Fix NoneType error of outputs

[QiuMike]

if --enable-log-outputs --enable-log-requests in cli, it will result an error of NoneType for none-stream requests

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Fix omni crash when start with --enable-log-outputs --enable-log-requests

Test Plan

python -m vllm_omni.entrypoints.cli.main serve /root/Jonathan1909/Ming-flash-omni-2.0/ --omni --port 8091 --allowed-local-media-path /root/vllm-omni --stage-configs-path vllm_omni/model_executor/stage_configs/ming_flash_omni.yaml --log-stats --enable-log-outputs --enable-log-requests

Test Result

(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] Chat completion failed: 'NoneType' object has no attribute 'outputs'
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] Traceback (most recent call last):
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] File "/root/vllm-omni/vllm_omni/entrypoints/openai/api_server.py", line 853, in create_chat_completion
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] generator = await handler.create_chat_completion(request, raw_request)
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] File "/root/vllm-omni/vllm_omni/entrypoints/openai/serving_chat.py", line 418, in create_chat_completion
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] return await self.chat_completion_full_generator(
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] File "/root/vllm-omni/vllm_omni/entrypoints/openai/serving_chat.py", line 1587, in chat_completion_full_generator
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] if choice.index < len(final_res.outputs):
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] AttributeError: 'NoneType' object has no attribute 'outputs'

final_res does not been initialized.

Essential Elements of an Effective PR Description Checklist

(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] Chat completion failed: 'NoneType' object has no attribute 'outputs'
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] Traceback (most recent call last):
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] File "/root/vllm-omni/vllm_omni/entrypoints/openai/api_server.py", line 853, in create_chat_completion
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] generator = await handler.create_chat_completion(request, raw_request)
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] File "/root/vllm-omni/vllm_omni/entrypoints/openai/serving_chat.py", line 418, in create_chat_completion
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] return await self.chat_completion_full_generator(
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] File "/root/vllm-omni/vllm_omni/entrypoints/openai/serving_chat.py", line 1587, in chat_completion_full_generator
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] if choice.index < len(final_res.outputs):
(APIServer pid=190855) ERROR 03-30 11:54:43 [api_server.py:855] AttributeError: 'NoneType' object has no attribute 'outputs'

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[hsliuustc0106]

@fake0fan PTAL

[david6666666]

Please add describtion and fix DCO

[QiuMike]

Please add describtion and fix DCO

@david6666666 Done

[fake0fan]

Good catch. LGTM.

But I’m just wondering if we should also add some tests for this path, so that this issue does not come back later. cc @Gaohan123 @yenuo26

[lishunyang12]

this fixes the text case but final_res is still None for audio/image-only responses. The logging block at L1588 (if choice.index < len(final_res.outputs)) will crash the same way when there's no text output.

Need to guard that access:

if final_res is not None and choice.index < len(final_res.outputs):

[lishunyang12]

Good — this was the missing assignment. But the same final_res is used in the logging block ~40 lines below without a None check, so audio/image-only requests still crash.

[QiuMike]

audio/image the output_text should be "", and will not crash, anything wrong?

[hsliuustc0106]

fix dco please

[QiuMike]

fix dco please

@hsliuustc0106 Done