mirascope 2.0.0a3__py3-none-any.whl → 2.0.0a5__py3-none-any.whl

mirascope/llm/providers/anthropic/_utils/beta_decode.py

@@ -0,0 +1,271 @@
+"""Beta Anthropic response decoding."""
+
+import json
+from typing import Any, TypeAlias, cast
+
+from anthropic.lib.streaming._beta_messages import (
+    BetaAsyncMessageStreamManager,
+    BetaMessageStreamManager,
+)
+from anthropic.types.beta import (
+    BetaContentBlock,
+    BetaRawMessageStreamEvent,
+    BetaRedactedThinkingBlockParam,
+    BetaTextBlockParam,
+    BetaThinkingBlockParam,
+    BetaToolUseBlockParam,
+)
+from anthropic.types.beta.parsed_beta_message import ParsedBetaMessage
+
+from ....content import (
+    AssistantContentPart,
+    Text,
+    TextChunk,
+    TextEndChunk,
+    TextStartChunk,
+    Thought,
+    ThoughtChunk,
+    ThoughtEndChunk,
+    ThoughtStartChunk,
+    ToolCall,
+    ToolCallChunk,
+    ToolCallEndChunk,
+    ToolCallStartChunk,
+)
+from ....messages import AssistantMessage
+from ....responses import (
+    AsyncChunkIterator,
+    ChunkIterator,
+    FinishReason,
+    FinishReasonChunk,
+    RawMessageChunk,
+    RawStreamEventChunk,
+    Usage,
+    UsageDeltaChunk,
+)
+from ..model_id import model_name
+from .decode import decode_usage
+
+BETA_FINISH_REASON_MAP = {
+    "max_tokens": FinishReason.MAX_TOKENS,
+    "refusal": FinishReason.REFUSAL,
+    "model_context_window_exceeded": FinishReason.CONTEXT_LENGTH_EXCEEDED,
+}
+
+
+def _decode_beta_assistant_content(content: BetaContentBlock) -> AssistantContentPart:
+    """Convert Beta content block to mirascope AssistantContentPart."""
+    if content.type == "text":
+        return Text(text=content.text)
+    elif content.type == "tool_use":
+        return ToolCall(
+            id=content.id,
+            name=content.name,
+            args=json.dumps(content.input),
+        )
+    elif content.type == "thinking":
+        return Thought(thought=content.thinking)
+    else:
+        raise NotImplementedError(
+            f"Support for beta content type `{content.type}` is not yet implemented."
+        )
+
+
+def beta_decode_response(
+    response: ParsedBetaMessage[Any],
+    model_id: str,
+) -> tuple[AssistantMessage, FinishReason | None, Usage]:
+    """Convert Beta message to mirascope AssistantMessage and usage."""
+    assistant_message = AssistantMessage(
+        content=[_decode_beta_assistant_content(part) for part in response.content],
+        provider_id="anthropic",
+        model_id=model_id,
+        provider_model_name=model_name(model_id),
+        raw_message={
+            "role": response.role,
+            "content": [
+                part.model_dump(exclude_none=True) for part in response.content
+            ],
+        },
+    )
+    finish_reason = (
+        BETA_FINISH_REASON_MAP.get(response.stop_reason)
+        if response.stop_reason
+        else None
+    )
+    usage = decode_usage(response.usage)
+    return assistant_message, finish_reason, usage
+
+
+BetaContentBlockParam: TypeAlias = (
+    BetaTextBlockParam
+    | BetaThinkingBlockParam
+    | BetaToolUseBlockParam
+    | BetaRedactedThinkingBlockParam
+)
+
+
+class _BetaChunkProcessor:
+    """Processes Beta stream events and maintains state across events."""
+
+    def __init__(self) -> None:
+        self.current_block_param: BetaContentBlockParam | None = None
+        self.accumulated_tool_json: str = ""
+        self.accumulated_blocks: list[BetaContentBlockParam] = []
+
+    def process_event(self, event: BetaRawMessageStreamEvent) -> ChunkIterator:
+        """Process a single Beta event and yield the appropriate content chunks."""
+        yield RawStreamEventChunk(raw_stream_event=event)
+
+        if event.type == "content_block_start":
+            content_block = event.content_block
+
+            if content_block.type == "text":
+                self.current_block_param = {
+                    "type": "text",
+                    "text": content_block.text,
+                }
+                yield TextStartChunk()
+            elif content_block.type == "tool_use":
+                self.current_block_param = {
+                    "type": "tool_use",
+                    "id": content_block.id,
+                    "name": content_block.name,
+                    "input": {},
+                }
+                self.accumulated_tool_json = ""
+                yield ToolCallStartChunk(
+                    id=content_block.id,
+                    name=content_block.name,
+                )
+            elif content_block.type == "thinking":
+                self.current_block_param = {
+                    "type": "thinking",
+                    "thinking": "",
+                    "signature": "",
+                }
+                yield ThoughtStartChunk()
+            elif content_block.type == "redacted_thinking":  # pragma: no cover
+                self.current_block_param = {
+                    "type": "redacted_thinking",
+                    "data": content_block.data,
+                }
+            else:
+                raise NotImplementedError(
+                    f"Support for beta content block type `{content_block.type}` "
