implement token counts (#5073)

src/phoenix/server/api/subscriptions.py

@@ -55,6 +55,7 @@
 )
 
 if TYPE_CHECKING:
+    from openai.types import CompletionUsage
     from openai.types.chat import (
         ChatCompletionMessageParam,
     )
@@ -161,6 +162,7 @@ async def chat_completion(
         self, info: Info[Context, None], input: ChatCompletionInput
     ) -> AsyncIterator[ChatCompletionChunk]:
         from openai import NOT_GIVEN, AsyncAzureOpenAI, AsyncOpenAI
+        from openai.types.chat import ChatCompletionStreamOptionsParam
 
         client: Union[AsyncAzureOpenAI, AsyncOpenAI]
 
@@ -208,14 +210,19 @@ async def chat_completion(
             text_chunks: List[TextChunk] = []
             tool_call_chunks: DefaultDict[ToolCallIndex, List[ToolCallChunk]] = defaultdict(list)
             role: Optional[str] = None
+            token_usage: Optional[CompletionUsage] = None
             async for chunk in await client.chat.completions.create(
                 messages=openai_messages,
                 model=input.model.name,
                 stream=True,
                 tools=input.tools or NOT_GIVEN,
+                stream_options=ChatCompletionStreamOptionsParam(include_usage=True),
                 **invocation_parameters,
             ):
                 response_chunks.append(chunk)
+                if (usage := chunk.usage) is not None:
+                    token_usage = usage
+                    continue
                 choice = chunk.choices[0]
                 delta = choice.delta
                 if role is None:
@@ -246,6 +253,7 @@ async def chat_completion(
                 dict(
                     chain(
                         _output_value_and_mime_type(response_chunks),
+                        _llm_token_counts(token_usage) if token_usage is not None else [],
                         _llm_output_messages(text_chunks, tool_call_chunks),
                     )
                 )
@@ -257,6 +265,8 @@ async def chat_completion(
         assert (attributes := finished_span.attributes) is not None
         start_time = _datetime(epoch_nanoseconds=finished_span.start_time)
         end_time = _datetime(epoch_nanoseconds=finished_span.end_time)
+        prompt_tokens = token_usage.prompt_tokens if token_usage is not None else 0
+        completion_tokens = token_usage.completion_tokens if token_usage is not None else 0
         trace_id = _hex(finished_span.context.trace_id)
         span_id = _hex(finished_span.context.span_id)
         status = finished_span.status
@@ -298,10 +308,10 @@ async def chat_completion(
                     status_code=status.status_code.name,
                     status_message=status.description or "",
                     cumulative_error_count=int(not status.is_ok),
-                    cumulative_llm_token_count_prompt=0,
-                    cumulative_llm_token_count_completion=0,
-                    llm_token_count_prompt=0,
-                    llm_token_count_completion=0,
+                    cumulative_llm_token_count_prompt=prompt_tokens,
+                    cumulative_llm_token_count_completion=completion_tokens,
+                    llm_token_count_prompt=prompt_tokens,
+                    llm_token_count_completion=completion_tokens,
                 )
             )
         info.context.event_queue.put(SpanInsertEvent(ids=(playground_project_id,)))
@@ -324,6 +334,12 @@ def _llm_tools(tools: List[JSONScalarType]) -> Iterator[Tuple[str, Any]]:
         yield f"{LLM_TOOLS}.{tool_index}.{TOOL_JSON_SCHEMA}", json.dumps(tool)
 
 
+def _llm_token_counts(usage: "CompletionUsage") -> Iterator[Tuple[str, Any]]:
+    yield LLM_TOKEN_COUNT_PROMPT, usage.prompt_tokens
+    yield LLM_TOKEN_COUNT_COMPLETION, usage.completion_tokens
+    yield LLM_TOKEN_COUNT_TOTAL, usage.total_tokens
+
+
 def _input_value_and_mime_type(input: ChatCompletionInput) -> Iterator[Tuple[str, Any]]:
     assert any(field.name == (api_key := "api_key") for field in fields(ChatCompletionInput))
     yield INPUT_MIME_TYPE, JSON
@@ -419,6 +435,9 @@ def _template_formatter(template_language: TemplateLanguage) -> TemplateFormatte
 LLM_MODEL_NAME = SpanAttributes.LLM_MODEL_NAME
 LLM_INVOCATION_PARAMETERS = SpanAttributes.LLM_INVOCATION_PARAMETERS
 LLM_TOOLS = SpanAttributes.LLM_TOOLS
+LLM_TOKEN_COUNT_PROMPT = SpanAttributes.LLM_TOKEN_COUNT_PROMPT
+LLM_TOKEN_COUNT_COMPLETION = SpanAttributes.LLM_TOKEN_COUNT_COMPLETION
+LLM_TOKEN_COUNT_TOTAL = SpanAttributes.LLM_TOKEN_COUNT_TOTAL
 
 MESSAGE_CONTENT = MessageAttributes.MESSAGE_CONTENT
 MESSAGE_ROLE = MessageAttributes.MESSAGE_ROLE