Max_num_tokens fixed for qwen3-32b

[yuxin.yang]

Qwen3-32B | NVIDIA NGC

what is the reasoning behind keeping only 3 model profiles, each of them have max_num_tokens=5255, no Blackwell support. Qwen3-32b is not listed on the multi-LLM container (Supported Architectures for Multi-LLM NIM — NVIDIA NIM for Large Language Models (LLMs))