Class swarmauri_standard.llms.LlamaCppModel.LlamaCppModel

swarmauri_standard.llms.LlamaCppModel.LlamaCppModel

LlamaCppModel(**data)

Bases: LLMBase

A class for interacting with LlamaCpp's model API for text generation.

This implementation uses httpx for both synchronous and asynchronous HTTP requests, providing support for predictions, streaming responses, and batch processing.

ATTRIBUTE	DESCRIPTION
api_key	API key for authentication (optional for local LlamaCpp) TYPE: Optional[SecretStr]
allowed_models	List of supported model identifiers TYPE: List[str]
name	The currently selected model name TYPE: str
type	Type identifier for the model class TYPE: Literal[LlamaCppModel]
timeout	Request timeout in seconds TYPE: float

Initializes the LlamaCppModel instance and sets up httpx clients for both sync and async operations.

PARAMETER	DESCRIPTION
**data	Keyword arguments for model initialization. TYPE: Dict[str, Any] DEFAULT: {}

Source code in swarmauri_standard/llms/LlamaCppModel.py

def __init__(self, **data):
    """
    Initializes the LlamaCppModel instance and sets up httpx clients
    for both sync and async operations.

    Args:
        **data (Dict[str, Any]): Keyword arguments for model initialization.
    """
    super().__init__(**data)
    if self.api_key:
        headers = {
            "Content-Type": "application/json",
            "Authorization": f"Bearer {self.api_key.get_secret_value()}",
        }
    else:
        headers = {"Content-Type": "application/json"}
    self._client = httpx.Client(
        headers=headers, base_url=self._BASE_URL, timeout=self.timeout
    )
    self._async_client = httpx.AsyncClient(
        headers=headers, base_url=self._BASE_URL, timeout=self.timeout
    )

    self.allowed_models = self.allowed_models or self.get_allowed_models()
    self.name = self.name or self.allowed_models[0]

allowed_models instance-attribute

allowed_models = allowed_models or get_allowed_models()

name instance-attribute

name = name or allowed_models[0]

type class-attribute instance-attribute

type = 'LLMBase'

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

resource class-attribute instance-attribute

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

version class-attribute instance-attribute

version = '0.1.0'

api_key class-attribute instance-attribute

api_key = None

timeout class-attribute instance-attribute

timeout = 600.0

include_usage class-attribute instance-attribute

include_usage = True

BASE_URL class-attribute instance-attribute

BASE_URL = None

predict

predict(
    conversation,
    temperature=0.7,
    max_tokens=256,
    enable_json=False,
    stop=None,
)

Sends a synchronous request to generate a response from the model.

PARAMETER	DESCRIPTION
conversation	The conversation object containing message history. TYPE: Conversation
temperature	Sampling temperature for response generation. TYPE: float DEFAULT: 0.7
max_tokens	Maximum number of tokens to generate. TYPE: int DEFAULT: 256
enable_json	Flag for enabling JSON response format. TYPE: bool DEFAULT: False
stop	Stop sequences for the response. TYPE: Optional[List[str]] DEFAULT: None

RETURNS	DESCRIPTION
Conversation	Updated conversation with the model's response. TYPE: Conversation

Source code in swarmauri_standard/llms/LlamaCppModel.py

@retry_on_status_codes((429, 529), max_retries=1)
def predict(
    self,
    conversation: Conversation,
    temperature: float = 0.7,
    max_tokens: int = 256,
    enable_json: bool = False,
    stop: Optional[List[str]] = None,
) -> Conversation:
    """
    Sends a synchronous request to generate a response from the model.

    Args:
        conversation (Conversation): The conversation object containing message history.
        temperature (float): Sampling temperature for response generation.
        max_tokens (int): Maximum number of tokens to generate.
        enable_json (bool): Flag for enabling JSON response format.
        stop (Optional[List[str]]): Stop sequences for the response.

