[ModelRunner] Fix stop and bad words list contiguous for offsets

[Marks101]

In our regression tests of the ModelRunner we noticed that in the current main branch (Jun 18, 2024) the stop_words_list feature does not work properly for batch_size > 1. The issue seems to be that the token arrays are not contiguously layed out in memory due to the transpose that is done in this line:

TensorRT-LLM/tensorrt_llm/runtime/generation.py

Line 104 in 2a115da

return np.array([flat_ids, offsets], dtype="int32").transpose((1, 0, 2))

This makes the array offsets that are created invalid.

In examples/run.py this features was deactivated for a long time, but it seems that originally the contiguous was implemented here:

TensorRT-LLM/examples/run.py

Line 403 in b777bd6

# stop_words_list = torch.Tensor(stop_words_list).to(torch.int32).to("cuda").contiguous()

Thanks for taking a look at this

[MartinMarciniszyn]

@Funatiq , could you please merge this into the main branch?

[nv-guomingz]

@Funatiq , could you please merge this into the main branch?

@byshiue already merged this PR into internal code base this morning.

[nv-guomingz]

@Marks101 thanks for your contribution to TRT-LLM, this MR has been merged into upstream now.