Investigate TRTLLM runtime repetitive issue

[Fridah-nv]

For models listed below, trtllm runtime generates repetitive outputs and less coherent than demollm's outputs:

EleutherAI/pythia-6.9b
HuggingFaceTB/SmolVLM2-2.2B-Instruct
allenai/OLMo-2-1124-7B-SFT
deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
meta-llama/CodeLlama-7b-Python-hf
meta-llama/Llama-3.2-1B-Instruct
mistralai/Mistral-Large-Instruct-2407
mistralai/Mistral-Nemo-Instruct-2407
nvidia/Llama-3.1-Minitron-4B-Width-Base

Let's take a closer look to see if there's any misconfiguration with trtllm runtime.
This analysis is based on Jun 1st dashboard run. View the coverage google sheet Tab 6/1/2025 for specific outputs.

[lucaslie]

NOTE: we will require to expose mixed_sampler in the build_and_run_ad.py script so we can correctly enable this option for models as needed.

By default, the ADConfig/TorchConfig will set it to False

[Fridah-nv]

With nv-auto-deploy#66, we can set mixed_sampler with --args.mixed-sampler True.
TODO: update the dashboard with mixed_sampler=True for the models listed above.