How to use the qwen-1.5-0.5b model?

[zjj6084bty]

When using the qwen-1.5-0.5b model, is the program used still demo_qwen_npu.cpp? However, this program requires the prefill model to be INT8 and the decoding model to be q4k, while the models provided at https://huggingface.co/mllmTeam/qwen-1.5-0.5b-mllm/tree/main only include qwen-1.5-0.5b-fp32.mllm and qwen-1.5-0.5b-q4_k.mllm. The INT8 model is not provided. Is it necessary to perform quantization ourselves?
在使用qwen-1.5-0.5b模型的时候所采用的程序仍是demo_qwen_npu.cpp吗？但是这个程序中要求预填充模型是INT8，解码模型是q4k，而https://huggingface.co/mllmTeam/qwen-1.5-0.5b-mllm/tree/main所提供的模型只有qwen-1.5-0.5b-fp32.mllm和qwen-1.5-0.5b-q4_k.mllm，没有提供INT8模型，是需要自己量化吗？

[chenghuaWang]

现在 MLLM 支持了 prefill 和 decode 都在 NPU 上执行。请参考 https://ubiquitouslearning.github.io/mllm/qnn_backend/aot_execute.html 。