distill::backends::anthropic_backend::AnthropicBackend¶

Inherits from distill.backends.base.TeacherBackend, ABC

Public Functions¶

	Name
	init(self self, dict endpoint_config, str model_id, str api_key)
str	backend_type(self self)
dict	generate(self self, list messages, int max_tokens, float temperature, ** kwargs)

Protected Functions¶

	Name
dict	_convert_messages(list messages)

Protected Attributes¶

	Name
	_client

Additional inherited members¶

Public Functions inherited from distill.backends.base.TeacherBackend

	Name
float	estimate_cost(int prompt_tokens, int completion_tokens)

Protected Attributes inherited from distill.backends.base.TeacherBackend

	Name
	_endpoint_config
	_model_id
	_api_key

Detailed Description¶

class distill::backends::anthropic_backend::AnthropicBackend;

Teacher backend that talks to the Anthropic Messages API.

Used for endpoints whose ``apiType`` is ``"anthropic"`` — either a
direct Anthropic API connection or an Anthropic-compatible proxy.

Public Functions Documentation¶

function init¶

__init__(
    self self,
    dict endpoint_config,
    str model_id,
    str api_key
)

Reimplements: distill::backends::base::TeacherBackend::init

Initialise the backend with connection details.

Args:
    endpoint_config: The resolved ``endpoints[name]`` dict from
        pipeline.yaml (contains ``url``, ``apiType``, etc.).
    model_id: The literal model identifier from the ``models``
        config block (e.g. ``"deepseek-v4-pro[1m]"``).
    api_key: The API key to authenticate requests.

function backend_type¶

str backend_type(
    self self
)

Reimplements: distill::backends::base::TeacherBackend::backend_type

Return a short identifier for this backend (e.g. ``"openai"``).```


### function generate

```python
dict generate(
    self self,
    list messages,
    int max_tokens,
    float temperature,
    ** kwargs
)

Reimplements: distill::backends::base::TeacherBackend::generate

Send a completion via the Anthropic Messages API.

Converts OpenAI-format messages to Anthropic format, extracts
the first system message as the top-level ``system`` parameter,
and normalises the response into the uniform dict.

Extended thinking (``thinking={"type": "enabled", ...}``) is
passed through in ``**kwargs`` if supplied.

Returns:
    Uniform dict with ``content``, ``prompt_tokens``,
    ``completion_tokens``, ``raw_response``.

Protected Functions Documentation¶

function _convert_messages¶

static dict _convert_messages(
    list messages
)

Convert an OpenAI-format message list to Anthropic API parameters.

Args:
    messages: List of dicts with ``role`` and ``content`` keys.
        Roles may be ``"system"``, ``"user"``, or ``"assistant"``.

Returns:
    dict with keys ``system`` (str or None) and ``messages`` (list
    of ``{"role": ..., "content": ...}`` dicts containing only
    ``"user"`` and ``"assistant"`` roles).

Protected Attributes Documentation¶

variable _client¶

_client;

Updated on 2026-07-25 at 22:56:57 +0000