Remove total_tokens

Author: drewby

instrumentation-genai/opentelemetry-instrumentation-openai-v2/src/opentelemetry/instrumentation/openai_v2/patch.py

@@ -340,6 +340,7 @@ def _record_metrics(
     )
 
     if result and getattr(result, "usage", None):
+        # Always record input tokens
         input_attributes = {
             **common_attributes,
             GenAIAttributes.GEN_AI_TOKEN_TYPE: GenAIAttributes.GenAiTokenTypeValues.INPUT.value,
@@ -349,19 +350,18 @@ def _record_metrics(
             attributes=input_attributes,
         )
 
-        token_count = (
-            result.usage.total_tokens
-            if operation_name
-            == GenAIAttributes.GenAiOperationNameValues.EMBEDDINGS.value
-            else result.usage.completion_tokens
-        )
-        attributes = {
-            **common_attributes,
-            GenAIAttributes.GEN_AI_TOKEN_TYPE: GenAIAttributes.GenAiTokenTypeValues.COMPLETION.value,
-        }
-        instruments.token_usage_histogram.record(
-            token_count, attributes=attributes
-        )
+        # For embeddings, don't record output tokens as all tokens are input tokens
+        if (
+            operation_name
+            != GenAIAttributes.GenAiOperationNameValues.EMBEDDINGS.value
+        ):
+            output_attributes = {
+                **common_attributes,
+                GenAIAttributes.GEN_AI_TOKEN_TYPE: GenAIAttributes.GenAiTokenTypeValues.COMPLETION.value,
+            }
+            instruments.token_usage_histogram.record(
+                result.usage.completion_tokens, attributes=output_attributes
+            )
 
 
 def _set_response_attributes(
@@ -436,11 +436,7 @@ def _set_embeddings_response_attributes(
             GenAIAttributes.GEN_AI_USAGE_INPUT_TOKENS,
             result.usage.prompt_tokens,
         )
-        set_span_attribute(
-            span,
-            GenAIAttributes.GEN_AI_USAGE_OUTPUT_TOKENS,
-            result.usage.total_tokens,
-        )
+        # Don't set output tokens for embeddings as all tokens are input tokens
 
     # Emit events for embeddings if content capture is enabled
     if capture_content:

instrumentation-genai/opentelemetry-instrumentation-openai-v2/tests/test_async_embeddings.py

@@ -243,25 +243,18 @@ async def test_async_embeddings_token_metrics(
 
     # Find the input token data point
     input_token_point = None
-    output_token_point = None
     for point in token_metric.data.data_points:
         if (
             point.attributes[GenAIAttributes.GEN_AI_TOKEN_TYPE]
             == GenAIAttributes.GenAiTokenTypeValues.INPUT.value
         ):
             input_token_point = point
-        elif (
-            point.attributes[GenAIAttributes.GEN_AI_TOKEN_TYPE]
-            == GenAIAttributes.GenAiTokenTypeValues.COMPLETION.value
-        ):
-            output_token_point = point
+            break
 
     assert input_token_point is not None, "Input token metric not found"
-    assert output_token_point is not None, "Total token metric not found"
 
     # Verify the token counts match what was reported in the response
     assert input_token_point.sum == response.usage.prompt_tokens
-    assert output_token_point.sum == response.usage.total_tokens
 
 
 def assert_embedding_attributes(
@@ -277,7 +270,7 @@ def assert_embedding_attributes(
         response_id=None,  # Embeddings don't have a response ID
         response_model=response.model,
         input_tokens=response.usage.prompt_tokens,
-        output_tokens=response.usage.total_tokens,  # Use total_tokens for output_tokens
+        output_tokens=None,  # Embeddings don't have separate output tokens
         operation_name="embeddings",
         server_address="api.openai.com",
     )
@@ -297,10 +290,6 @@ def assert_embedding_attributes(
         span.attributes[GenAIAttributes.GEN_AI_USAGE_INPUT_TOKENS]
         == response.usage.prompt_tokens
     )
-    assert (
-        span.attributes[GenAIAttributes.GEN_AI_USAGE_OUTPUT_TOKENS]
-        == response.usage.total_tokens
-    )
 
 
 def assert_all_attributes(

instrumentation-genai/opentelemetry-instrumentation-openai-v2/tests/test_embeddings.py

@@ -376,25 +376,18 @@ def test_embeddings_token_metrics(
 
     # Find the input token data point
     input_token_point = None
-    output_token_point = None
     for point in token_metric.data.data_points:
         if (
             point.attributes[GenAIAttributes.GEN_AI_TOKEN_TYPE]
             == GenAIAttributes.GenAiTokenTypeValues.INPUT.value
         ):
             input_token_point = point
-        elif (
-            point.attributes[GenAIAttributes.GEN_AI_TOKEN_TYPE]
-            == GenAIAttributes.GenAiTokenTypeValues.COMPLETION.value
-        ):
-            output_token_point = point
+            break
 
     assert input_token_point is not None, "Input token metric not found"
-    assert output_token_point is not None, "Output token metric not found"
 
     # Verify the token counts match what was reported in the response
     assert input_token_point.sum == response.usage.prompt_tokens
-    assert output_token_point.sum == response.usage.total_tokens
 
 
 def assert_embedding_attributes(
@@ -410,7 +403,7 @@ def assert_embedding_attributes(
         response_id=None,  # Embeddings don't have a response ID
         response_model=response.model,
         input_tokens=response.usage.prompt_tokens,
-        output_tokens=response.usage.total_tokens,  # Use total_tokens for output_tokens
+        output_tokens=None,  # Embeddings don't have separate output tokens
         operation_name="embeddings",
         server_address="api.openai.com",
     )
@@ -430,10 +423,6 @@ def assert_embedding_attributes(
         span.attributes[GenAIAttributes.GEN_AI_USAGE_INPUT_TOKENS]
         == response.usage.prompt_tokens
     )
-    assert (
-        span.attributes[GenAIAttributes.GEN_AI_USAGE_OUTPUT_TOKENS]
-        == response.usage.total_tokens
-    )
 
 
 def assert_all_attributes(