155 lines
6.4 KiB
Python
155 lines
6.4 KiB
Python
"""根目录、config.yaml、.env.local。所有路径解析只在这里做。"""
|
||
from __future__ import annotations
|
||
|
||
import os
|
||
import hashlib
|
||
import json
|
||
from functools import lru_cache
|
||
from pathlib import Path
|
||
|
||
import yaml
|
||
|
||
def resolve_installation(root=None, config_path=None, environ=None) -> tuple[Path, Path, Path]:
|
||
"""Pure path resolution shared by runtime and offline inventory.
|
||
|
||
Explicit root/config wins over environment. Relative override paths follow
|
||
the caller's working directory, as the existing runtime config path does.
|
||
No config/secret contents, database or runtime directories are touched.
|
||
"""
|
||
env = os.environ if environ is None else environ
|
||
source = Path(root or env.get("BENCH_ROOT") or Path(__file__).resolve().parent.parent).expanduser().resolve()
|
||
config_file = Path(config_path or env.get("BENCH_CONFIG") or source / "config.yaml").expanduser().resolve()
|
||
state = Path(env.get("BENCH_STATE_DIR") or source / "state").expanduser().resolve()
|
||
return source, config_file, state
|
||
|
||
|
||
BENCH_ROOT, CONFIG_PATH, STATE = resolve_installation()
|
||
PROJECT_ROOT = BENCH_ROOT.parent
|
||
REGISTRY = BENCH_ROOT / "registry"
|
||
DATA = BENCH_ROOT / "data"
|
||
CACHE = BENCH_ROOT / "cache"
|
||
RUNS = BENCH_ROOT / "runs"
|
||
REPORTS = BENCH_ROOT / "reports"
|
||
DATASETS = BENCH_ROOT / "datasets"
|
||
AGENT = BENCH_ROOT / "agent"
|
||
DB_PATH = STATE / "jobs.sqlite"
|
||
|
||
STAGE_NAMES = [
|
||
"01-fetch", "02-baseline", "03-export", # model-level → cache/
|
||
"04-calibrate", "05-compile", "06-hbm", "07-board", "08-eval", # target-level → job dir
|
||
]
|
||
MODEL_LEVEL = STAGE_NAMES[:3]
|
||
TARGET_LEVEL = STAGE_NAMES[3:]
|
||
TASKS = ["cls", "det", "seg", "iseg", "pose", "mde", "stereo", "flow", "feat", "occ"]
|
||
MARCHES = ["nash-e", "nash-m", "nash-p"]
|
||
|
||
|
||
def expand(p: str | os.PathLike | None, base: Path = BENCH_ROOT) -> Path | None:
|
||
if p is None:
|
||
return None
|
||
p = os.path.expandvars(os.path.expanduser(str(p)))
|
||
q = Path(p)
|
||
return q if q.is_absolute() else (base / q).resolve()
|
||
|
||
|
||
@lru_cache(maxsize=1)
|
||
def config() -> dict:
|
||
with open(CONFIG_PATH, encoding="utf-8") as f:
|
||
cfg = yaml.safe_load(f) or {}
|
||
cfg.setdefault("paths", {})
|
||
if not cfg["paths"].get("project_root"):
|
||
cfg["paths"]["project_root"] = str(PROJECT_ROOT)
|
||
return cfg
|
||
|
||
|
||
def venv_python(backend: str, job: dict | None = None) -> Path:
|
||
"""按 backend 选解释器。这是 harness 的职责,agent 无权改。"""
|
||
if job and job.get("toolchain") and backend == job.get("backend"):
|
||
from .toolchains import validate
|
||
return Path(validate(job)["venv"]) / "bin" / "python"
|
||
v = config()["paths"]["venv"].get(backend)
|
||
if not v:
|
||
raise KeyError(f"config.yaml paths.venv 没有 backend={backend}")
|
||
return expand(v) / "bin" / "python"
|
||
|
||
|
||
def venv_bin(backend: str, tool: str, job: dict | None = None) -> Path:
|
||
return venv_python(backend, job).parent / tool
|
||
|
||
|
||
def model_runtime_path(value: str | os.PathLike) -> Path:
|
||
"""Keep the venv launch path: different venvs often symlink the same Python binary."""
