| """Execute the notebook in a fresh CPU kernel, with an 8 GiB address-space cap.""" |
| import argparse |
| import json |
| import os |
| from pathlib import Path |
| import resource |
| import sys |
| import time |
|
|
| import nbformat |
| from nbclient import NotebookClient |
|
|
|
|
| def main(): |
| parser = argparse.ArgumentParser() |
| parser.add_argument("--remote", action="store_true") |
| parser.add_argument("--revision") |
| parser.add_argument("--output", type=Path, default=Path("/tmp/trajectory_eda_executed.ipynb")) |
| args = parser.parse_args() |
| root = Path(__file__).resolve().parents[1] |
| os.environ.update(MPLCONFIGDIR="/tmp/trajectory_mpl", OPENBLAS_NUM_THREADS="1", OMP_NUM_THREADS="1", |
| CUDA_VISIBLE_DEVICES="", HF_HUB_DISABLE_IMPLICIT_TOKEN="1", HF_HOME="/tmp/trajectory_anonymous_hf") |
| os.environ.pop("HF_TOKEN", None) |
| os.environ["CORPUS_EXPORT_DIR"] = "/tmp/trajectory_eda_exports" |
| if args.remote: |
| os.environ.pop("CORPUS_LOCAL_ROOT", None) |
| else: |
| os.environ["CORPUS_LOCAL_ROOT"] = str(root) |
| if args.revision: |
| os.environ["CORPUS_REVISION"] = args.revision |
| resource.setrlimit(resource.RLIMIT_AS, (8 * 1024**3, 8 * 1024**3)) |
| kernel_dir = Path("/tmp/trajectory_jupyter/kernels/trajectory-cpu") |
| kernel_dir.mkdir(parents=True, exist_ok=True) |
| (kernel_dir / "kernel.json").write_text(json.dumps({ |
| "argv":[sys.executable,"-m","ipykernel_launcher","-f","{connection_file}"], |
| "display_name":"Trajectory CPU validation", "language":"python"})) |
| os.environ["JUPYTER_PATH"] = "/tmp/trajectory_jupyter" |
| nb = nbformat.read(root / "trajectory_eda_colab.ipynb", as_version=4) |
| nb.cells.append(nbformat.v4.new_code_cell(""" |
| # Validation-only checks for lazy examples and the interactive callback. |
| process = example_summary.loc[(example_summary.example_kind == 'process') & (example_summary.truncated == True)].iloc[0] |
| assert inspect_example(process.example_id)['metadata']['truncated'] is True |
| packet = example_summary.loc[example_summary.example_kind == 'experiment1_judge_packet'].iloc[0] |
| assert inspect_example(packet.example_id)['packet_id'] == packet.example_id |
| controls['collection'].value = 'olmo' |
| rejected = episodes.loc[(episodes.collection == 'olmo') & (episodes.rejected_responses > 0)].sort_values('events').iloc[0] |
| picker.value = rejected.episode_id |
| canonical_only.value = False |
| only_rejected.value = True |
| show_timeline() |
| assert selected_bundle['episode']['episode_id'] == rejected.episode_id |
| assert any(e['rejected_response'] for e in selected_bundle['events']) |
| controls['collection'].value = 'qwen' |
| assert all(eid in set(episodes.loc[episodes.collection == 'qwen','episode_id']) for label,eid in picker.options) |
| print('Inspector filters, rejected-response callback, truncated example, and packet loads passed.') |
| """)) |
| start = time.monotonic() |
| client = NotebookClient(nb, timeout=240, kernel_name="trajectory-cpu", resources={"metadata":{"path":"/tmp"}}) |
| client.execute() |
| nbformat.write(nb, args.output) |
| report = {"status":"passed", "mode":"anonymous_remote" if args.remote else "local_publication_copy", |
| "code_cells_executed":sum(c.cell_type == "code" for c in nb.cells), |
| "elapsed_seconds":round(time.monotonic()-start,2), "address_space_cap_gib":8, |
| "peak_child_rss_mib":round(resource.getrusage(resource.RUSAGE_CHILDREN).ru_maxrss/1024,2), |
| "gpu_visible":False, "model_calls":0, "login_required":False, |
| "dataset_revision":args.revision or "resolved_inside_notebook", "executed_notebook":str(args.output)} |
| target = root / "provenance" / ("notebook_anonymous_execution.json" if args.remote else "notebook_cpu_execution.json") |
| target.write_text(json.dumps(report,indent=2)+"\n") |
| print(json.dumps(report,indent=2)) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
