src/submission/student_queue.py

import json
import os
from datetime import datetime, timezone
from typing import Optional

from huggingface_hub import CommitOperationAdd

from src.envs import API, QUEUE_REPO, PROJ_DIR, PROJ_ID


def _assert_file_exists(path: Optional[str], name: str, required: bool = True) -> Optional[str]:
    if not path:
        if required:
            raise FileNotFoundError(f"Missing required file: {name}")
        return None

    if not os.path.isfile(path):
        raise FileNotFoundError(
            f"Provided path for {name} is not a file on the local file system: {path}"
        )

    if os.path.getsize(path) == 0:
        raise ValueError(f"{name} is empty: {path}")

    return path


def queue_student_submission(
    group_id: str,
    alias: Optional[str],
    state_dict_file: Optional[str],
    model_py_file: str,
    preproc_py_file: str,
) -> tuple[str, str]:
    """
    Upload submitted files to the private queue dataset for offline evaluation.

    Layout in repo:
      {PROJ_DIR}/{group_id alias}/{timestamp}/model.py
      {PROJ_DIR}/{group_id alias}/{timestamp}/preprocess.py
      {PROJ_DIR}/{group_id alias}/{timestamp}/model.pt  (optional)
      {PROJ_DIR}/{group_id alias}/{timestamp}/request.json
    """
    if not group_id or not group_id.strip():
        raise ValueError("Group ID is required.")

    group_id = group_id.strip()
    alias = alias.strip() if isinstance(alias, str) else alias

    ts = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%SZ")
    alias_clean = alias.strip() if isinstance(alias, str) and alias.strip() else "NA"
    folder_name = f"{group_id} {alias_clean}"
    base_path = f"{PROJ_DIR}/{folder_name}/{ts}"

    model_py_file = _assert_file_exists(model_py_file, "model.py", required=True)
    preproc_py_file = _assert_file_exists(preproc_py_file, "preprocess.py", required=True)
    state_dict_file = _assert_file_exists(state_dict_file, "model.pt", required=False)

    request_json = {
        "proj": PROJ_ID,
        "group_id": group_id,
        "alias": alias,
        "timestamp": ts,
        "status": "PENDING",
        "datasets": ["url_val"],
        "has_weights": bool(state_dict_file),
    }

    operations = [
        CommitOperationAdd(
            path_in_repo=f"{base_path}/model.py",
            path_or_fileobj=model_py_file,
        ),
        CommitOperationAdd(
            path_in_repo=f"{base_path}/preprocess.py",
            path_or_fileobj=preproc_py_file,
        ),
    ]

    if state_dict_file:
        operations.append(
            CommitOperationAdd(
                path_in_repo=f"{base_path}/model.pt",
                path_or_fileobj=state_dict_file,
            )
        )

    # Add request.json last. This is the worker's "ready" signal.
    operations.append(
        CommitOperationAdd(
            path_in_repo=f"{base_path}/request.json",
            path_or_fileobj=json.dumps(request_json).encode("utf-8"),
        )
    )

    API.create_commit(
        repo_id=QUEUE_REPO,
        repo_type="dataset",
        operations=operations,
        commit_message=f"Queue submission {group_id}/{ts}",
    )

    return (
        f"Submission queued for Group '{group_id}' at {ts}. "
        "Your model will be evaluated shortly.",
        ts,
    )