deepeval 3.6.0__tar.gz → 3.6.1__tar.gz

{deepeval-3.6.0 → deepeval-3.6.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: deepeval
-Version: 3.6.0
+Version: 3.6.1
 Summary: The LLM Evaluation Framework
 Home-page: https://github.com/confident-ai/deepeval
 License: Apache-2.0

deepeval-3.6.1/deepeval/_version.py

@@ -0,0 +1 @@
+__version__: str = "3.6.1"

{deepeval-3.6.0 → deepeval-3.6.1}/deepeval/openai_agents/callback_handler.py

@@ -83,7 +83,9 @@ class DeepEvalTracingProcessor(TracingProcessor):
         if not span.started_at:
             return
         current_span = current_span_context.get()
-        if current_span and isinstance(current_span, LlmSpan):
+        if current_span and isinstance(
+            current_span, LlmSpan
+        ):  # llm span started by
             return
 
         span_type = self.get_span_kind(span.span_data)
@@ -101,10 +103,15 @@ class DeepEvalTracingProcessor(TracingProcessor):
             current_trace_context.get(), span.span_data
         )
 
+        span_type = self.get_span_kind(span.span_data)
         current_span = current_span_context.get()
-        if current_span and isinstance(current_span, LlmSpan):
+        if (
+            current_span
+            and isinstance(current_span, LlmSpan)
+            and span_type == "llm"
+        ):  # addtional check if the span kind data is llm too
             update_span_properties(current_span, span.span_data)
-            return
+
         observer = self.span_observers.pop(span.span_id, None)
         if observer:
             observer.__exit__(None, None, None)

{deepeval-3.6.0 → deepeval-3.6.1}/deepeval/tracing/otel/exporter.py

@@ -42,6 +42,7 @@ from deepeval.tracing.types import TraceAttributes
 from deepeval.test_case import ToolCall
 from dataclasses import dataclass
 import deepeval
+from deepeval.tracing.utils import make_json_serializable_for_metadata
 
 
 @dataclass
@@ -360,6 +361,8 @@ class ConfidentSpanExporter(SpanExporter):
             raw_trace_expected_tools
         )
         trace_metadata = self._parse_json_string(raw_trace_metadata)
+        if trace_metadata:
+            trace_metadata = make_json_serializable_for_metadata(trace_metadata)
         trace_metric_collection = parse_string(raw_trace_metric_collection)
 
         base_span_wrapper.trace_input = trace_input
@@ -426,6 +429,9 @@ class ConfidentSpanExporter(SpanExporter):
         span_tools_called = self._parse_list_of_tools(raw_span_tools_called)
         span_expected_tools = self._parse_list_of_tools(raw_span_expected_tools)
         span_metadata = self._parse_json_string(raw_span_metadata)
+        if span_metadata:
+            span_metadata = make_json_serializable_for_metadata(span_metadata)
+
         span_metric_collection = parse_string(raw_span_metric_collection)
 
         # Set Span Attributes

{deepeval-3.6.0 → deepeval-3.6.1}/deepeval/tracing/otel/utils.py

@@ -4,6 +4,8 @@ from deepeval.tracing import trace_manager, BaseSpan
 from opentelemetry.sdk.trace.export import ReadableSpan
 import json
 
+from deepeval.tracing.utils import make_json_serializable
+
 GEN_AI_OPERATION_NAMES = ["chat", "generate_content", "task_completion"]
 
 
@@ -103,10 +105,13 @@ def check_llm_input_from_gen_ai_attributes(
     output = None
     try:
         input = json.loads(span.attributes.get("gen_ai.input.messages"))
+        input = _flatten_input(input)
+
     except Exception as e:
         pass
     try:
         output = json.loads(span.attributes.get("gen_ai.output.messages"))
+        output = _flatten_input(output)
     except Exception as e:
         pass
 
@@ -127,6 +132,61 @@ def check_llm_input_from_gen_ai_attributes(
     return input, output
 
