decoding

游雁

2024-06-12 be26169447b2e5f8f38c97af8f5f6a201bc6ce40

 funasr/train_utils/load_pretrained_model.py

@@ -7,120 +7,91 @@
import torch
import torch.nn
import torch.optim


def filter_state_dict(
    dst_state: Dict[str, Union[float, torch.Tensor]],
    src_state: Dict[str, Union[float, torch.Tensor]],
):
    """Filter name, size mismatch instances between dicts.

    Args:
        dst_state: reference state dict for filtering
        src_state: target state dict for filtering

    """
    match_state = {}
    for key, value in src_state.items():
        if key in dst_state and (dst_state[key].size() == src_state[key].size()):
            match_state[key] = value
        else:
            if key not in dst_state:
                logging.warning(
                    f"Filter out {key} from pretrained dict"
                    + " because of name not found in target dict"
                )
            else:
                logging.warning(
                    f"Filter out {key} from pretrained dict"
                    + " because of size mismatch"
                    + f"({dst_state[key].size()}-{src_state[key].size()})"
                )
    return match_state
import pdb


def load_pretrained_model(
    init_param: str,
    path: str,
    model: torch.nn.Module,
    ignore_init_mismatch: bool,
    ignore_init_mismatch: bool = True,
    map_location: str = "cpu",
    oss_bucket=None,
    scope_map=[],
    excludes=None,
    **kwargs,
):
    """Load a model state and set it to the model.

    Args:
        init_param: <file_path>:<src_key>:<dst_key>:<exclude_Keys>
            init_param: <file_path>:<src_key>:<dst_key>:<exclude_Keys>

    Examples:
        >>> load_pretrained_model("somewhere/model.pb", model)
        >>> load_pretrained_model("somewhere/model.pb:decoder:decoder", model)
        >>> load_pretrained_model("somewhere/model.pb:decoder:decoder:", model)
        >>> load_pretrained_model(
        ...     "somewhere/model.pb:decoder:decoder:decoder.embed", model
        ... )
        >>> load_pretrained_model("somewhere/decoder.pb::decoder", model)

    """
    sps = init_param.split(":", 4)
    if len(sps) == 4:
        path, src_key, dst_key, excludes = sps
    elif len(sps) == 3:
        path, src_key, dst_key = sps
        excludes = None
    elif len(sps) == 2:
        path, src_key = sps
        dst_key, excludes = None, None
    else:
        (path,) = sps
        src_key, dst_key, excludes = None, None, None
    if src_key == "":
        src_key = None
    if dst_key == "":
        dst_key = None

    if dst_key is None:
        obj = model
    else:
    obj = model
    dst_state = obj.state_dict()

        def get_attr(obj: Any, key: str):
            """Get an nested attribute.

            >>> class A(torch.nn.Module):
            ...     def __init__(self):
            ...         super().__init__()
            ...         self.linear = torch.nn.Linear(10, 10)
            >>> a = A()
            >>> assert A.linear.weight is get_attr(A, 'linear.weight')

            """
            if key.strip() == "":
                return obj
            for k in key.split("."):
                obj = getattr(obj, k)
            return obj

        obj = get_attr(model, dst_key)
    logging.info(f"ckpt: {path}")

    if oss_bucket is None:
        src_state = torch.load(path, map_location=map_location)
    else:
        buffer = BytesIO(oss_bucket.get_object(path).read())
        src_state = torch.load(buffer, map_location=map_location)

    src_state = src_state["state_dict"] if "state_dict" in src_state else src_state
    src_state = src_state["model_state_dict"] if "model_state_dict" in src_state else src_state
    src_state = src_state["model"] if "model" in src_state else src_state

    if isinstance(scope_map, str):
        scope_map = scope_map.split(",")
    scope_map += ["module.", "None"]
    logging.info(f"scope_map: {scope_map}")

    if excludes is not None:
        for e in excludes.split(","):
            src_state = {k: v for k, v in src_state.items() if not k.startswith(e)}
        if isinstance(excludes, str):
            excludes = excludes.split(",")
    logging.info(f"excludes: {excludes}")

    if src_key is not None:
        src_state = {
            k[len(src_key) + 1 :]: v
            for k, v in src_state.items()
            if k.startswith(src_key)
        }
    for k in dst_state.keys():

    dst_state = obj.state_dict()
    if ignore_init_mismatch:
        src_state = filter_state_dict(dst_state, src_state)
        for k_ex in excludes:
            if k.startswith(k_ex):
                logging.info(f"key: {{k}} matching: {k_ex}, excluded")
                continue

    logging.debug("Loaded src_state keys: {}".format(src_state.keys()))
    logging.debug("Loaded dst_state keys: {}".format(dst_state.keys()))
    dst_state.update(src_state)
    obj.load_state_dict(dst_state)
        k_src = k

        if scope_map is not None:
            src_prefix = ""
            dst_prefix = ""
            for i in range(0, len(scope_map), 2):
                src_prefix = scope_map[i] if scope_map[i].lower() != "none" else ""
                dst_prefix = scope_map[i + 1] if scope_map[i + 1].lower() != "none" else ""

                if dst_prefix == "" and (src_prefix + k) in src_state.keys():
                    k_src = src_prefix + k
                    if not k_src.startswith("module."):
                        logging.info(f"init param, map: {k} from {k_src} in ckpt")
                elif (
                    k.startswith(dst_prefix)
                    and k.replace(dst_prefix, src_prefix, 1) in src_state.keys()
                ):
                    k_src = k.replace(dst_prefix, src_prefix, 1)
                    if not k_src.startswith("module."):
                        logging.info(f"init param, map: {k} from {k_src} in ckpt")

        if k_src in src_state.keys():
            if ignore_init_mismatch and dst_state[k].shape != src_state[k_src].shape:
                logging.info(
                    f"ignore_init_mismatch:{ignore_init_mismatch}, dst: {k, dst_state[k].shape}, src: {k_src, src_state[k_src].shape}"
                )
            else:
                dst_state[k] = src_state[k_src]

        else:
            logging.info(f"Warning, miss key in ckpt: {k}, mapped: {k_src}")

    flag = obj.load_state_dict(dst_state, strict=True)
    logging.info(f"Loading ckpt: {path}, status: {flag}")