[NPU] Slow Token Generation with Latest NPU Driver 32.0.100.3053 on LNL 226V series

[climh]

Description

Observed less than < 1 token per second generation for model >7B parameters using 32.0.100.3053 driver with latest ipex-llm[npu] on LNL 226V series laptop.

Model tested:

1. Qwen/Qwen2.5-7B-Instruct
2. meta-llama/Llama-2-7b-hf
3. meta-llama/Meta-Llama-3-8B-Instruct

ipex-llm[npu] = 2.2.0b20241022
npu driver = 32.0.10.3053