Class swarmauri_parser_bertembedding.BERTEmbeddingParser.BERTEmbeddingParser

swarmauri_parser_bertembedding.BERTEmbeddingParser.BERTEmbeddingParser

BERTEmbeddingParser(**kwargs)

Bases: ParserBase

A parser that transforms input text into document embeddings using BERT.

This parser tokenizes the input text, passes it through a pre-trained BERT model, and uses the resulting embeddings as the document content.

Initializes the BERTEmbeddingParser with a specific BERT model.

Parameters: - model_name (str): The name of the pre-trained BERT model to use.

Source code in swarmauri_parser_bertembedding/BERTEmbeddingParser.py

def __init__(self, **kwargs):
    """
    Initializes the BERTEmbeddingParser with a specific BERT model.

    Parameters:
    - model_name (str): The name of the pre-trained BERT model to use.
    """
    super().__init__(**kwargs)
    self._tokenizer = BertTokenizer.from_pretrained(self.parser_model_name)
    self._model = BertModel.from_pretrained(self.parser_model_name)
    self._model.eval()  # Set model to evaluation mode

parser_model_name class-attribute instance-attribute

parser_model_name = 'bert-base-uncased'

type class-attribute instance-attribute

type = 'BERTEmbeddingParser'

model_config class-attribute instance-attribute

model_config = ConfigDict(
    extra="allow", arbitrary_types_allowed=True
)

id class-attribute instance-attribute

id = Field(default_factory=generate_id)

members class-attribute instance-attribute

members = None

owners class-attribute instance-attribute

owners = None

host class-attribute instance-attribute

host = None

default_logger class-attribute

default_logger = None

logger class-attribute instance-attribute

logger = None

name class-attribute instance-attribute

name = None

resource class-attribute instance-attribute

resource = Field(default=PARSER.value, frozen=True)

version class-attribute instance-attribute

version = '0.1.0'

parse

parse(data)

Tokenizes input data and generates embeddings using a BERT model.

Parameters: - data (Union[str, Any]): Input data, expected to be a single string or batch of strings.

Returns: - List[IDocument]: A list containing a single IDocument instance with BERT embeddings as content.

Source code in swarmauri_parser_bertembedding/BERTEmbeddingParser.py

def parse(self, data: Union[str, Any]) -> List[IDocument]:
    """
    Tokenizes input data and generates embeddings using a BERT model.

    Parameters:
    - data (Union[str, Any]): Input data, expected to be a single string or batch of strings.

    Returns:
    - List[IDocument]: A list containing a single IDocument instance with BERT embeddings as content.
    """

    # Tokenization
    inputs = self._tokenizer(
        data, return_tensors="pt", padding=True, truncation=True, max_length=512
    )

    # Generate embeddings
    with torch.no_grad():
        outputs = self._model(**inputs)

    # Use the last hidden state as document embeddings (batch_size, sequence_length, hidden_size)
    embeddings = outputs.last_hidden_state

    # Convert to list of numpy arrays
    embeddings = embeddings.detach().cpu().numpy()

    # For simplicity, let's consider the mean of embeddings across tokens to represent the document
    doc_embeddings = embeddings.mean(axis=1)

    # Creating document object(s)
    documents = []
    for i, emb in enumerate(doc_embeddings):
        # Store the original input text as content and embeddings in metadata
        input_text = data[i] if isinstance(data, list) else data
        doc = Document(content=input_text)
        # Add the embedding and source to metadata
        doc.metadata = {"source": "BERTEmbeddingParser", "embedding": emb}
        # Set an ID if needed in your application
        doc.id = str(i)
        documents.append(doc)

    return documents

register_model classmethod

register_model()

Decorator to register a base model in the unified registry.

RETURNS	DESCRIPTION
Callable	A decorator function that registers the model class. TYPE: Callable[[Type[BaseModel]], Type[BaseModel]]

Source code in swarmauri_base/DynamicBase.py

@classmethod
def register_model(cls) -> Callable[[Type[BaseModel]], Type[BaseModel]]:
    """
    Decorator to register a base model in the unified registry.

    Returns:
        Callable: A decorator function that registers the model class.
    """

    def decorator(model_cls: Type[BaseModel]):
        """Register ``model_cls`` as a base model."""
        model_name = model_cls.__name__
        if model_name in cls._registry:
            glogger.warning(
                "Model '%s' is already registered; skipping duplicate.", model_name
            )
            return model_cls

        cls._registry[model_name] = {"model_cls": model_cls, "subtypes": {}}
        glogger.debug("Registered base model '%s'.", model_name)
        DynamicBase._recreate_models()
        return model_cls

    return decorator

register_type classmethod

register_type(resource_type=None, type_name=None)

Decorator to register a subtype under one or more base models in the unified registry.

PARAMETER	DESCRIPTION
resource_type	The base model(s) under which to register the subtype. If None, all direct base classes (except DynamicBase) are used. TYPE: Optional[Union[Type[T], List[Type[T]]]] DEFAULT: None
type_name	An optional custom type name for the subtype. TYPE: Optional[str] DEFAULT: None

RETURNS	DESCRIPTION
Callable	A decorator function that registers the subtype. TYPE: Callable[[Type[DynamicBase]], Type[DynamicBase]]

Source code in swarmauri_base/DynamicBase.py

@classmethod
def register_type(
    cls,
    resource_type: Optional[Union[Type[T], List[Type[T]]]] = None,
    type_name: Optional[str] = None,
) -> Callable[[Type["DynamicBase"]], Type["DynamicBase"]]:
    """
    Decorator to register a subtype under one or more base models in the unified registry.

    Parameters:
        resource_type (Optional[Union[Type[T], List[Type[T]]]]):
            The base model(s) under which to register the subtype. If None, all direct base classes (except DynamicBase)
            are used.
        type_name (Optional[str]): An optional custom type name for the subtype.

