File size: 7,804 Bytes
c793be8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
from __future__ import annotations

import ast
import json
from collections import Counter
from pathlib import Path


ROOT = Path(__file__).resolve().parents[1]
DATA = ROOT / "data" / "ClawBenchPro_base100_hard100_quality"
VERL = ROOT / "verl"


def require_file(path: Path) -> None:
    if not path.is_file():
        raise FileNotFoundError(path)


def main() -> None:
    required_bundle_files = [
        ROOT / "v14" / "0708_new" / "inference_clawbenchpro.sh",
        ROOT / "v14" / "0708_new" / "score_clawbenchpro.sh",
        ROOT / "v14" / "0710" / "inference.sh",
        ROOT / "scripts" / "setup_full_npu_environment.sh",
        VERL / "recipe" / "nanoclaw" / "inference.py",
        VERL / "recipe" / "nanoclaw" / "nanoclaw.py",
        VERL / "recipe" / "nanoclaw" / "nanoclaw_tool_config.yaml",
        VERL / "recipe" / "nanoclaw" / "score_clawbenchpro.py",
        VERL / "verl" / "experimental" / "agent_loop" / "agent_loop.py",
        VERL / "verl" / "experimental" / "agent_loop" / "tool_agent_loop.py",
        DATA / "benchmark_manifest.json",
        DATA / "quality_selection_report.json",
        DATA / "quality_selection_manifest.jsonl",
        DATA / "_SUCCESS",
        ROOT / "BUNDLE_VALIDATION.json",
    ]
    for path in required_bundle_files:
        require_file(path)

    inference_shell = (ROOT / "v14" / "0708_new" / "inference_clawbenchpro.sh").read_text(encoding="utf-8")
    inference_core_shell = (ROOT / "v14" / "0710" / "inference.sh").read_text(encoding="utf-8")
    scoring_shell = (ROOT / "v14" / "0708_new" / "score_clawbenchpro.sh").read_text(encoding="utf-8")
    scoring_python = (VERL / "recipe" / "nanoclaw" / "score_clawbenchpro.py").read_text(encoding="utf-8")
    setup_shell = (ROOT / "scripts" / "setup_full_npu_environment.sh").read_text(encoding="utf-8")
    if "SETUP_ENVIRONMENT=${SETUP_ENVIRONMENT:-1}" not in inference_shell:
        raise ValueError("inference entrypoint must default to full environment installation")
    required_multi_model_markers = (
        "MODEL_CHECKPOINTS",
        "MODEL_PATH_LIST",
        "MODEL_NAME_LIST",
        "MODEL_INPUT_VALIDATE_ONLY",
        "INPUT_MODEL_PATHS",
        "CONTINUE_ON_MODEL_ERROR",
        "MODEL_SWITCH_COOLDOWN",
        "MODEL_RESOURCE_RELEASE_TIMEOUT",
        "model checkpoint path must be absolute",
    )
    missing_multi_model_markers = [marker for marker in required_multi_model_markers if marker not in inference_shell]
    if missing_multi_model_markers:
        raise ValueError(f"multi-checkpoint inference entrypoint is incomplete: {missing_multi_model_markers}")
    required_core_multi_model_markers = (
        'for model_index in "${!MODEL_PATHS[@]}"',
        'model_output_root=${OUTPUT_ROOT}/${model_name}',
        "wait_for_ray_available_npu_resources",
        "MODEL_RESOLVED_NAMES",
        "duplicate model output name",
    )
    missing_core_markers = [marker for marker in required_core_multi_model_markers if marker not in inference_core_shell]
    if missing_core_markers:
        raise ValueError(f"core multi-checkpoint loop is incomplete: {missing_core_markers}")
    required_scoring_multi_model_markers = (
        "discover_models",
        "for model_dir in models",
        'output_root / model_dir.name',
        'output_root / "leaderboard.csv"',
    )
    missing_scoring_markers = [marker for marker in required_scoring_multi_model_markers if marker not in scoring_python]
    if missing_scoring_markers:
        raise ValueError(f"multi-model scoring is incomplete: {missing_scoring_markers}")
    if "SETUP_ENVIRONMENT=${SETUP_ENVIRONMENT:-1}" not in scoring_shell:
        raise ValueError("scoring entrypoint must default to full environment installation")
    if 'source "${BUNDLE_ROOT}/scripts/setup_full_npu_environment.sh"' not in scoring_shell:
        raise ValueError("scoring entrypoint does not source the full NPU environment installer")
    required_install_markers = (
        "torch==2.9.0",
        "torch-npu==2.9.0",
        "VLLM_TARGET_DEVICE=empty pip install .",
        "triton==3.5.0",
        "pip install -r requirements-npu.txt",
        "transformers==5.3.0",
        "openai httpx",
    )
    missing_install_markers = [marker for marker in required_install_markers if marker not in setup_shell]
    if missing_install_markers:
        raise ValueError(f"full NPU environment installer is incomplete: {missing_install_markers}")

