+            echo "ERROR: model name may only contain letters, digits, dot, underscore and hyphen: ${model_name}" >&2
+            exit 2
+        fi
+        if [ -n "${INPUT_MODEL_NAME_SEEN[${model_name}]:-}" ]; then
+            echo "ERROR: duplicate configured model name: ${model_name}" >&2
+            exit 2
+        fi
+        INPUT_MODEL_NAME_SEEN[${model_name}]=1
+    else
+        model_name='<auto>'
+    fi
+    echo "MODEL_INPUT[$model_index] name=${model_name} path=${model_path}"
+done
+
+printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_PATHS[@]}"
+export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED}
+if [ "${#INPUT_MODEL_NAMES[@]}" -gt 0 ]; then
+    printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${INPUT_MODEL_NAMES[@]}"
+    export MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}
+else
+    unset MODEL_NAME_LIST
+fi
+unset MODEL_PATH MODEL_NAME
+
+echo "MODEL_INPUT_COUNT=${#INPUT_MODEL_PATHS[@]}"
+if [ "${MODEL_INPUT_VALIDATE_ONLY}" = "1" ]; then
+    echo "MODEL_INPUT_VALIDATE_ONLY=1: model configuration is valid; inference not started."
+    exit 0
+fi
+
+# 必须把训练时修改过的整份 VERL v12 代码同步到此目录；不能只安装上游 VERL。
+export WORK_DIR=${WORK_DIR:-${BUNDLE_ROOT}/verl}
+
+# 已离线筛选并适配好的训练兼容数据：base 100 + hard 100。这里直接读取
+# manifest-backed data_* bundle，不在推理节点重新扫描或适配完整 1000 题数据。
+# 部署到共享存储后，可通过 BASE_TASKS 覆盖为共享目录中的副本。
+export BASE_TASKS=${BASE_TASKS:-${BUNDLE_ROOT}/data/ClawBenchPro_base100_hard100_quality}
+
+# v14/0710/inference.sh 只有在该变量非空时才会启动全量 ClawBenchPro 适配器。
+# 专用入口固定使用上面的精选子集，避免意外退回 991/1000 题路径。
+export CLAWBENCHPRO_ROOT=
+export CLAWBENCHPRO_ADAPTED_ROOT=
+
+# 输出严格为 OUTPUT_ROOT/<model_name>/step_1/<task_id>_sample_<n>/。
+export OUTPUT_ROOT=${OUTPUT_ROOT:-/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/output/qwen35_4b}
+export OVERWRITE_OUTPUT=${OVERWRITE_OUTPUT:-True}
+export CONTINUE_ON_MODEL_ERROR=${CONTINUE_ON_MODEL_ERROR:-1}
+export MODEL_SWITCH_COOLDOWN=${MODEL_SWITCH_COOLDOWN:-20}
+export MODEL_RESOURCE_RELEASE_TIMEOUT=${MODEL_RESOURCE_RELEASE_TIMEOUT:-600}
+
+# 默认单机 8 NPU、TP=4，即 2 个 vLLM rollout replica。多机时所有节点提交
+# 同一脚本，ModelArts 通过 VC_TASK_INDEX 区分 rank。
+export INFER_NNODES=${INFER_NNODES:-1}
+export NPUS_PER_NODE=${NPUS_PER_NODE:-8}
+export INFER_TP=${INFER_TP:-4}
+
+# 评测默认每题 1 条轨迹；如需复现训练时的 GRPO 采样数量可设为 8。
+export N_RESP_PER_PROMPT=${N_RESP_PER_PROMPT:-1}
+export PROMPT_BATCH_SIZE=${PROMPT_BATCH_SIZE:-16}
+export AGENT_NUM_WORKERS=${AGENT_NUM_WORKERS:-32}
+export CALCULATE_LOG_PROBS=${CALCULATE_LOG_PROBS:-False}
+
+# 与 v14/0708_new/half_turn.sh 的 actor rollout 完全对齐。
+export MAX_TURNS=${MAX_TURNS:-35}
+export MAX_PROMPT_LENGTH=${MAX_PROMPT_LENGTH:-8192}
+export MAX_RESPONSE_LENGTH=${MAX_RESPONSE_LENGTH:-22768}
+export MAX_ASSISTANT_RESPONSE_LENGTH=${MAX_ASSISTANT_RESPONSE_LENGTH:-16384}
