Token Usage Always Zero in Accumulation Stream Example from OpenAI Go SDK

[EchoHoo]

Hello OpenAI team,

I've been using the OpenAI Go SDK and specifically running the stream example provided at https://github.com/openai/openai-go/blob/main/examples/chat-completion-accumulating/main.go. However, I've encountered an issue where the token usage is always reported as zero, despite the example running correctly and generating outputs.

Steps to Reproduce:

1. Clone the OpenAI Go SDK from the main branch.
2. Navigate to the examples/chat-completion-accumulating directory.
3. Run the example using go run main.go.
4. Observe the outputs, specifically the token usage statistics.

Expected Behavior:
The token usage should accurately reflect the number of tokens used during the chat completions.

Actual Behavior:
The token usage is consistently reported as zero.

Output Details:
`
> Begin a very brief introduction of Greece, then incorporate the local weather of a few towns

{"role":"assistant","content":null}
{"tool_calls":[{"index":0,"id":"call_TqDP3Izo7KIyfFgmEzkkWZdn","type":"function","function":{"name":"get_live_weather","arguments":""}}]}
{"tool_calls":[{"index":0,"function":{"arguments":"{\"lo"}}]}
{"tool_calls":[{"index":0,"function":{"arguments":"catio"}}]}
{"tool_calls":[{"index":0,"function":{"arguments":"n\": \"A"}}]}
{"tool_calls":[{"index":0,"function":{"arguments":"then"}}]}
{"tool_calls":[{"index":0,"function":{"arguments":"s\"}"}}]}
Tool call stream finished: 0 get_live_weather {"location": "Athens"}

{"tool_calls":[{"index":1,"id":"call_eRCQGBbzI0eqWrFkxyE6HJaj","type":"function","function":{"name":"get_live_weather","arguments":""}}]}
{"tool_calls":[{"index":1,"function":{"arguments":"{\"lo"}}]}
{"tool_calls":[{"index":1,"function":{"arguments":"catio"}}]}
{"tool_calls":[{"index":1,"function":{"arguments":"n\": \"T"}}]}
{"tool_calls":[{"index":1,"function":{"arguments":"hess"}}]}
{"tool_calls":[{"index":1,"function":{"arguments":"aloni"}}]}
{"tool_calls":[{"index":1,"function":{"arguments":"ki\"}"}}]}
Tool call stream finished: 1 get_live_weather {"location": "Thessaloniki"}

{"tool_calls":[{"index":2,"id":"call_5wx6hceKR9z5gr3w4s0b9iE8","type":"function","function":{"name":"get_live_weather","arguments":""}}]}
{"tool_calls":[{"index":2,"function":{"arguments":"{\"lo"}}]}
{"tool_calls":[{"index":2,"function":{"arguments":"catio"}}]}
{"tool_calls":[{"index":2,"function":{"arguments":"n\": \"H"}}]}
{"tool_calls":[{"index":2,"function":{"arguments":"erak"}}]}
{"tool_calls":[{"index":2,"function":{"arguments":"lion\""}}]}
{"tool_calls":[{"index":2,"function":{"arguments":"}"}}]}
Tool call stream finished: 2 get_live_weather {"location": "Heraklion"}

{}
Total Tokens: 0
Finish Reason: tool_calls

`

Could you please look into this? It seems like either the token usage data is not being captured accurately, or there might be an issue with how it's being reported in the example.

Thank you for your assistance!

[afraidjpg]

Same problem, and I wrote a test for this:

func Test_OpenAI(t *testing.T) {
	secret := "api-secret"
	client := openai.NewClient(
		option.WithAPIKey(secret),
	)

	var message []openai.ChatCompletionMessageParamUnion
	message = append(message, openai.UserMessage("hello"))

	ctx := context.Background()
	params := openai.ChatCompletionNewParams{
		Messages: openai.F(message),
		Model:    openai.F(openai.ChatModelGPT3_5Turbo),
		StreamOptions: openai.F(openai.ChatCompletionStreamOptionsParam{
			IncludeUsage: openai.F(true),
		}),
	}

	stream := client.Chat.Completions.NewStreaming(
		ctx,
		params,
	)

	// optionally, an accumulator helper can be used
	acc := &openai.ChatCompletionAccumulator{}

	for stream.Next() {
		chunk := stream.Current()
		acc.AddChunk(chunk)

		if _, ok := acc.JustFinishedContent(); ok {
			break
		}
		// if using tool calls
		if _, ok := acc.JustFinishedToolCall(); ok {
			acc.Usage = chunk.Usage
			break
		}

		if _, ok := acc.JustFinishedRefusal(); ok {
			acc.Usage = chunk.Usage
			break
		}

		// it's best to use chunks after handling JustFinished events
		if len(chunk.Choices) > 0 {
			// ... write to SSE response

		}
	}

	if err := stream.Err(); err != nil {
		t.Logf("stream error: %v", err)
	}

	b, _ := json.MarshalIndent(v, "", "  ")
	fmt.Println(string(b))
}

with resposne:

{
  "id": "xxx",
  "choices": [
    {
      "finish_reason": "stop",
      "index": 0,
      "logprobs": {
        "content": null,
        "refusal": null
      },
      "message": {
        "role": "assistant",
        "content": "Hello! How can I assist you today?"
      }
    }
  ],
  "created": 1736786016,
  "model": "gpt-3.5-turbo-0125",
  "object": "chat.completion.chunk",
  "service_tier": "",
  "system_fingerprint": "fp_808245b034",
  "usage": {
    "completion_tokens": 0,
    "prompt_tokens": 0,
    "total_tokens": 0,
    "completion_tokens_details": {
      "accepted_prediction_tokens": 0,
      "audio_tokens": 0,
      "reasoning_tokens": 0,
      "rejected_prediction_tokens": 0
    },
    "prompt_tokens_details": {
      "audio_tokens": 0,
      "cached_tokens": 0
    }
  }
}

[kujtimiihoxha]

@jacobzim-stl this is marked as completed but I am still seeing this issue, is there a workaround for this ?

[kujtimiihoxha]

Ok found the solution https://github.com/openai/openai-[go/pull/125/files#diff-f4664a715018362d741cda39d5d652d2822b5c7a97b7f42b59c1871eb6a96a63R49](https://www.golinks.io/pull/125/files#diff-f4664a715018362d741cda39d5d652d2822b5c7a97b7f42b59c1871eb6a96a63R49?trackSource=github)