执行example.py train官方样例数据 加载完模型后无报错退出,未进入训练阶段

[SunYusheng]

aistudio环境, A100 40G单卡 模型使用 rocketqa/zh_dureader_de_v2 训练数据使用examples/data/dual.train.tsv 即: train_dual_encoder('zh_dureader_de_v2', './examples/data/dual.train.tsv') 执行到[ init.py: 74]: Load pretraining parameters from /home/aistudio/.rocketqa/zh_dureader_de_v2/dual_params. 后退出 ` [INFO] 2022-11-29 10:26:59,291 [ args.py: 71]: verbose: False INFO:root:vocab_path: /home/aistudio/.rocketqa/zh_dureader_de_v2/zh_vocab.txt [INFO] 2022-11-29 10:26:59,291 [ args.py: 71]: vocab_path: /home/aistudio/.rocketqa/zh_dureader_de_v2/zh_vocab.txt INFO:root:warmup_proportion: 0.1 [INFO] 2022-11-29 10:26:59,291 [ args.py: 71]: warmup_proportion: 0.1 INFO:root:weight_decay: 0.01 [INFO] 2022-11-29 10:26:59,291 [ args.py: 71]: weight_decay: 0.01 INFO:root:------------------------------------------------

INFO:root:Device count: 1 [INFO] 2022-11-29 10:26:59,330 [dual_encoder.py: 291]: Device count: 1 INFO:root:Num train examples: 112 [INFO] 2022-11-29 10:26:59,331 [dual_encoder.py: 292]: Num train examples: 112 INFO:root:Max train steps: 7 [INFO] 2022-11-29 10:26:59,331 [dual_encoder.py: 293]: Max train steps: 7 INFO:root:Num warmup steps: 0 [INFO] 2022-11-29 10:26:59,331 [dual_encoder.py: 294]: Num warmup steps: 0 INFO:root:Learning rate: 0.000010 [INFO] 2022-11-29 10:26:59,331 [dual_encoder.py: 295]: Learning rate: 0.000010 WARNING:root:paddle.fluid.layers.py_reader() may be deprecated in the near future. Please use paddle.fluid.io.DataLoader.from_generator() instead. [WARNING] 2022-11-29 10:26:59,331 [ io.py: 721]: paddle.fluid.layers.py_reader() may be deprecated in the near future. Please use paddle.fluid.io.DataLoader.from_generator() instead. INFO:rocketqa.utils.init:Load pretraining parameters from /home/aistudio/.rocketqa/zh_dureader_de_v2/dual_params. [INFO] 2022-11-29 10:27:05,570 [ init.py: 74]: Load pretraining parameters from /home/aistudio/.rocketqa/zh_dureader_de_v2/dual_params. `

[yslu-TW]

80

兄弟 你遇到的問題跟我一樣，其實是有訓練的只是沒有儲存model跟列出log而已，要去他library的程式碼做修改，可參考我的回覆。