+                    "is not yet implemented."
+                )
+
+        elif event.type == "content_block_delta":
+            if self.current_block_param is None:  # pragma: no cover
+                raise RuntimeError("Received delta without a current block")
+
+            delta = event.delta
+            if delta.type == "text_delta":
+                if self.current_block_param["type"] != "text":  # pragma: no cover
+                    raise RuntimeError(
+                        f"Received text_delta for {self.current_block_param['type']} block"
+                    )
+                self.current_block_param["text"] += delta.text
+                yield TextChunk(delta=delta.text)
+            elif delta.type == "input_json_delta":
+                if self.current_block_param["type"] != "tool_use":  # pragma: no cover
+                    raise RuntimeError(
+                        f"Received input_json_delta for {self.current_block_param['type']} block"
+                    )
+                self.accumulated_tool_json += delta.partial_json
+                yield ToolCallChunk(delta=delta.partial_json)
+            elif delta.type == "thinking_delta":
+                if self.current_block_param["type"] != "thinking":  # pragma: no cover
+                    raise RuntimeError(
+                        f"Received thinking_delta for {self.current_block_param['type']} block"
+                    )
+                self.current_block_param["thinking"] += delta.thinking
+                yield ThoughtChunk(delta=delta.thinking)
+            elif delta.type == "signature_delta":
+                if self.current_block_param["type"] != "thinking":  # pragma: no cover
+                    raise RuntimeError(
+                        f"Received signature_delta for {self.current_block_param['type']} block"
+                    )
+                self.current_block_param["signature"] += delta.signature
+            else:
+                raise RuntimeError(
+                    f"Received unsupported delta type: {delta.type}"
+                )  # pragma: no cover
+
+        elif event.type == "content_block_stop":
+            if self.current_block_param is None:  # pragma: no cover
+                raise RuntimeError("Received stop without a current block")
+
+            block_type = self.current_block_param["type"]
+
+            if block_type == "text":
+                yield TextEndChunk()
+            elif block_type == "tool_use":
+                if self.current_block_param["type"] != "tool_use":  # pragma: no cover
+                    raise RuntimeError(
+                        f"Block type mismatch: stored {self.current_block_param['type']}, expected tool_use"
+                    )
+                self.current_block_param["input"] = (
+                    json.loads(self.accumulated_tool_json)
+                    if self.accumulated_tool_json
+                    else {}
+                )
+                yield ToolCallEndChunk()
+            elif block_type == "thinking":
+                yield ThoughtEndChunk()
+            else:
+                raise NotImplementedError
+
+            self.accumulated_blocks.append(self.current_block_param)
+            self.current_block_param = None
+
+        elif event.type == "message_delta":
+            if event.delta.stop_reason:
+                finish_reason = BETA_FINISH_REASON_MAP.get(event.delta.stop_reason)
+                if finish_reason is not None:
+                    yield FinishReasonChunk(finish_reason=finish_reason)
+
+            # Emit usage delta
+            usage = event.usage
+            yield UsageDeltaChunk(
+                input_tokens=usage.input_tokens or 0,
+                output_tokens=usage.output_tokens,
+                cache_read_tokens=usage.cache_read_input_tokens or 0,
+                cache_write_tokens=usage.cache_creation_input_tokens or 0,
+                reasoning_tokens=0,
+            )
+
+    def raw_message_chunk(self) -> RawMessageChunk:
+        return RawMessageChunk(
+            raw_message=cast(
+                dict[str, Any],
+                {
+                    "role": "assistant",
+                    "content": self.accumulated_blocks,
+                },
+            )
+        )
+
+
+def beta_decode_stream(
+    beta_stream_manager: BetaMessageStreamManager[Any],
+) -> ChunkIterator:
+    """Returns a ChunkIterator converted from a Beta MessageStreamManager."""
+    processor = _BetaChunkProcessor()
+    with beta_stream_manager as stream:
+        for event in stream._raw_stream:  # pyright: ignore[reportPrivateUsage]
+            yield from processor.process_event(event)
+    yield processor.raw_message_chunk()
+
+
+async def beta_decode_async_stream(
+    beta_stream_manager: BetaAsyncMessageStreamManager[Any],
+) -> AsyncChunkIterator:
+    """Returns an AsyncChunkIterator converted from a Beta MessageStreamManager."""
+    processor = _BetaChunkProcessor()
+    async with beta_stream_manager as stream:
+        async for event in stream._raw_stream:  # pyright: ignore[reportPrivateUsage]
+            for item in processor.process_event(event):
+                yield item
+    yield processor.raw_message_chunk()