+export MAX_TOOL_RESPONSE_LENGTH=${MAX_TOOL_RESPONSE_LENGTH:-8192}
+export ROLLOUT_MAX_NUM_BATCHED_TOKENS=${ROLLOUT_MAX_NUM_BATCHED_TOKENS:-16384}
+export ROLLOUT_GPU_MEMORY_UTILIZATION=${ROLLOUT_GPU_MEMORY_UTILIZATION:-0.70}
+export ROLLOUT_TEMPERATURE=${ROLLOUT_TEMPERATURE:-1.0}
+export ROLLOUT_TOP_P=${ROLLOUT_TOP_P:-0.95}
+export ROLLOUT_TOP_K=${ROLLOUT_TOP_K:-20}
+export ROLLOUT_MIN_P=${ROLLOUT_MIN_P:-0.0}
+export ROLLOUT_PRESENCE_PENALTY=${ROLLOUT_PRESENCE_PENALTY:-0.0}
+export ROLLOUT_FREQUENCY_PENALTY=${ROLLOUT_FREQUENCY_PENALTY:-0.0}
+export ROLLOUT_REPETITION_PENALTY=${ROLLOUT_REPETITION_PENALTY:-1.0}
+export ROLLOUT_FREE_CACHE_ENGINE=${ROLLOUT_FREE_CACHE_ENGINE:-True}
+export ROLLOUT_ENFORCE_EAGER=${ROLLOUT_ENFORCE_EAGER:-False}
+
+# 与训练相同：thinking actor、Qwen3-Coder parser、同一 tool YAML、受限 bash、
+# 不保存 workspace_before，并严格禁止输出路径静默追加 request-id 后缀。
+export TOOL_CONFIG_PATH=${TOOL_CONFIG_PATH:-recipe/nanoclaw/nanoclaw_tool_config.yaml}
+export NANOCLAW_MAX_STEPS=${NANOCLAW_MAX_STEPS:-}
+export NANOCLAW_CLEANUP_WORKSPACES=False
+export NANOCLAW_KEEP_FAILED_WORKSPACES=False
+export NANOCLAW_ENV_BUILDER_TIMEOUT=${NANOCLAW_ENV_BUILDER_TIMEOUT:-600}
+export NANOCLAW_ALLOW_BASH=True
+export NANOCLAW_STRICT_RESULT_DIR=True
+export NANOCLAW_SAVE_WORKSPACE_BEFORE=False
+
+# 默认安装与训练一致的 GCC/CANN/torch-npu/vLLM/Triton/VERL 依赖。
+export SETUP_ENVIRONMENT=${SETUP_ENVIRONMENT:-1}
+
+# 只重跑失败或残缺的 checkpoint。已经完整通过磁盘输出审计的模型会从传给
+# inference.sh 的列表中移除，因此即使 OVERWRITE_OUTPUT=True 也绝不会删除或
+# 重跑这些成功模型；覆盖只作用于筛选后仍然残缺的模型目录。
+export RESUME_SKIP_COMPLETED_MODELS=1
+if [ "${RESUME_SKIP_COMPLETED_MODELS}" = "1" ]; then
+    if [ "${#INPUT_MODEL_NAMES[@]}" -ne "${#INPUT_MODEL_PATHS[@]}" ]; then
+        echo "ERROR: RESUME_SKIP_COMPLETED_MODELS=1 requires an explicit name for every model." >&2
+        exit 2
+    fi
+
+    declare -a FILTERED_MODEL_PATHS=()
+    declare -a FILTERED_MODEL_NAMES=()
+    for model_index in "${!INPUT_MODEL_PATHS[@]}"; do
+        model_path=${INPUT_MODEL_PATHS[model_index]}
+        model_name=${INPUT_MODEL_NAMES[model_index]}
+        model_output_root="${OUTPUT_ROOT}/${model_name}"
+
+        if python3 - "${BASE_TASKS}/benchmark_manifest.json" "${model_output_root}" "${N_RESP_PER_PROMPT}" <<'PY'
+import json
+import sys
+from pathlib import Path
+
+manifest_path = Path(sys.argv[1])
+model_root = Path(sys.argv[2])
+rollout_n = int(sys.argv[3])
+
+def load_json(path: Path):
+    try:
+        value = json.loads(path.read_text(encoding="utf-8"))
+    except Exception:
+        return None
+    return value if isinstance(value, dict) else None
+
+try:
+    manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
+    task_ids = [str(item["task_id"]) for item in manifest["tasks"]]
