EASTERNTIGER

老师你好，使用mengzi-t5的模型就行就测试时发现一个问题，模型会有输出结果不全或者把原文内容没有完全输出的情况。 例如：error_sentence='不要去选一些冒进的或激进企业。',correct_sent='不要去选一些冒进的或激进的企。' error_sentence='当人们看到这幅作品时，就自然到某位摄影家。',correct_sent='当人们看到这幅作品时，就自然想到某位摄影。' error_sentence='很成年人会与父母共同居住生活。',correct_sent='很多成年人会与父母共同居住生。' 总会有这种明明可以纠正正确，但是结果输出不全的情况，请问这种情况该怎么解决？

RuntimeError: Can't enable access between nodes 1 and 0

1

Hi， I tried to convert T5 model to tensorrt. I have a 4 GPUs devices.In the python convert_checkpoint.py step,I set tp_size=4,pp_size=1.Then I got tensorrt model successfully.However,when I use command :mpirun...

How can I transfer t5 decoder_model_merged.onnx to tensorrt

2

Hi,when I try to use command: trtexec --onnx=decoder_model_merged.onnx --saveEngine=decoder_model_merged.trt in linux,it showed: [07/30/2024-02:00:24] [E] Error[4]: [graphShapeAnalyzer.cpp::analyzeShapes::2084] Error Code 4: Miscellaneous (IConditionalOutputLayer optimum::if_OutputLayer_4945: optimum::if_OutputLayer_4945: dimensions not compatible for if-conditional outputs) [07/30/2024-02:00:24]...