mirascope/llm/providers/anthropic/_utils/beta_encode.py

@@ -0,0 +1,216 @@
+"""Beta Anthropic message encoding and request preparation."""
+
+from collections.abc import Sequence
+from typing import Any, TypedDict, cast
+from typing_extensions import Required
+
+from anthropic import Omit
+from anthropic.types.anthropic_beta_param import AnthropicBetaParam
+from anthropic.types.beta import (
+    BetaContentBlockParam,
+    BetaMessageParam,
+    BetaTextBlockParam,
+    BetaThinkingConfigParam,
+    BetaToolChoiceParam,
+    BetaToolParam,
+)
+from pydantic import BaseModel
+
+from ....content import ContentPart
+from ....exceptions import FormattingModeNotSupportedError
+from ....formatting import (
+    Format,
+    FormattableT,
+    _utils as _formatting_utils,
+    resolve_format,
+)
+from ....messages import AssistantMessage, Message, UserMessage
+from ....tools import AnyToolSchema, BaseToolkit
+from ...base import Params, _utils as _base_utils
+from ..model_id import model_name
+from ..model_info import MODELS_WITHOUT_STRICT_STRUCTURED_OUTPUTS
+from .encode import (
+    DEFAULT_MAX_TOKENS,
+    FORMAT_TOOL_NAME,
+    convert_tool_to_tool_param,
+    encode_content,
+    process_params,
+)
+
+DEFAULT_FORMAT_MODE = "strict"
+
+
+class BetaParseKwargs(TypedDict, total=False):
+    """Kwargs for Anthropic beta.messages.parse method."""
+
+    model: Required[str]
+    max_tokens: Required[int]
+    messages: Sequence[BetaMessageParam]
+    system: Sequence[BetaTextBlockParam] | Omit
+    tools: Sequence[BetaToolParam] | Omit
+    tool_choice: BetaToolChoiceParam | Omit
+    temperature: float | Omit
+    top_p: float | Omit
+    top_k: int | Omit
+    stop_sequences: list[str] | Omit
+    thinking: BetaThinkingConfigParam | Omit
+    betas: list[AnthropicBetaParam]
+    output_format: type[BaseModel]
+
+
+def _beta_encode_content(
+    content: Sequence[ContentPart],
+    encode_thoughts: bool,
+    add_cache_control: bool = False,
+) -> str | Sequence[BetaContentBlockParam]:
+    """Convert mirascope content to Beta Anthropic content format."""
+    result = encode_content(content, encode_thoughts, add_cache_control)
+    if isinstance(result, str):
+        return result
+    return cast(Sequence[BetaContentBlockParam], result)
+
+
+def _beta_encode_message(
+    message: UserMessage | AssistantMessage,
+    model_id: str,
+    encode_thoughts: bool,
+    add_cache_control: bool = False,
+) -> BetaMessageParam:
+    """Convert user or assistant Message to Beta MessageParam format.
+
+    Args:
+        message: The message to encode
+        model_id: The Anthropic model ID
+        encode_thoughts: Whether to encode thought blocks as text
+        add_cache_control: Whether to add cache_control to the last content block
+    """
+    if (
+        message.role == "assistant"
+        and message.provider_id == "anthropic"
+        and message.model_id == model_id
+        and message.raw_message
+        and not encode_thoughts
+        and not add_cache_control
+    ):
+        raw = cast(dict[str, Any], message.raw_message)
+        return BetaMessageParam(
+            role=raw["role"],
+            content=raw["content"],
+        )
+
+    content = _beta_encode_content(message.content, encode_thoughts, add_cache_control)
+
+    return BetaMessageParam(
+        role=message.role,
+        content=content,
+    )
+
+
+def _beta_encode_messages(
+    messages: Sequence[UserMessage | AssistantMessage],
+    model_id: str,
+    encode_thoughts: bool,
+) -> Sequence[BetaMessageParam]:
+    """Encode messages and add cache control for multi-turn conversations.
+
+    If the conversation contains assistant messages (indicating multi-turn),
+    adds cache_control to the last content block of the last message.
+    """
+    # Detect multi-turn conversations by checking for assistant messages
