dadyita

训练bge-reranker-large时，加入参数--gradient_checkpointing 后报错

1

命令行如下 #!/bin/sh CUDA_VISIBLE_DEVICES=3 \ torchrun --nproc_per_node 1 \ -m FlagEmbedding.reranker.run \ --output_dir /home/dayita/model/rerank/train3 \ --model_name_or_path /llms/models/bge-reranker-large/ \ --train_data /home/dayita/muyu/testForRerank2.jsonl \ --learning_rate 1e-5 \ --gradient_checkpointing \ --fp16 \ --num_train_epochs 5 \...