mirascope 2.0.0a6__py3-none-any.whl → 2.0.2__py3-none-any.whl

mirascope/ops/_internal/instrumentation/llm/cost.py

@@ -0,0 +1,190 @@
+"""Cost calculation utilities for LLM instrumentation."""
+
+from __future__ import annotations
+
+import logging
+import re
+from typing import Any
+
+from .....api._generated.token_cost import (
+    TokenCostCalculateRequestUsage,
+    TokenCostCalculateResponse,
+)
+from .....api.client import get_async_client, get_sync_client
+from .....llm.providers import get_provider_for_model
+from .....llm.providers.mirascope import MirascopeProvider
+from .....llm.responses.usage import Usage
+
+logger = logging.getLogger(__name__)
+
+
+def _is_via_router(model_id: str) -> bool:
+    """Check if the model_id routes through MirascopeProvider (Mirascope Router).
+
+    Args:
+        model_id: The full model ID (e.g., "openai/gpt-4o").
+
+    Returns:
+        True if the registered provider for this model is MirascopeProvider.
+    """
+    try:
+        provider = get_provider_for_model(model_id)
+        return isinstance(provider, MirascopeProvider)
+    except Exception:
+        return False
+
+
+def _normalize_model_id(model_id: str) -> str:
+    """Normalize model_id by stripping provider prefix and any :suffix.
+
+    Args:
+        model_id: The full model ID (e.g., "openai/gpt-4o:responses").
+
+    Returns:
+        The base model name (e.g., "gpt-4o").
+    """
+    # Strip provider prefix (e.g., "openai/gpt-4o" -> "gpt-4o")
+    if "/" in model_id:
+        model_id = model_id.split("/", 1)[1]
+    # Strip any :suffix (e.g., ":responses", ":completions")
+    return re.sub(r":.*$", "", model_id)
+
+
+def extract_cache_write_breakdown(usage: Usage) -> dict[str, float] | None:
+    """Extract cache write breakdown from Anthropic's raw usage data.
+
+    Returns a dict like {"ephemeral5m": 100, "ephemeral1h": 50} if available,
+    or None if the breakdown isn't available.
+    """
+    raw = usage.raw
+    if raw is None:
+        return None
+
+    # Check for Anthropic's cache_creation breakdown
+    cache_creation = getattr(raw, "cache_creation", None)
+    if cache_creation is None:
+        return None
+
+    ephemeral_5m = getattr(cache_creation, "ephemeral_5m_input_tokens", None)
+    ephemeral_1h = getattr(cache_creation, "ephemeral_1h_input_tokens", None)
+
+    if ephemeral_5m is None and ephemeral_1h is None:
+        return None
+
+    breakdown: dict[str, float] = {}
+    if ephemeral_5m is not None and ephemeral_5m > 0:
+        breakdown["ephemeral5m"] = float(ephemeral_5m)
+    if ephemeral_1h is not None and ephemeral_1h > 0:
+        breakdown["ephemeral1h"] = float(ephemeral_1h)
+
+    return breakdown if breakdown else None
+
+
+def calculate_cost_sync(
+    provider_id: str,
+    model_id: str,
+    usage: Usage,
+) -> TokenCostCalculateResponse | None:
+    """Calculate cost synchronously. Returns None on any failure."""
+    normalized_model = _normalize_model_id(model_id)
+    via_router = _is_via_router(model_id)
+    logger.debug(
+        "Calculating cost for provider=%s, model=%s (normalized=%s), "
+        "via_router=%s, usage=(input=%s, output=%s, cache_read=%s, cache_write=%s)",
+        provider_id,
+        model_id,
+        normalized_model,
+        via_router,
+        usage.input_tokens,
+        usage.output_tokens,
+        usage.cache_read_tokens,
+        usage.cache_write_tokens,
+    )
+    try:
+        client = get_sync_client()
+        logger.debug("Got sync client, making API request")
+        cache_breakdown = extract_cache_write_breakdown(usage)
+        if cache_breakdown:
+            logger.debug("Cache write breakdown: %s", cache_breakdown)
+        # Build usage kwargs, only including optional fields if they have values
+        usage_kwargs: dict[str, Any] = {
+            "input_tokens": usage.input_tokens,
+            "output_tokens": usage.output_tokens,
+        }
+        if usage.cache_read_tokens:
+            usage_kwargs["cache_read_tokens"] = usage.cache_read_tokens
+        if usage.cache_write_tokens:
+            usage_kwargs["cache_write_tokens"] = usage.cache_write_tokens
+        if cache_breakdown:
+            usage_kwargs["cache_write_breakdown"] = cache_breakdown
+        result = client.token_cost.calculate(
+            provider=provider_id,
+            model=normalized_model,
+            usage=TokenCostCalculateRequestUsage(**usage_kwargs),
+            via_router=via_router,
+        )
+        logger.debug(
+            "Cost calculation result: input=%s, output=%s, total=%s centicents",
+            result.input_cost_centicents,
+            result.output_cost_centicents,
+            result.total_cost_centicents,
+        )
+        return result
+    except Exception:
+        logger.debug("Failed to calculate cost", exc_info=True)
+        return None
+
+
+async def calculate_cost_async(
+    provider_id: str,
+    model_id: str,
+    usage: Usage,
+) -> TokenCostCalculateResponse | None:
+    """Calculate cost asynchronously. Returns None on any failure."""
+    normalized_model = _normalize_model_id(model_id)
+    via_router = _is_via_router(model_id)
+    logger.debug(
+        "Calculating cost (async) for provider=%s, model=%s (normalized=%s), "
+        "via_router=%s, usage=(input=%s, output=%s, cache_read=%s, cache_write=%s)",
+        provider_id,
+        model_id,
+        normalized_model,
+        via_router,
+        usage.input_tokens,
+        usage.output_tokens,
+        usage.cache_read_tokens,
+        usage.cache_write_tokens,
+    )
+    try:
+        client = get_async_client()
+        logger.debug("Got async client, making API request")
+        cache_breakdown = extract_cache_write_breakdown(usage)
+        if cache_breakdown:
+            logger.debug("Cache write breakdown: %s", cache_breakdown)
+        # Build usage kwargs, only including optional fields if they have values
+        usage_kwargs: dict[str, Any] = {
+            "input_tokens": usage.input_tokens,
+            "output_tokens": usage.output_tokens,
+        }
+        if usage.cache_read_tokens:
+            usage_kwargs["cache_read_tokens"] = usage.cache_read_tokens
+        if usage.cache_write_tokens:
+            usage_kwargs["cache_write_tokens"] = usage.cache_write_tokens
+        if cache_breakdown:
+            usage_kwargs["cache_write_breakdown"] = cache_breakdown
+        result = await client.token_cost.calculate(
+            provider=provider_id,
+            model=normalized_model,
+            usage=TokenCostCalculateRequestUsage(**usage_kwargs),
+            via_router=via_router,
+        )
+        logger.debug(
+            "Cost calculation result (async): input=%s, output=%s, total=%s centicents",
+            result.input_cost_centicents,
+            result.output_cost_centicents,
+            result.total_cost_centicents,
+        )
+        return result
+    except Exception:
+        logger.debug("Failed to calculate cost (async)", exc_info=True)
+        return None

mirascope/ops/_internal/instrumentation/llm/encode.py

@@ -91,7 +91,7 @@ def _serialize_message_parts(
                 tool_output_part: gen_ai_types.ToolCallResponsePart = {
                     "type": "tool_call_response",
                     "id": part.id,
-                    "response": part.value,
+                    "response": part.result,
                 }
                 serialized.append(tool_output_part)
             case Thought():