[Bug]: Using vllm072 to start deepseek-r1-awq generates an error, RuntimeError：Failed to infer device type

[hyyuananran]

Your current environment

The output of `python collect_env.py`

python -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --port 12345 --max-model-len 65536 --trust-remote-code --tensor-parallel-size 8 --quantization moe_wna16 --gpu-memory-utilization 0.97 --kv-cache-dtype fp8_e5m2 --calculate-kv-scales --served-model-name deepseek-reasoner --model cognitivecomputations/DeepSeek-R1-AWQ

Your output of `python collect_env.py` here

🐛 Describe the bug

python -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --port 12345 --max-model-len 65536 --trust-remote-code --tensor-parallel-size 8 --quantization moe_wna16 --gpu-memory-utilization 0.97 --kv-cache-dtype fp8_e5m2 --calculate-kv-scales --served-model-name deepseek-reasoner --model cognitivecomputations/DeepSeek-R1-AWQ

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[robertgshaw2-redhat]

We are aware of the issue. Upgrading pynvml to 12+ will fix for now. This fixes #12963 the issue permanently