    Returns:
        Conversation: Updated conversation with the model's response.
    """
    formatted_messages = self._format_messages(conversation.history)
    payload = self._create_request_payload(
        formatted_messages, temperature, max_tokens, enable_json, stop
    )

    response = self._client.post("/chat/completions", json=payload)
    response.raise_for_status()

    result = response.json()
    message_content = result["choices"][0]["message"]["content"]
    conversation.add_message(AgentMessage(content=message_content))

    return conversation

apredict async

apredict(
    conversation,
    temperature=0.7,
    max_tokens=256,
    enable_json=False,
    stop=None,
)

Sends an asynchronous request to generate a response from the model.

PARAMETER	DESCRIPTION
conversation	The conversation object containing message history. TYPE: Conversation
temperature	Sampling temperature for response generation. TYPE: float DEFAULT: 0.7
max_tokens	Maximum number of tokens to generate. TYPE: int DEFAULT: 256
enable_json	Flag for enabling JSON response format. TYPE: bool DEFAULT: False
stop	Stop sequences for the response. TYPE: Optional[List[str]] DEFAULT: None

RETURNS	DESCRIPTION
Conversation	Updated conversation with the model's response. TYPE: Conversation

Source code in swarmauri_standard/llms/LlamaCppModel.py

@retry_on_status_codes((429, 529), max_retries=1)
async def apredict(
    self,
    conversation: Conversation,
    temperature: float = 0.7,
    max_tokens: int = 256,
    enable_json: bool = False,
    stop: Optional[List[str]] = None,
) -> Conversation:
    """
    Sends an asynchronous request to generate a response from the model.

    Args:
        conversation (Conversation): The conversation object containing message history.
        temperature (float): Sampling temperature for response generation.
        max_tokens (int): Maximum number of tokens to generate.
        enable_json (bool): Flag for enabling JSON response format.
        stop (Optional[List[str]]): Stop sequences for the response.

    Returns:
        Conversation: Updated conversation with the model's response.
    """
    formatted_messages = self._format_messages(conversation.history)
    payload = self._create_request_payload(
        formatted_messages, temperature, max_tokens, enable_json, stop
    )

    response = await self._async_client.post("/chat/completions", json=payload)
    response.raise_for_status()

    result = response.json()
    message_content = result["choices"][0]["message"]["content"]
    conversation.add_message(AgentMessage(content=message_content))

    return conversation

stream

stream(
    conversation, temperature=0.7, max_tokens=256, stop=None
)

Streams response content from the model synchronously.

PARAMETER	DESCRIPTION
conversation	The conversation object containing message history. TYPE: Conversation
temperature	Sampling temperature for response generation. TYPE: float DEFAULT: 0.7
max_tokens	Maximum number of tokens to generate. TYPE: int DEFAULT: 256
stop	Stop sequences for the response. TYPE: Optional[List[str]] DEFAULT: None

YIELDS	DESCRIPTION
str	Chunks of content from the model's response. TYPE:: str

Source code in swarmauri_standard/llms/LlamaCppModel.py

@retry_on_status_codes((429, 529), max_retries=1)
def stream(
    self,
    conversation: Conversation,
    temperature: float = 0.7,
    max_tokens: int = 256,
    stop: Optional[List[str]] = None,
) -> Iterator[str]:
    """
    Streams response content from the model synchronously.

    Args:
        conversation (Conversation): The conversation object containing message history.
        temperature (float): Sampling temperature for response generation.
        max_tokens (int): Maximum number of tokens to generate.
        stop (Optional[List[str]]): Stop sequences for the response.

