@RunLLM vllm used to support both “served-model-name” and ““model”” as launch arguments, is there any change in recent versions of vLLM?

Yes, there is a change: in recent versions of vLLM, the recommended way to specify the model when launching the server is to provide the model as a positional argument to vllm serve (e.g., vllm serve ), rather than using the --model option. The --model option is deprecated and will be removed in v0.…

[image] RunLLM: e API-visible model name(s) and remains the correct way to set alternative model names for the API server. If both are provided, @RunLLM tell me which pr deprecated --model option

The PR that deprecated the --model option for vllm serve is PR #17944 (“[Misc] remove --model from vllm serve usage”) . This PR enforces using the model as a positional argument or via a config file, and raises an error if --model is used directly with vllm serve. Would you like more details on the …

@RunLLM is v1 engine in recent vllm versions working well with mig GPUs?

No, vLLM V1 engine does not officially support MIG (Multi-Instance GPU) configurations. The maintainers have stated that MIG is not a supported use case, and users are advised to “hide the heterogeneity through docker” as a workaround, but even this is not guaranteed to work reliably. There are know…

@RunLLM I think it worked fine with older version like v0.11?

"served-model-name" and "model"

General

hustxiayang January 26, 2026, 5:39pm 3

@RunLLM tell me which pr deprecated --model option

Topic		Replies	Views
How to use local model when using vllm serve? General	3	8457	July 22, 2025
Jina AI VLM - Support help General	3	91	October 12, 2025
Make install easier General	11	665	July 24, 2025
How to serve two vLLM instance using docker? General	3	597	February 26, 2026
torch.OutOfMemoryError: CUDA out of memory General	15	1713	March 3, 2026

"served-model-name" and "model"

Related topics