agenta 0.25.4a1__tar.gz → 0.25.4a2__tar.gz

{agenta-0.25.4a1 → agenta-0.25.4a2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: agenta
-Version: 0.25.4a1
+Version: 0.25.4a2
 Summary: The SDK for agenta is an open-source LLMOps platform.
 Home-page: https://agenta.ai
 Keywords: LLMOps,LLM,evaluation,prompt engineering
@@ -14,6 +14,7 @@ Classifier: Programming Language :: Python :: 3.9
 Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Software Development :: Libraries
 Requires-Dist: cachetools (>=5.3.3,<6.0.0)
 Requires-Dist: click (>=8.1.3,<9.0.0)
@@ -22,6 +23,7 @@ Requires-Dist: fastapi (>=0.100.0)
 Requires-Dist: httpx (>=0.24,<0.28)
 Requires-Dist: importlib-metadata (>=8.0.0,<9.0)
 Requires-Dist: ipdb (>=0.13)
+Requires-Dist: litellm (>=1.48.0,<2.0.0)
 Requires-Dist: opentelemetry-api (>=1.27.0,<2.0.0)
 Requires-Dist: opentelemetry-exporter-otlp (>=1.27.0,<2.0.0)
 Requires-Dist: opentelemetry-sdk (>=1.27.0,<2.0.0)

{agenta-0.25.4a1 → agenta-0.25.4a2}/agenta/sdk/decorators/routing.py

@@ -656,8 +656,7 @@ class entrypoint:
 
         SHOW_DETAILS = True
         SHOW_DATA = False
-        SHOW_SPANS = True
-        SHOW_SPAN_ATTRIBUTES = False
+        SHOW_TRACE = False
 
         log.info("\n========= Result =========\n")
 
@@ -665,14 +664,14 @@ class entrypoint:
         if SHOW_DETAILS:
             log.info(f"latency:  {result.trace.get('latency')}")
             log.info(f"cost:     {result.trace.get('cost')}")
-            log.info(f"tokens:   {list(result.trace.get('tokens', {}).values())}")
+            log.info(f"usage:   {list(result.trace.get('usage', {}).values())}")
 
         if SHOW_DATA:
             log.info(" ")
             log.info(f"data:")
             log.info(json.dumps(result.data, indent=2))
 
-        if SHOW_SPANS:
+        if SHOW_TRACE:
             log.info(" ")
             log.info(f"trace:")
             log.info(f"----------------")

{agenta-0.25.4a1 → agenta-0.25.4a2}/agenta/sdk/litellm/litellm.py

@@ -1,5 +1,7 @@
 import agenta as ag
 
+from opentelemetry.trace import SpanKind
+
 from agenta.sdk.tracing.spans import CustomSpan
 from agenta.sdk.utils.exceptions import suppress
 from agenta.sdk.utils.logging import log
@@ -42,10 +44,10 @@ def litellm_handler():
                 else "embedding"
             )
 
-            kind = "CLIENT"
+            kind = SpanKind.CLIENT
 
             self.span = CustomSpan(
-                ag.tracer.start_span(name=f"litellm_{kind.lower()}", kind=kind)
+                ag.tracer.start_span(name=f"litellm_{kind.name.lower()}", kind=kind)
             )
 
             self.span.set_attributes(

{agenta-0.25.4a1 → agenta-0.25.4a2}/agenta/sdk/tracing/inline.py

@@ -999,7 +999,7 @@ def parse_to_agenta_span_dto(
 ########################################
 
 
-from copy import deepcopy
+from litellm import cost_calculator
 from opentelemetry.sdk.trace import ReadableSpan
 
 
@@ -1029,6 +1029,8 @@ def parse_inline_trace(
     ### services.observability.service.ingest/query() ###
     #####################################################
 
+    _calculate_cost(span_idx)
+
     ###############################################
     ### services.observability.service.ingest() ###
     ### --------------------------------------- ###
@@ -1083,7 +1085,7 @@ def parse_inline_trace(
             "trace_id": trace_id,
             "latency": latency,
             "cost": cost,
-            "tokens": tokens,
+            "usage": tokens,
             "spans": spans,
         }
 
@@ -1267,3 +1269,39 @@ def _parse_to_legacy_span(span: SpanDTO) -> CreateSpan:
     )
 
     return legacy_span
+
+
+PAYING_TYPES = [
+    "embedding",
+    "query",
+    "completion",
+    "chat",
+    "rerank",
+]
+
+
+def _calculate_cost(span_idx: Dict[str, SpanCreateDTO]):
+    for span in span_idx.values():
+        if span.node.type.name.lower() in PAYING_TYPES and span.meta and span.metrics:
+
+            try:
+                costs = cost_calculator.cost_per_token(
+                    model=span.meta.get("response.model"),
+                    prompt_tokens=span.metrics.get("unit.tokens.prompt", 0.0),
+                    completion_tokens=span.metrics.get("unit.tokens.completion", 0.0),
+                    call_type=span.node.type.name.lower(),
+                    response_time_ms=span.time.span // 1_000,
+                )
+
+                if not costs:
+                    continue
+
+                prompt_cost, completion_cost = costs
+                total_cost = prompt_cost + completion_cost
+
+                span.metrics["unit.costs.prompt"] = prompt_cost
+                span.metrics["unit.costs.completion"] = completion_cost
+                span.metrics["unit.costs.total"] = total_cost
+
+            except:
+                pass

{agenta-0.25.4a1 → agenta-0.25.4a2}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "agenta"
-version = "0.25.4a1"
+version = "0.25.4a2"
 description = "The SDK for agenta is an open-source LLMOps platform."
 readme = "README.md"
 authors = ["Mahmoud Mabrouk <mahmoud@agenta.ai>"]
@@ -36,6 +36,7 @@ pyyaml = "^6.0.2"
 opentelemetry-api = "^1.27.0"
 opentelemetry-sdk = "^1.27.0"
 opentelemetry-exporter-otlp = "^1.27.0"
+litellm ="^1.48.0"
 
 [tool.poetry.dev-dependencies]
 pytest = "^8.3"