您好，我在编译一个encoder模型，他的yaml如下
# encoder config
calibration_parameters:
cal_data_dir: ./calibration_data_dir_cup/encoder_inputs/input_ids;./calibration_data_dir_cup/encoder_inputs/attn_mask_2d;./calibration_data_dir_cup/encoder_inputs/token_type_ids;./calibration_data_dir_cup/encoder_inputs/imgs;./calibration_data_dir_cup/encoder_inputs/depths;./calibration_data_dir_cup/encoder_inputs/pixels;./calibration_data_dir_cup/encoder_inputs/projection_mat_inv;./calibration_data_dir_cup/encoder_inputs/hist_robot_state;./calibration_data_dir_cup/encoder_inputs/joint_scale_shift;./calibration_data_dir_cup/encoder_inputs/joint_mask;./calibration_data_dir_cup/encoder_inputs/joint_relative_pos
quant_config: {
"model_config": {
"all_node_type": "int16"
}
}
compiler_parameters:
advice: 0
balance_factor: 0
cache_mode: enable
cache_path: 'cache'
compile_mode: latency
core_num: 1
jobs: 64
max_time_per_fc: 0
optimize_level: O2
input_parameters:
input_layout_train: NCHW;NCHW;NCHW;NCHW;NCHW;NCHW;NCHW;NCHW;NCHW;NCHW;NCHW
input_name: input_ids;attn_mask_2d;token_type_ids;imgs;depths;pixels;projection_mat_inv;hist_robot_state;joint_scale_shift;joint_mask;joint_relative_pos
input_shape: 1x14;1x14;1x14;1x4x3x256x320;1x4x1x256x320;400x1x2;1x4x4x4;1x1x14x8;1x14x2;1x14;1x14x14
input_type_rt: featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap
input_type_train: featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap;featuremap
mean_value: ''
scale_value: ''
separate_batch: false
std_value: ''
model_parameters:
debug_mode: ''
march: nash-e
onnx_model: test_onnx_model_cup/full_encoder_e2e1.onnx
output_model_file_prefix: encoder_opt
output_nodes: ''
remove_node_name: ''
remove_node_type: Quantize;Dequantize;Cast;Softmax
# remove_node_type: Quantize;Dequantize
working_dir: ./model_output_calib_encoder_cup1
在3.2.0的OE和gpu docker下正确，但是3.7.0的OE和cpu docker报错
麻烦看下是什么情况 十分感谢，由于开发机只支持到cuda12.2,3.7.0的GPU Docker需要12.4以上的cuda，所以只用了cpu,有最新支持cuda12.2的gpu docker等资料也麻烦分享下，十分感谢