    Yields:
        str: Chunks of content from the model's response.
    """
    formatted_messages = self._format_messages(conversation.history)
    payload = self._create_request_payload(
        formatted_messages, temperature, max_tokens, False, stop, stream=True
    )

    with self._client.stream("POST", "/chat/completions", json=payload) as response:
        response.raise_for_status()
        collected_content = []

        for line in response.iter_lines():
            # Convert bytes to string if necessary
            if isinstance(line, bytes):
                line = line.decode("utf-8")

            if line.startswith("data: "):
                line = line[6:]  # Remove 'data: ' prefix
                if line != "[DONE]":
                    chunk = json.loads(line)
                    if chunk["choices"][0]["delta"].get("content"):
                        content = chunk["choices"][0]["delta"]["content"]
                        collected_content.append(content)
                        yield content

    full_content = "".join(collected_content)
    conversation.add_message(AgentMessage(content=full_content))

astream async

astream(
    conversation, temperature=0.7, max_tokens=256, stop=None
)

Streams response content from the model asynchronously.

PARAMETER	DESCRIPTION
conversation	The conversation object containing message history. TYPE: Conversation
temperature	Sampling temperature for response generation. TYPE: float DEFAULT: 0.7
max_tokens	Maximum number of tokens to generate. TYPE: int DEFAULT: 256
stop	Stop sequences for the response. TYPE: Optional[List[str]] DEFAULT: None

YIELDS	DESCRIPTION
str	Chunks of content from the model's response. TYPE:: AsyncIterator[str]

Source code in swarmauri_standard/llms/LlamaCppModel.py

@retry_on_status_codes((429, 529), max_retries=1)
async def astream(
    self,
    conversation: Conversation,
    temperature: float = 0.7,
    max_tokens: int = 256,
    stop: Optional[List[str]] = None,
) -> AsyncIterator[str]:
    """
    Streams response content from the model asynchronously.

    Args:
        conversation (Conversation): The conversation object containing message history.
        temperature (float): Sampling temperature for response generation.
        max_tokens (int): Maximum number of tokens to generate.
        stop (Optional[List[str]]): Stop sequences for the response.

    Yields:
        str: Chunks of content from the model's response.
    """
    formatted_messages = self._format_messages(conversation.history)
    payload = self._create_request_payload(
        formatted_messages, temperature, max_tokens, False, stop, stream=True
    )

    async with self._async_client.stream(
        "POST", "/chat/completions", json=payload
    ) as response:
        response.raise_for_status()
        collected_content = []

        async for line in response.aiter_lines():
            if line.startswith("data: "):
                line = line[6:]  # Remove 'data: ' prefix
                if line != "[DONE]":
                    chunk = json.loads(line)
                    if chunk["choices"][0]["delta"].get("content"):
                        content = chunk["choices"][0]["delta"]["content"]
                        collected_content.append(content)
                        yield content

    full_content = "".join(collected_content)
    conversation.add_message(AgentMessage(content=full_content))

batch

batch(
    conversations,
    temperature=0.7,
    max_tokens=256,
    enable_json=False,
    stop=None,
)

Processes multiple conversations in batch synchronously.

PARAMETER	DESCRIPTION
conversations	List of conversation objects. TYPE: List[Conversation]
temperature	Sampling temperature for response generation. TYPE: float DEFAULT: 0.7
max_tokens	Maximum number of tokens to generate. TYPE: int DEFAULT: 256
enable_json	Flag for enabling JSON response format. TYPE: bool DEFAULT: False
stop	Stop sequences for responses. TYPE: Optional[List[str]] DEFAULT: None

RETURNS	DESCRIPTION
List[Conversation]	List[Conversation]: List of updated conversations with model responses.

Source code in swarmauri_standard/llms/LlamaCppModel.py

def batch(
    self,
    conversations: List[Conversation],
    temperature: float = 0.7,
    max_tokens: int = 256,
    enable_json: bool = False,
    stop: Optional[List[str]] = None,
) -> List[Conversation]:
    """
    Processes multiple conversations in batch synchronously.

    Args:
        conversations (List[Conversation]): List of conversation objects.
        temperature (float): Sampling temperature for response generation.
        max_tokens (int): Maximum number of tokens to generate.
        enable_json (bool): Flag for enabling JSON response format.
        stop (Optional[List[str]]): Stop sequences for responses.