 
+def _flatten_input(input: list) -> list:
+    if input and isinstance(input, list):
+        try:
+            result: List[dict] = []
+            for m in input:
+                if isinstance(m, dict):
+                    role = m.get("role")
+                    if not role:
+                        role = "assistant"
+                    parts = m.get("parts")
+                    if parts:
+                        for part in parts:
+                            if isinstance(part, dict):
+                                ptype = part.get("type")
+                                if ptype == "text":
+                                    result.append(
+                                        {
+                                            "role": role,
+                                            "content": part.get("content"),
+                                        }
+                                    )
+                                else:
+                                    result.append(
+                                        {
+                                            "role": role,
+                                            "content": make_json_serializable(
+                                                part
+                                            ),
+                                        }
+                                    )
+                            else:
+                                result.append(
+                                    {
+                                        "role": role,
+                                        "content": make_json_serializable(part),
+                                    }
+                                )
+                    else:
+                        result.append(
+                            {"role": role, "content": m.get("content")}
+                        )  # no parts
+                else:
+                    result.append(
+                        {
+                            "role": "assistant",
+                            "content": make_json_serializable(m),
+                        }
+                    )
+            return result
+        except Exception as e:
+            return input
+
+    return input
+
+
 def check_tool_name_from_gen_ai_attributes(span: ReadableSpan) -> Optional[str]:
     try:
         gen_ai_tool_name = span.attributes.get("gen_ai.tool.name")
@@ -374,6 +434,8 @@ def check_pydantic_ai_agent_input_output(
     except Exception:
         pass
 
+    input_val = _flatten_input(input_val)
+    output_val = _flatten_input(output_val)
     return input_val, output_val
 
 

{deepeval-3.6.0 → deepeval-3.6.1}/deepeval/tracing/utils.py

@@ -100,6 +100,60 @@ def make_json_serializable(obj):
     return _serialize(obj)
 
 
+def make_json_serializable_for_metadata(obj):
+    """
+    Recursively converts an object to a JSON‐serializable form,
+    replacing circular references with "<circular>".
+    """
+    seen = set()  # Store `id` of objects we've visited
+
+    def _serialize(o):
+        oid = id(o)
+
+        # strip Nulls
+        if isinstance(o, str):
+            return _strip_nul(o)
+
+        # Primitive types are already serializable
+        if isinstance(o, (str, int, float, bool)) or o is None:
+            return str(o)
+
+        # Detect circular reference
+        if oid in seen:
+            return "<circular>"
+
+        # Mark current object as seen
+        seen.add(oid)
+
+        # Handle containers
+        if isinstance(o, (list, tuple, set, deque)):  # TODO: check if more
+            serialized = []
+            for item in o:
+                serialized.append(_serialize(item))
+
+            return serialized
+
+        if isinstance(o, dict):
+            result = {}
+            for key, value in o.items():
+                # Convert key to string (JSON only allows string keys)
+                result[str(key)] = _serialize(value)
+            return result
+
+        # Handle objects with __dict__
+        if hasattr(o, "__dict__"):
+            result = {}
+            for key, value in vars(o).items():
+                if not key.startswith("_"):
+                    result[key] = _serialize(value)
+            return result
+
+        # Fallback: convert to string
+        return _strip_nul(str(o))
+
+    return _serialize(obj)
+
+
 def to_zod_compatible_iso(
     dt: datetime, microsecond_precision: bool = False
 ) -> str:

{deepeval-3.6.0 → deepeval-3.6.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "deepeval"
-version = "3.6.0"
+version = "3.6.1"
 description = "The LLM Evaluation Framework"
 authors = ["Jeffrey Ip <jeffreyip@confident-ai.com>"]
 license = "Apache-2.0"

deepeval-3.6.0/deepeval/_version.py

@@ -1 +0,0 @@
-__version__: str = "3.6.0"