lmnr 0.7.15__tar.gz → 0.7.17__tar.gz

{lmnr-0.7.15 → lmnr-0.7.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lmnr
-Version: 0.7.15
+Version: 0.7.17
 Summary: Python SDK for Laminar
 Author: lmnr.ai
 Author-email: lmnr.ai <founders@lmnr.ai>
@@ -26,6 +26,7 @@ Requires-Dist: grpcio>=1
 Requires-Dist: httpx>=0.24.0
 Requires-Dist: orjson>=3.0.0
 Requires-Dist: packaging>=22.0
+Requires-Dist: opentelemetry-instrumentation-threading>=0.57b0
 Requires-Dist: opentelemetry-instrumentation-alephalpha>=0.47.1 ; extra == 'alephalpha'
 Requires-Dist: opentelemetry-instrumentation-alephalpha>=0.47.1 ; extra == 'all'
 Requires-Dist: opentelemetry-instrumentation-bedrock>=0.47.1 ; extra == 'all'

{lmnr-0.7.15 → lmnr-0.7.17}/pyproject.toml

@@ -6,7 +6,7 @@
 
 [project]
 name = "lmnr"
-version = "0.7.15"
+version = "0.7.17"
 description = "Python SDK for Laminar"
 authors = [
   { name = "lmnr.ai", email = "founders@lmnr.ai" }
@@ -39,6 +39,7 @@ dependencies = [
   "httpx (>=0.24.0)",
   "orjson (>=3.0.0)",
   "packaging (>=22.0)",
+  "opentelemetry-instrumentation-threading>=0.57b0",
 ]
 # poetry would auto-generate these based on requires-python, but other build
 # systems don't, so we need to specify them manually.
@@ -145,4 +146,4 @@ build-backend = "uv_build"
 members = ["examples/fastapi-app"]
 
 [tool.ruff]
-target-version = "py310"
+target-version = "py310"

{lmnr-0.7.15 → lmnr-0.7.17}/src/lmnr/opentelemetry_lib/litellm/__init__.py

@@ -406,7 +406,15 @@ try:
                         details.get("cached_tokens"),
                     )
                 # TODO: add audio/image/text token details
-            # TODO: add completion tokens details (reasoning tokens)
+            if usage_dict.get("completion_tokens_details"):
+                details = usage_dict.get("completion_tokens_details", {})
+                details = model_as_dict(details)
+                if details.get("reasoning_tokens"):
+                    set_span_attribute(
+                        span,
+                        "gen_ai.usage.reasoning_tokens",
+                        details.get("reasoning_tokens"),
+                    )
 
         def _process_tool_calls(self, span, tool_calls, choice_index, is_response=True):
             """Process and set tool call attributes on the span"""
@@ -467,17 +475,56 @@ try:
                 content = message.get("content", "")
                 if content is None:
                     continue
+                reasoning_content = message.get("reasoning_content")
+                if reasoning_content:
+                    if isinstance(reasoning_content, str):
+                        reasoning_content = [
+                            {
+                                "type": "text",
+                                "text": reasoning_content,
+                            }
+                        ]
+                    elif not isinstance(reasoning_content, list):
+                        reasoning_content = [
+                            {
+                                "type": "text",
+                                "text": str(reasoning_content),
+                            }
+                        ]
+                else:
+                    reasoning_content = []
                 if isinstance(content, str):
-                    set_span_attribute(span, f"gen_ai.completion.{i}.content", content)
+                    if reasoning_content:
+                        set_span_attribute(
+                            span,
+                            f"gen_ai.completion.{i}.content",
+                            json.dumps(
+                                reasoning_content
+                                + [
+                                    {
+                                        "type": "text",
+                                        "text": content,
+                                    }
+                                ]
+                            ),
+                        )
+                    else:
+                        set_span_attribute(
+                            span,
+                            f"gen_ai.completion.{i}.content",
+                            content,
+                        )
                 elif isinstance(content, list):
                     set_span_attribute(
-                        span, f"gen_ai.completion.{i}.content", json.dumps(content)
+                        span,
+                        f"gen_ai.completion.{i}.content",
+                        json.dumps(reasoning_content + content),
                     )
                 else:
                     set_span_attribute(
                         span,
                         f"gen_ai.completion.{i}.content",
-                        json.dumps(model_as_dict(content)),
+                        json.dumps(reasoning_content + [model_as_dict(content)]),
                     )
 
         def _process_content_part(self, content_part: dict) -> dict:

{lmnr-0.7.15 → lmnr-0.7.17}/src/lmnr/opentelemetry_lib/opentelemetry/instrumentation/google_genai/__init__.py

