Fix model selection

[huydhn]

There is a bug in the script where I wrongly used models variable as both the input list of target models and to de-dup the list of selected models.

Testing

https://github.com/pytorch/pytorch-integration-testing/actions/runs/15940471919/job/44967831823

python .github/scripts/generate_vllm_benchmark_matrix.py --benchmark-configs-dir vllm-benchmarks/benchmarks --models "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8" --gpus h100

returns only one job now

::set-output name=benchmark_matrix::{'include': [{'runner': 'linux.aws.h100.8', 'models': 'meta-llama/llama-4-maverick-17b-128e-instruct-fp8'}]}

[yangw-dev]

LGTM!