from __future__ import annotations
import logging
import torch
from .configuration_riemann import RiemannConfig
from .modeling_riemann import RiemannForCausalLM

logger = logging.getLogger("riemann.openpeer_weights")
OPENPEER_REPO = "OpenPeerAI/OpenPeerLLM"
OPENPEER_WEIGHTS_FILE = "model.safetensors"


def load_openpeer_state_dict() -> dict:
    from huggingface_hub import hf_hub_download
    from safetensors.torch import load_file

    path = hf_hub_download(OPENPEER_REPO, OPENPEER_WEIGHTS_FILE)
    return load_file(path)


def build_stage0_from_openpeer_pretrained(config: RiemannConfig) -> RiemannForCausalLM:
    expected = dict(
        vocab_size=50257,
        hidden_size=768,
        num_hidden_layers=12,
        num_attention_heads=12,
        intermediate_size=3072,
        max_position_embeddings=1024,
    )
    for key, value in expected.items():
        if getattr(config, key) != value:
            raise ValueError(
                f"build_stage0_from_openpeer_pretrained needs {key}={value} to match the published OpenPeerLLM checkpoint shape, got {getattr(config, key)}"
            )
    model = RiemannForCausalLM(config)
    state_dict = load_openpeer_state_dict()
    missing, unexpected = model.load_state_dict(state_dict, strict=False)
    if missing or unexpected:
        logger.warning(
            "loading OpenPeerLLM weights: missing=%s unexpected=%s", missing, unexpected
        )
    else:
        logger.info(
            "loaded all %d tensors from %s straight from the Hub, no conversion needed",
            len(state_dict),
            OPENPEER_REPO,
        )
    return model


if __name__ == "__main__":
    logging.basicConfig(
        level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s"
    )
    cfg = RiemannConfig()
    m = build_stage0_from_openpeer_pretrained(cfg)
    n_params = sum((p.numel() for p in m.parameters()))
    print(f"built stage0 from real OpenPeerLLM weights, {n_params:,} params")
    ids = torch.randint(0, cfg.vocab_size, (1, 16))
    out = m(input_ids=ids, labels=ids)
    print("sanity forward pass loss:", out.loss.item())