|
||
path = Path(os.path.expandvars(os.path.expanduser(str(value))))
|
||
return Path(os.path.abspath(path if path.is_absolute() else BENCH_ROOT / path))
|
||
|
||
|
||
def model_python(job: dict) -> Path:
|
||
"""New model recipes may require an isolated upstream-compatible runtime."""
|
||
runtime = job.get("model_spec", {}).get("runtime_python")
|
||
frozen = job.get("model_runtime")
|
||
if frozen:
|
||
current = (config().get("model_runtimes") or {}).get(job.get("model"), frozen["configured"])
|
||
matches = str(model_runtime_path(current)) == frozen["python"]
|
||
if not matches and "cache_contract" not in job:
|
||
matches = str(expand(current)) == frozen["python"] # Historical snapshots may have followed symlinks.
|
||
if not matches:
|
||
raise ValueError("frozen model runtime changed; create a new run")
|
||
return Path(frozen["python"])
|
||
if runtime:
|
||
return model_runtime_path(runtime) if "cache_contract" in job else expand(runtime)
|
||
return venv_python("bench")
|
||
|
||
|
||
def freeze_cache_contract(job: dict) -> dict:
|
||
"""New jobs explicitly opt into versioned caches, including the default Python."""
|
||
from .tools.export_onnx import exporter_version
|
||
runtime = job.get("model_spec", {}).get("runtime_python")
|
||
python = model_runtime_path(runtime) if runtime and not job.get("model_runtime") else model_python(job)
|
||
return {"version": 1, "model_python": str(model_runtime_path(python)),
|
||
"exporter_version": "task-recipe-v1" if job.get("model_recipe") else exporter_version(job["model"])}
|
||
|
||
|
||
def cache_contract(job: dict, *, validate: bool = False) -> dict | None:
|
||
"""Missing means legacy, never an invitation to relabel an existing cache."""
|
||
if "cache_contract" not in job:
|
||
if validate and job.get("model_runtime"):
|
||
model_python(job)
|
||
return None
|
||
contract = job["cache_contract"]
|
||
if (not isinstance(contract, dict) or type(contract.get("version")) is not int or contract["version"] != 1
|
||
or not isinstance(contract.get("model_python"), str) or not Path(contract["model_python"]).is_absolute()
|
||
or not isinstance(contract.get("exporter_version"), str) or not contract["exporter_version"]):
|
||
raise ValueError("invalid or unsupported cache contract")
|
||
if validate and str(model_runtime_path(model_python(job))) != contract["model_python"]:
|
||
raise ValueError("frozen cache runtime changed; create a new run")
|
||
return contract
|
||
|
||
|
||
def cache_key(identity: dict) -> str:
|
||
return hashlib.sha256(json.dumps(identity, sort_keys=True, separators=(",", ":")).encode()).hexdigest()
|
||
|
||
|
||
@lru_cache(maxsize=1)
|
||
def load_env_local() -> dict:
|
||
"""读取 bench/.env.local。只给 harness 用;rollout 的 agent 子进程环境由 stagerun/rollout 单独构造,永远不含这些。"""
|
||
env = {}
|
||
p = BENCH_ROOT / ".env.local"
|
||
if p.exists():
|
||
for line in p.read_text(encoding="utf-8").splitlines():
|
||
line = line.strip()
|
||
if not line or line.startswith("#") or "=" not in line:
|
||
continue
|
||
k, v = line.split("=", 1)
|
||
env[k.strip()] = v.strip().strip('"').strip("'")
|
||
return env
|
||
|
||
|
||
def secret(name: str) -> str | None:
|
||
return os.environ.get(name) or load_env_local().get(name)
|
||
|
||
|
||
def ensure_dirs() -> None:
|
||
for d in (STATE, RUNS, REPORTS, DATASETS, DATA / "pool", DATA / "t1", DATA / "calib",
|
||
CACHE / "weights", CACHE / "onnx", CACHE / "baseline"):
|
||
d.mkdir(parents=True, exist_ok=True)
|