fix multi-GPU validation to include pipeline parallelism

[nightflight-dk]

This PR updates the validation logic in TritonPythonModel to correctly identify multi-GPU configurations that use Pipeline Parallelism (PP).

Previously, the check only considered Tensor Parallelism (TP) size (tensor_parallel_size > 1) when verifying if a model configured with KIND_GPU (intended for single-GPU) was attempting to run in a multi-GPU setup. This meant that models using Pipeline Parallelism (where pipeline_parallel_size > 1) could potentially bypass this check if tensor_parallel_size was 1.

Changes:

Retrieve pipeline_parallel_size from vllm_engine_config (defaulting to 1).
Update the validation condition to check if the total parallelism (tp_size * pp_size) exceeds 1.
Ensures that users are correctly prompted to use KIND_MODEL for any multi-GPU configuration, whether it involves TP, PP, or both.
Testing:

Verified that setting pipeline_parallel_size > 1 with
KIND_GPU now raises the expected ValueError.

[nightflight-dk]

@yinggeh @mc-nv pls review, thanks

[yinggeh]

LGTM

[yinggeh]

Can you also update the copyright to 2026 in the first line?

[nightflight-dk]

@yinggeh done, ready to merge

[yinggeh]

@nightflight-dk You should only update copyrights of the file you edited, which is src/model.py.

[nightflight-dk]

@nightflight-dk You should only update copyrights of the file you edited, which is src/model.py.
done

[yinggeh]

Thanks for contributing. This change will be included in Triton 26.01 release.