+except Exception as exc:
+    print(f"[resume_audit_error] manifest={manifest_path} error={type(exc).__name__}: {exc}", file=sys.stderr)
+    raise SystemExit(2)
+
+issues = []
+step_dir = model_root / "step_1"
+for task_id in task_ids:
+    for sample_index in range(rollout_n):
+        result_dir = step_dir / f"{task_id}_sample_{sample_index}"
+        if not result_dir.is_dir():
+            issues.append(f"missing_result:{result_dir.name}")
+            continue
+        if not (result_dir / "workspace_after").is_dir():
+            issues.append(f"missing_workspace:{result_dir.name}")
+        metadata = load_json(result_dir / "nanoclaw_metadata.json")
+        if metadata is None:
+            issues.append(f"invalid_metadata:{result_dir.name}")
+        elif metadata.get("status") != "ready":
+            issues.append(f"metadata_status={metadata.get('status')}:{result_dir.name}")
+        if load_json(result_dir / "conversation_history.json") is None:
+            issues.append(f"invalid_conversation:{result_dir.name}")
+        if load_json(result_dir / "trajectory.json") is None:
+            issues.append(f"invalid_trajectory:{result_dir.name}")
+
+if issues:
+    preview = ", ".join(issues[:8])
+    print(
+        f"[resume_rerun_incomplete] model={model_root.name} issues={len(issues)} preview={preview}",
+        file=sys.stderr,
+    )
+    raise SystemExit(1)
+
+print(
+    f"[resume_skip_complete] model={model_root.name} tasks={len(task_ids)} rollout_n={rollout_n}",
+    file=sys.stderr,
+)
+PY
+        then
+            echo "SKIP_COMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        else
+            audit_rc=$?
+            if [ "${audit_rc}" -eq 2 ]; then
+                echo "ERROR: unable to audit existing model output: ${model_name}" >&2
+                exit 2
+            fi
+            FILTERED_MODEL_PATHS+=("${model_path}")
+            FILTERED_MODEL_NAMES+=("${model_name}")
+            echo "RERUN_INCOMPLETE_MODEL name=${model_name} output=${model_output_root}"
+        fi
+    done
+
+    if [ "${#FILTERED_MODEL_PATHS[@]}" -eq 0 ]; then
+        echo "ALL_CONFIGURED_MODELS_ALREADY_COMPLETE: no inference will be submitted."
+        exit 0
+    fi
+
+    printf -v MODEL_PATH_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_PATHS[@]}"
+    printf -v MODEL_NAME_LIST_NORMALIZED '%s\n' "${FILTERED_MODEL_NAMES[@]}"
+    export MODEL_PATH_LIST=${MODEL_PATH_LIST_NORMALIZED}
+    export MODEL_NAME_LIST=${MODEL_NAME_LIST_NORMALIZED}
+    export OVERWRITE_OUTPUT=True
+    echo "RESUME_RERUN_MODEL_COUNT=${#FILTERED_MODEL_PATHS[@]}"
+fi
+
+exec bash "/opt/huawei/dataset/zyr_yuyin/lyf/datasets/testClawBenchPro/upload_clawbenchpro_base100_hard100_npu/v14/0710/inference.sh"