"""Resolve a Kaggle-fitting HF base model (never GGUF)."""
from __future__ import annotations

import json
import os
from pathlib import Path
from typing import Any

DEFAULT_CANDIDATES = [
    "Qwen/Qwen2.5-Coder-1.5B-Instruct",
    "Qwen/Qwen2.5-Coder-0.5B-Instruct",
    "Qwen/Qwen2.5-1.5B-Instruct",
    "HuggingFaceTB/SmolLM2-1.7B-Instruct",
    "google/gemma-2-2b-it",
]


def _probe(repo_id: str, token: str | None) -> bool:
    from huggingface_hub import HfApi

    api = HfApi(token=token)
    try:
        info = api.model_info(repo_id)
    except Exception:
        return False
    siblings = [s.rfilename for s in (info.siblings or [])]
    # Reject pure-GGUF repos
    has_weight = any(
        f.endswith((".safetensors", ".bin", ".pt"))
        or f == "pytorch_model.bin.index.json"
        or f.endswith(".safetensors.index.json")
        for f in siblings
    )
    only_gguf = siblings and all(
        f.endswith(".gguf") or f.endswith(".md") or "license" in f.lower() or f.startswith(".")
        for f in siblings
    )
    if only_gguf and not has_weight:
        return False
    return has_weight or True  # allow repos that list via remote code


def resolve_train_model(
    preferred: list[str] | None = None,
    *,
    override: str | None = None,
    token: str | None = None,
    out_path: Path | None = None,
) -> dict[str, Any]:
    token = token or os.getenv("HF_TOKEN")
    override = override or os.getenv("TRAIN_MODEL_ID") or None
    candidates = []
    if override:
        candidates.append(override)
    candidates.extend(preferred or DEFAULT_CANDIDATES)

    seen: set[str] = set()
    errors: list[dict[str, str]] = []
    for repo_id in candidates:
        if repo_id in seen:
            continue
        seen.add(repo_id)
        try:
            if _probe(repo_id, token):
                result = {
                    "train_model_id": repo_id,
                    "candidates_tried": list(seen),
                    "errors": errors,
                }
                if out_path:
                    out_path.parent.mkdir(parents=True, exist_ok=True)
                    out_path.write_text(json.dumps(result, indent=2), encoding="utf-8")
                return result
        except Exception as e:  # noqa: BLE001
            errors.append({"repo_id": repo_id, "error": f"{type(e).__name__}: {e}"})
            continue
        errors.append({"repo_id": repo_id, "error": "probe_failed"})

    raise RuntimeError(
        "No trainable HF model resolved. Tried: "
        + ", ".join(seen)
        + f" errors={errors}"
    )