Function load_local_files

bertopic/_save_utils.py:174–221

Load local BERTopic files.

(path)

def load_local_files(path):
    """Load local BERTopic files."""
    # Load json configs
    topics = load_cfg_from_json(path / TOPICS_NAME)
    params = load_cfg_from_json(path / CONFIG_NAME)

    # Load Topic Embeddings
    safetensor_path = path / HF_SAFE_WEIGHTS_NAME
    if safetensor_path.is_file():
        tensors = load_safetensors(safetensor_path)
    else:
        torch_path = path / HF_WEIGHTS_NAME
        if torch_path.is_file():
            tensors = torch.load(torch_path, map_location="cpu")
            tensors = {k: v.numpy() for k, v in tensors.items()}

    # c-TF-IDF
    try:
        ctfidf_tensors = None
        safetensor_path = path / CTFIDF_SAFE_WEIGHTS_NAME
        if safetensor_path.is_file():
            ctfidf_tensors = load_safetensors(safetensor_path)
        else:
            torch_path = path / CTFIDF_WEIGHTS_NAME
            if torch_path.is_file():
                ctfidf_tensors = torch.load(torch_path, map_location="cpu")
                ctfidf_tensors = {k: v.numpy() for k, v in ctfidf_tensors.items()}
        ctfidf_config = load_cfg_from_json(path / CTFIDF_CFG_NAME)
    except:  # noqa: E722
        ctfidf_config, ctfidf_tensors = None, None

    # Load images
    images = None
    if _has_vision:
        try:
            Image.open(path / "images/0.jpg")
            _has_images = True
        except:  # noqa: E722
            _has_images = False

        if _has_images:
            topic_list = list(topics["topic_representations"].keys())
            images = {}
            for topic in topic_list:
                image = Image.open(path / f"images/{topic}.jpg")
                images[int(topic)] = image

    return topics, params, tensors, ctfidf_tensors, ctfidf_config, images


def load_files_from_hf(path):