Qwen2.5-Omni-7B-MNN模型运行问题

[HarryHang]

使用Model scope下载的 Qwen2.5-Omni-7B-MNN模型，在config.json里把后端改为cuda，在有两块4080的linux服务器上运行llm_demo，出现以下两个问题： 1、算子不支持 CUDABackend Don't support type Attention, Attention/Reshape_8_output_0 CUDABackend Don't support type Attention, Attention/Reshape_17_output_0 CUDABackend Don't support type Attention, Attention/Reshape_26_output_0 CUDABackend Don't support type Attention, Attention/Reshape_35_output_0 CUDABackend Don't support type Attention, Attention/Reshape_44_output_0 CUDABackend Don't support type Attention, Attention/Reshape_53_output_0 CUDABackend Don't support type Attention, Attention/Reshape_62_output_0 CUDABackend Don't support type Attention, Attention/Reshape_71_output_0 CUDABackend Don't support type Attention, Attention/Reshape_80_output_0 CUDABackend Don't support type Attention, Attention/Reshape_89_output_0 CUDABackend Don't support type Attention, Attention/Reshape_98_output_0 CUDABackend Don't support type Attention, Attention/Reshape_107_output_0 CUDABackend Don't support type Attention, Attention/Reshape_116_output_0 CUDABackend Don't support type Attention, Attention/Reshape_125_output_0 CUDABackend Don't support type Attention, Attention/Reshape_134_output_0 CUDABackend Don't support type Attention, Attention/Reshape_143_output_0 CUDABackend Don't support type Attention, Attention/Reshape_152_output_0 CUDABackend Don't support type Attention, Attention/Reshape_161_output_0 CUDABackend Don't support type Attention, Attention/Reshape_170_output_0 CUDABackend Don't support type Attention, Attention/Reshape_179_output_0 CUDABackend Don't support type Attention, Attention/Reshape_188_output_0 CUDABackend Don't support type Attention, Attention/Reshape_197_output_0 CUDABackend Don't support type Attention, Attention/Reshape_206_output_0 CUDABackend Don't support type Attention, Attention/Reshape_215_output_0 CUDABackend Don't support type Attention, Attention/Reshape_224_output_0 CUDABackend Don't support type Attention, Attention/Reshape_233_output_0 CUDABackend Don't support type Attention, Attention/Reshape_242_output_0 CUDABackend Don't support type Attention, Attention/Reshape_251_output_0 等等 2、显存不够 Out of memory when gamma is acquired in CudaLayerNorm. Out of memory when beta is acquired in CudaLayerNorm.

Out of memory when gamma is acquired in CudaLayerNorm. Out of memory when beta is acquired in CudaLayerNorm. Out of memory when gamma is acquired in CudaLayerNorm. Out of memory when beta is acquired in CudaLayerNorm. 等等。。。后面就段错误了 nvidia-smi观察，只有一张卡的显存增长。 请问第一个问题如何解决？ 第二个问题，有办法配置成不够的时候自动分配到另一张卡上吗？ 谢谢！