@@ -21,6 +21,7 @@ from .schema_utils import SchemaJSONEncoder, process_schema
 from .utils import (
     dont_throw,
     get_content,
+    merge_text_parts,
     process_content_union,
     process_stream_chunk,
     role_from_content_union,
@@ -272,6 +273,16 @@ def _set_response_attributes(span, response: types.GenerateContentResponse):
 
     if response.usage_metadata:
         usage_dict = to_dict(response.usage_metadata)
+        candidates_token_count = usage_dict.get("candidates_token_count")
+        # unlike OpenAI, and unlike input cached tokens, thinking tokens are
+        # not counted as part of candidates token count, so we need to add them
+        # separately for consistency with other instrumentations
+        thoughts_token_count = usage_dict.get("thoughts_token_count")
+        output_token_count = (
+            (candidates_token_count or 0) + (thoughts_token_count or 0)
+            if candidates_token_count is not None or thoughts_token_count is not None
+            else None
+        )
         set_span_attribute(
             span,
             gen_ai_attributes.GEN_AI_USAGE_INPUT_TOKENS,
@@ -280,7 +291,7 @@ def _set_response_attributes(span, response: types.GenerateContentResponse):
         set_span_attribute(
             span,
             gen_ai_attributes.GEN_AI_USAGE_OUTPUT_TOKENS,
-            usage_dict.get("candidates_token_count"),
+            output_token_count,
         )
         set_span_attribute(
             span,
@@ -292,6 +303,11 @@ def _set_response_attributes(span, response: types.GenerateContentResponse):
             SpanAttributes.LLM_USAGE_CACHE_READ_INPUT_TOKENS,
             usage_dict.get("cached_content_token_count"),
         )
+        set_span_attribute(
+            span,
+            SpanAttributes.LLM_USAGE_REASONING_TOKENS,
+            thoughts_token_count,
+        )
 
     if should_send_prompts():
         set_span_attribute(
@@ -388,7 +404,7 @@ def _build_from_streaming_response(
             candidates=[
                 {
                     "content": {
-                        "parts": final_parts,
+                        "parts": merge_text_parts(final_parts),
                         "role": role,
                     },
                 }
@@ -438,7 +454,7 @@ async def _abuild_from_streaming_response(
             candidates=[
                 {
                     "content": {
-                        "parts": final_parts,
+                        "parts": merge_text_parts(final_parts),
                         "role": role,
                     },
                 }

{lmnr-0.7.15 → lmnr-0.7.17}/src/lmnr/opentelemetry_lib/opentelemetry/instrumentation/google_genai/utils.py

@@ -40,6 +40,56 @@ class ProcessChunkResult(TypedDict):
     model_version: str | None
 
 
+def merge_text_parts(
+    parts: list[types.PartDict | types.File | types.Part | str],
+) -> list[types.Part]:
+    if not parts:
+        return []
+
+    merged_parts: list[types.Part] = []
+    accumulated_text = ""
+
+    for part in parts:
+        # Handle string input - treat as text
+        if isinstance(part, str):
+            accumulated_text += part
+        # Handle File objects - they are not text, so don't merge
+        elif isinstance(part, types.File):
+            # Flush any accumulated text first
+            if accumulated_text:
+                merged_parts.append(types.Part(text=accumulated_text))
+                accumulated_text = ""
+            # Add the File as-is (wrapped in a Part if needed)
+            # Note: File objects should be passed through as-is in the original part
+            merged_parts.append(part)
+        # Handle Part and PartDict (dicts)
+        else:
+            part_dict = to_dict(part)
+
+            # Check if this is a text part
+            if part_dict.get("text") is not None:
+                accumulated_text += part_dict.get("text")
+            else:
+                # Non-text part (inline_data, function_call, etc.)
+                # Flush any accumulated text first
+                if accumulated_text:
+                    merged_parts.append(types.Part(text=accumulated_text))
+                    accumulated_text = ""
+
+                # Add the non-text part as-is
+                if isinstance(part, types.Part):
+                    merged_parts.append(part)
+                elif isinstance(part, dict):
+                    # Convert dict to Part object
+                    merged_parts.append(types.Part(**part_dict))
+
+    # Don't forget to add any remaining accumulated text
+    if accumulated_text:
+        merged_parts.append(types.Part(text=accumulated_text))
+
+    return merged_parts
+
+
 def set_span_attribute(span: Span, name: str, value: Any):
     if value is not None and value != "":
         span.set_attribute(name, value)

{lmnr-0.7.15 → lmnr-0.7.17}/src/lmnr/version.py

@@ -3,7 +3,7 @@ import httpx
 from packaging import version
 
 
-__version__ = "0.7.15"
+__version__ = "0.7.17"
 PYTHON_VERSION = f"{sys.version_info.major}.{sys.version_info.minor}"