    Returns:
        Callable: A decorator function that registers the subtype.
    """

    def decorator(subclass: Type["DynamicBase"]):
        """Register ``subclass`` as a subtype."""
        if resource_type is None:
            resource_types = [
                base for base in subclass.__bases__ if base is not cls
            ]
        elif not isinstance(resource_type, list):
            resource_types = [resource_type]
        else:
            resource_types = resource_type

        for rt in resource_types:
            if not issubclass(subclass, rt):
                raise TypeError(
                    f"'{subclass.__name__}' must be a subclass of '{rt.__name__}'."
                )
            final_type_name = type_name or getattr(
                subclass, "_type", subclass.__name__
            )
            base_model_name = rt.__name__

            if base_model_name not in cls._registry:
                cls._registry[base_model_name] = {"model_cls": rt, "subtypes": {}}
                glogger.debug(
                    "Created new registry entry for base model '%s'.",
                    base_model_name,
                )

            subtypes_dict = cls._registry[base_model_name]["subtypes"]
            if final_type_name in subtypes_dict:
                glogger.warning(
                    "Type '%s' already exists under '%s'; skipping duplicate.",
                    final_type_name,
                    base_model_name,
                )
                continue

            subtypes_dict[final_type_name] = subclass
            glogger.debug(
                "Registered '%s' as '%s' under '%s'.",
                subclass.__name__,
                final_type_name,
                base_model_name,
            )

        DynamicBase._recreate_models()
        return subclass

    return decorator

model_validate_toml classmethod

model_validate_toml(toml_data)

Validate a model from a TOML string.

Source code in swarmauri_base/TomlMixin.py

@classmethod
def model_validate_toml(cls, toml_data: str):
    """Validate a model from a TOML string."""
    try:
        # Parse TOML into a Python dictionary
        toml_content = tomllib.loads(toml_data)

        # Convert the dictionary to JSON and validate using Pydantic
        return cls.model_validate_json(json.dumps(toml_content))
    except tomllib.TOMLDecodeError as e:
        raise ValueError(f"Invalid TOML data: {e}")
    except ValidationError as e:
        raise ValueError(f"Validation failed: {e}")

model_dump_toml

model_dump_toml(
    fields_to_exclude=None, api_key_placeholder=None
)

Return a TOML representation of the model.

Source code in swarmauri_base/TomlMixin.py

def model_dump_toml(self, fields_to_exclude=None, api_key_placeholder=None):
    """Return a TOML representation of the model."""
    if fields_to_exclude is None:
        fields_to_exclude = []

    # Load the JSON string into a Python dictionary
    json_data = json.loads(self.model_dump_json())

    # Function to recursively remove specific keys and handle api_key placeholders
    def process_fields(data, fields_to_exclude):
        """Recursively filter fields and apply placeholders."""
        if isinstance(data, dict):
            return {
                key: (
                    api_key_placeholder
                    if key == "api_key" and api_key_placeholder is not None
                    else process_fields(value, fields_to_exclude)
                )
                for key, value in data.items()
                if key not in fields_to_exclude
            }
        elif isinstance(data, list):
            return [process_fields(item, fields_to_exclude) for item in data]
        else:
            return data

    # Filter the JSON data
    filtered_data = process_fields(json_data, fields_to_exclude)

    # Convert the filtered data into TOML
    return toml.dumps(filtered_data)

model_validate_yaml classmethod

model_validate_yaml(yaml_data)

Validate a model from a YAML string.

Source code in swarmauri_base/YamlMixin.py

@classmethod
def model_validate_yaml(cls, yaml_data: str):
    """Validate a model from a YAML string."""
    try:
        # Parse YAML into a Python dictionary
        yaml_content = yaml.safe_load(yaml_data)

        # Convert the dictionary to JSON and validate using Pydantic
        return cls.model_validate_json(json.dumps(yaml_content))
    except yaml.YAMLError as e:
        raise ValueError(f"Invalid YAML data: {e}")
    except ValidationError as e:
        raise ValueError(f"Validation failed: {e}")

model_dump_yaml

model_dump_yaml(
    fields_to_exclude=None, api_key_placeholder=None
)

Return a YAML representation of the model.

Source code in swarmauri_base/YamlMixin.py

def model_dump_yaml(self, fields_to_exclude=None, api_key_placeholder=None):
    """Return a YAML representation of the model."""
    if fields_to_exclude is None:
        fields_to_exclude = []

    # Load the JSON string into a Python dictionary
    json_data = json.loads(self.model_dump_json())

    # Function to recursively remove specific keys and handle api_key placeholders
    def process_fields(data, fields_to_exclude):
        """Recursively filter fields and apply placeholders."""
        if isinstance(data, dict):
            return {
                key: (
                    api_key_placeholder
                    if key == "api_key" and api_key_placeholder is not None
                    else process_fields(value, fields_to_exclude)
                )
                for key, value in data.items()
                if key not in fields_to_exclude
            }
        elif isinstance(data, list):
            return [process_fields(item, fields_to_exclude) for item in data]
        else:
            return data

    # Filter the JSON data
    filtered_data = process_fields(json_data, fields_to_exclude)

    # Convert the filtered data into YAML using safe mode
    return yaml.safe_dump(filtered_data, default_flow_style=False)

model_post_init

model_post_init(logger=None)

Assign a logger instance after model initialization.

Source code in swarmauri_base/LoggerMixin.py

def model_post_init(self, logger: Optional[FullUnion[LoggerBase]] = None) -> None:
    """Assign a logger instance after model initialization."""

    # Directly assign the provided FullUnion[LoggerBase] or fallback to the
    # class-level default.
    self.logger = self.logger or logger or self.default_logger