感谢回复, 但是如原帖所述, 从HalfTensor到IntTensor都尝试过. 都有问题.
设置后, 和conv3d算子有关的错误信息如下: CNML: 7.10.2 0a592c0 CNRT: 4.10.1 a884a9a [INFO][/pytorch/catch/torch_mlu/csrc/aten/operators/op_methods.cpp][line:32][init][thread:140028256171776][process:35236]: Using MLU device: 0 [INFO][/pytorch/catch/torch_mlu/csrc/aten/operators/op_methods.cpp][line:33][init][thread:140028256171776][process:35236]: Catch RunningMode: RunningMode::CNML_EAGER [DEBUG][/pytorch/catch/torch_mlu/csrc/aten/operators/cnml_ops.cpp][line:226][copy_][thread:140028256171776][process:35236]: self[shape: [1, 3, 16, 224, 224], device: mlu:0, dtype: Float] src[shape: [1, 3, 16, 224, 224], device: cpu, dtype: Float] non_blocking[false] [DEBUG][/pytorch/catch/torch_mlu/csrc/aten/operators/cnml_ops.cpp][line:226][copy_][thread:140028256171776][process:35236]: self[shape: [64, 3, 7, 7, 7], device: mlu:0, dtype: Float] src[shape: [64, 3, 7, 7, 7], device: cpu, dtype: Float] non_blocking[false] ... [8-10 19:5:54] [LOG_CNMLTRACE] [cnmlError] FIRSTCONV3D: check_layer: FirstConv3D 53: given input onchip datatype, input data type unsupport !! current input onchip datatype is (INT8, pos -5, scale 1.50331), input datatype is FLOAT32 [8-10 19:5:54] [LOG_MATCHER] [cnmlError] Launch kernel failed. [8-10 19:5:54] [LOG_MATCHER] [cnmlError] Layer group id: 60 layer type: CONV3D 2021-08-10 19:05:54.611169: [cnmlError] MATCHER: launchTpl: No match template! [ERROR][/pytorch/catch/torch_mlu/csrc/aten/operators/cnml/internal/conv_generic_internal.cpp][line:193][cnml_generic_conv_internal][thread:140028256171776][process:35236]: CNML error: CNML_STATUS_UNSUPPORT [WARNING][/pytorch/catch/torch_mlu/csrc/aten/operators/op_methods.cpp][line:1018][conv3d][thread:140028256171776][process:35236]: conv3d Op cannot run on MLU device, start running on CPU!
你可以设置环境变量 export TORCH_MIN_CNLOG_LEVEL=-1,看一下算子的输入输出信息展开
请登录后评论