    manifest = json.loads((DATA / "benchmark_manifest.json").read_text(encoding="utf-8"))
    tasks = manifest.get("tasks")
    if not isinstance(tasks, list) or manifest.get("task_count") != 200 or len(tasks) != 200:
        raise ValueError("benchmark_manifest.json must declare exactly 200 tasks")
    categories = Counter(str(item.get("quality_category")) for item in tasks)
    if categories != {"base": 100, "hard": 100}:
        raise ValueError(f"unexpected category counts: {dict(categories)}")

    task_dirs = sorted(path for path in DATA.glob("data_*") if path.is_dir())
    if len(task_dirs) != 200:
        raise ValueError(f"expected 200 task directories, found {len(task_dirs)}")
    if {path.name for path in task_dirs} != {str(item.get("task_id")) for item in tasks}:
        raise ValueError("task directories do not match benchmark manifest")

    llm_judge = 0
    required_task_keys = ("env_builder", "prompt", "verifier", "yaml")
    for task_dir in task_dirs:
        task_manifest = json.loads((task_dir / "manifest.json").read_text(encoding="utf-8"))
        files = task_manifest.get("files")
        if not isinstance(files, dict):
            raise ValueError(f"invalid task manifest: {task_dir}")
        for key in required_task_keys:
            require_file(task_dir / str(files.get(key) or "<missing>"))
        require_file(task_dir / "_env_builder_impl.py")
        verifier = task_dir / str(files["verifier"])
        verifier_text = verifier.read_text(encoding="utf-8", errors="replace")
        if any(marker in verifier_text for marker in ("OpenAI(", "client.chat.completions", "httpx.Client")):
            llm_judge += 1
        for python_file in (task_dir / str(files["env_builder"]), task_dir / "_env_builder_impl.py", verifier):
            ast.parse(python_file.read_text(encoding="utf-8"), filename=str(python_file))

    if llm_judge != 134:
        raise ValueError(f"expected 134 LLM Judge verifiers, found {llm_judge}")
    symlinks = [path for path in ROOT.rglob("*") if path.is_symlink()]
    if symlinks:
        raise ValueError(f"bundle contains symlinks and is not self-contained: {symlinks[:10]}")

    for python_file in (
        VERL / "recipe" / "nanoclaw" / "inference.py",
        VERL / "recipe" / "nanoclaw" / "score_clawbenchpro.py",
    ):
        ast.parse(python_file.read_text(encoding="utf-8"), filename=str(python_file))

    print(
        json.dumps(
            {
                "bundle_root": str(ROOT),
                "tasks": 200,
                "categories": dict(categories),
                "deterministic_verifiers": 66,
                "llm_judge_verifiers": llm_judge,
                "inference_full_environment_install_default": True,
                "inference_multi_checkpoint_input": True,
                "script_top_model_array_smoke_count": 3,
                "absolute_checkpoint_paths_required": True,
                "script_top_model_strings": True,
                "tabular_model_config_removed": True,
                "per_model_output_directories": True,
                "scoring_full_environment_install_default": True,
                "scoring_multi_model_discovery": True,
                "symlinks": 0,
                "status": "ok",
            },
            ensure_ascii=False,
            indent=2,
        )
    )


if __name__ == "__main__":
    main()