    Returns:
        List[Conversation]: List of updated conversations with model responses.
    """
    return [
        self.predict(
            conv,
            temperature=temperature,
            max_tokens=max_tokens,
            enable_json=enable_json,
            stop=stop,
        )
        for conv in conversations
    ]

abatch async

abatch(
    conversations,
    temperature=0.7,
    max_tokens=256,
    enable_json=False,
    stop=None,
    max_concurrent=5,
)

Processes multiple conversations asynchronously, with concurrency control.

PARAMETER	DESCRIPTION
conversations	List of conversation objects. TYPE: List[Conversation]
temperature	Sampling temperature for response generation. TYPE: float DEFAULT: 0.7
max_tokens	Maximum number of tokens to generate. TYPE: int DEFAULT: 256
enable_json	Flag for enabling JSON response format. TYPE: bool DEFAULT: False
stop	Stop sequences for responses. TYPE: Optional[List[str]] DEFAULT: None
max_concurrent	Maximum number of concurrent tasks. TYPE: int DEFAULT: 5

RETURNS	DESCRIPTION
List[Conversation]	List[Conversation]: List of updated conversations with model responses.

Source code in swarmauri_standard/llms/LlamaCppModel.py

async def abatch(
    self,
    conversations: List[Conversation],
    temperature: float = 0.7,
    max_tokens: int = 256,
    enable_json: bool = False,
    stop: Optional[List[str]] = None,
    max_concurrent: int = 5,
) -> List[Conversation]:
    """
    Processes multiple conversations asynchronously, with concurrency control.

    Args:
        conversations (List[Conversation]): List of conversation objects.
        temperature (float): Sampling temperature for response generation.
        max_tokens (int): Maximum number of tokens to generate.
        enable_json (bool): Flag for enabling JSON response format.
        stop (Optional[List[str]]): Stop sequences for responses.
        max_concurrent (int): Maximum number of concurrent tasks.

    Returns:
        List[Conversation]: List of updated conversations with model responses.
    """
    semaphore = asyncio.Semaphore(max_concurrent)

    async def process_conversation(conv: Conversation) -> Conversation:
        async with semaphore:
            return await self.apredict(
                conv,
                temperature=temperature,
                max_tokens=max_tokens,
                enable_json=enable_json,
                stop=stop,
            )

    tasks = [process_conversation(conv) for conv in conversations]
    return await asyncio.gather(*tasks)

get_allowed_models

get_allowed_models()

Queries the LLMProvider API endpoint to get the list of allowed models.

RETURNS	DESCRIPTION
List[str]	List[str]: List of allowed model identifiers.

Source code in swarmauri_standard/llms/LlamaCppModel.py

def get_allowed_models(self) -> List[str]:
    """
    Queries the LLMProvider API endpoint to get the list of allowed models.

    Returns:
        List[str]: List of allowed model identifiers.
    """
    response = self._client.get("/models")
    response.raise_for_status()
    models = response.json()
    return [model["id"] for model in models]

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

add_allowed_model

add_allowed_model(model)

Add a new model to the list of allowed models.

RAISES	DESCRIPTION
ValueError	If the model is already in the allowed models list.

Source code in swarmauri_base/llms/LLMBase.py

def add_allowed_model(self, model: str) -> None:
    """
    Add a new model to the list of allowed models.

    Raises:
        ValueError: If the model is already in the allowed models list.
    """
    if model in self.allowed_models:
        raise ValueError(f"Model '{model}' is already allowed.")
    self.allowed_models.append(model)

remove_allowed_model

remove_allowed_model(model)

Remove a model from the list of allowed models.

RAISES	DESCRIPTION
ValueError	If the model is not in the allowed models list.

Source code in swarmauri_base/llms/LLMBase.py

def remove_allowed_model(self, model: str) -> None:
    """
    Remove a model from the list of allowed models.

    Raises:
        ValueError: If the model is not in the allowed models list.
    """
    if model not in self.allowed_models:
        raise ValueError(f"Model '{model}' is not in the allowed models list.")
    self.allowed_models.remove(model)