Paul

albert_tiny 使用自己的数据 run_pretraining.py 问题

2

albert_tiny 使用自己的数据 run_pretraining.py 做fine-tune，然后下游任务训练出的模型，f1提升4%，但预测是时间为原来的三倍,是哪里的参数不对，还是啥原因？大神

### Description ```shell In file included from /data/mt/hbl/FasterTransformer_v5_tf/src/fastertransformer/tf_op/bert/BertOp.cc:18: /data/mt/hbl/FasterTransformer_v5_tf/src/fastertransformer/tf_op/BaseOp.h: In member function ‘fastertransformer::Tensor BaseOp::convert_tensor(tensorflow::Tensor)’: /data/mt/hbl/FasterTransformer_v5_tf/src/fastertransformer/tf_op/BaseOp.h:105:36: error: ‘bfloat16’ is not a member of ‘Eigen’ 105 | if (std::is_same::value == true) {...

mT5 with fp16 RuntimeError: [FT][ERROR] CUDA runtime error: an illegal memory access was encountered FasterTransformer_v5.1/src/fastertransformer/utils/memory_utils.cu:96

5

### Description ```shell branch: v5.1 docker_image: nvidia/pytorch:21.11-py3 gpu: T4 Error: Traceback (most recent call last): File "../examples/pytorch/t5/summarization.py", line 382, in main() File "../examples/pytorch/t5/summarization.py", line 289, in main summary_ft, _ =...