+    has_assistant_message = any(msg.role == "assistant" for msg in messages)
+
+    # Encode messages, adding cache_control to the last message if multi-turn
+    encoded_messages: list[BetaMessageParam] = []
+    for i, message in enumerate(messages):
+        is_last = i == len(messages) - 1
+        add_cache = has_assistant_message and is_last
+        encoded_messages.append(
+            _beta_encode_message(message, model_id, encode_thoughts, add_cache)
+        )
+    return encoded_messages
+
+
+def _beta_convert_tool_to_tool_param(tool: AnyToolSchema) -> BetaToolParam:
+    """Convert a single Mirascope tool to Beta Anthropic tool format."""
+    return cast(BetaToolParam, convert_tool_to_tool_param(tool))
+
+
+def beta_encode_request(
+    *,
+    model_id: str,
+    messages: Sequence[Message],
+    tools: Sequence[AnyToolSchema] | BaseToolkit[AnyToolSchema] | None,
+    format: type[FormattableT] | Format[FormattableT] | None,
+    params: Params,
+) -> tuple[Sequence[Message], Format[FormattableT] | None, BetaParseKwargs]:
+    """Prepares a request for the Anthropic beta.messages.parse method."""
+
+    processed = process_params(params, DEFAULT_MAX_TOKENS)
+    encode_thoughts = processed.pop("encode_thoughts", False)
+    max_tokens = processed.pop("max_tokens", DEFAULT_MAX_TOKENS)
+
+    kwargs: BetaParseKwargs = BetaParseKwargs(
+        {
+            "model": model_name(model_id),
+            "max_tokens": max_tokens,
+            "betas": ["structured-outputs-2025-11-13"],
+            **processed,
+        }
+    )
+
+    tools = tools.tools if isinstance(tools, BaseToolkit) else tools or []
+    anthropic_tools = [_beta_convert_tool_to_tool_param(tool) for tool in tools]
+    format = resolve_format(format, default_mode=DEFAULT_FORMAT_MODE)
+
+    if format is not None:
+        if format.mode == "strict":
+            if model_name(model_id) in MODELS_WITHOUT_STRICT_STRUCTURED_OUTPUTS:
+                raise FormattingModeNotSupportedError(
+                    formatting_mode=format.mode,
+                    provider_id="anthropic",
+                    model_id=model_id,
+                )
+            else:
+                kwargs["output_format"] = cast(type[BaseModel], format.formattable)
+
+        if format.mode == "tool":
+            format_tool_schema = _formatting_utils.create_tool_schema(format)
+            anthropic_tools.append(_beta_convert_tool_to_tool_param(format_tool_schema))
+            if tools:
+                kwargs["tool_choice"] = {"type": "any"}
+            else:
+                kwargs["tool_choice"] = {
+                    "type": "tool",
+                    "name": FORMAT_TOOL_NAME,
+                    "disable_parallel_tool_use": True,
+                }
+
+        if format.formatting_instructions:
+            messages = _base_utils.add_system_instructions(
+                messages, format.formatting_instructions
+            )
+
+    if anthropic_tools:
+        # Add cache control to the last tool for prompt caching
+        last_tool = anthropic_tools[-1]
+        last_tool["cache_control"] = {"type": "ephemeral"}
+        kwargs["tools"] = anthropic_tools
+
+    system_message_content, remaining_messages = _base_utils.extract_system_message(
+        messages
+    )
+
+    kwargs["messages"] = _beta_encode_messages(
+        remaining_messages, model_id, encode_thoughts
+    )
+
+    if system_message_content:
+        kwargs["system"] = [
+            BetaTextBlockParam(
+                type="text",
+                text=system_message_content,
+                cache_control={"type": "ephemeral"},
+            )
+        ]
+
+    return messages, format, kwargs

mirascope/llm/providers/anthropic/_utils/decode.py

@@ -1,10 +1,11 @@
-"""Anthropic response decoding."""
+"""Standard Anthropic response decoding."""
 
 import json
 from typing import Any, TypeAlias, cast
 
 from anthropic import types as anthropic_types
 from anthropic.lib.streaming import AsyncMessageStreamManager, MessageStreamManager
+from anthropic.types.beta import BetaUsage
 
 from ....content import (
     AssistantContentPart,
@@ -29,6 +30,8 @@ from ....responses import (
     FinishReasonChunk,
     RawMessageChunk,
     RawStreamEventChunk,
+    Usage,
+    UsageDeltaChunk,
 )
 from ..model_id import AnthropicModelId, model_name
 
@@ -58,11 +61,30 @@ def _decode_assistant_content(
         )
 
 
+def decode_usage(
+    usage: anthropic_types.Usage | BetaUsage,
+) -> Usage:
+    """Convert Anthropic Usage (or BetaUsage) to Mirascope Usage."""
+
+    cache_read_tokens = usage.cache_read_input_tokens or 0
+    cache_write_tokens = usage.cache_creation_input_tokens or 0
+    input_tokens = usage.input_tokens + cache_read_tokens + cache_write_tokens
+    output_tokens = usage.output_tokens
+    return Usage(
+        input_tokens=input_tokens,
+        output_tokens=output_tokens,
+        cache_read_tokens=cache_read_tokens,
+        cache_write_tokens=cache_write_tokens,
+        reasoning_tokens=0,
+        raw=usage,
+    )
+
+
 def decode_response(
     response: anthropic_types.Message,
     model_id: AnthropicModelId,
-) -> tuple[AssistantMessage, FinishReason | None]:
-    """Convert Anthropic message to mirascope AssistantMessage."""
+) -> tuple[AssistantMessage, FinishReason | None, Usage]:
+    """Convert Anthropic message to mirascope AssistantMessage and usage."""
     assistant_message = AssistantMessage(
         content=[_decode_assistant_content(part) for part in response.content],
         provider_id="anthropic",
@@ -78,14 +100,14 @@ def decode_response(
         if response.stop_reason
         else None
     )
-    return assistant_message, finish_reason
+    usage = decode_usage(response.usage)
+    return assistant_message, finish_reason, usage
 
 
 ContentBlock: TypeAlias = (
     anthropic_types.TextBlockParam
     | anthropic_types.ThinkingBlockParam
     | anthropic_types.ToolUseBlockParam
-    | anthropic_types.ThinkingBlockParam
     | anthropic_types.RedactedThinkingBlockParam
 )
 
@@ -210,6 +232,16 @@ class _AnthropicChunkProcessor:
                 if finish_reason is not None:
                     yield FinishReasonChunk(finish_reason=finish_reason)
 
+            # Emit usage delta
+            usage = event.usage
+            yield UsageDeltaChunk(
+                input_tokens=usage.input_tokens or 0,
+                output_tokens=usage.output_tokens,
+                cache_read_tokens=usage.cache_read_input_tokens or 0,
+                cache_write_tokens=usage.cache_creation_input_tokens or 0,
+                reasoning_tokens=0,
+            )
+
     def raw_message_chunk(self) -> RawMessageChunk:
         return RawMessageChunk(
             raw_message=cast(
@@ -225,7 +257,7 @@ class _AnthropicChunkProcessor:
 def decode_stream(
     anthropic_stream_manager: MessageStreamManager,
 ) -> ChunkIterator:
-    """Returns a ChunkIterator converted from an Anthropic MessageStreamManager"""
+    """Returns a ChunkIterator converted from an Anthropic MessageStreamManager."""
     processor = _AnthropicChunkProcessor()
     with anthropic_stream_manager as stream:
         for event in stream._raw_stream:  # pyright: ignore[reportPrivateUsage]
@@ -236,7 +268,7 @@ def decode_stream(
 async def decode_async_stream(
     anthropic_stream_manager: AsyncMessageStreamManager,
 ) -> AsyncChunkIterator:
-    """Returns an AsyncChunkIterator converted from an Anthropic MessageStreamManager"""
+    """Returns an AsyncChunkIterator converted from an Anthropic MessageStreamManager."""
     processor = _AnthropicChunkProcessor()
     async with anthropic_stream_manager as stream:
         async for event in stream._raw_stream:  # pyright: ignore[reportPrivateUsage]