使用提供的例子进行训练无法输出模型

[wenlincheng]

日志如下

E:\IdeaProjects\knowledge-model\rocketqa_es>python example.py
RocketQA model [zh_dureader_de]
WARNING:root:paddle.fluid.layers.py_reader() may be deprecated in the near future. Please use paddle.fluid.io.DataLoader.from_generator() instead.
W1222 14:00:13.174715  6936 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 7.5, Driver API Version: 11.7, Runtime API Version: 11.7
W1222 14:00:13.178715  6936 gpu_resources.cc:91] device: 0, cuDNN Version: 8.4.
Load model done
INFO:root:-----------  Configuration Arguments -----------
[INFO] 2022-12-22 14:00:16,089 [     args.py:   69]:    -----------  Configuration Arguments -----------
INFO:root:batch_size: 8
[INFO] 2022-12-22 14:00:16,089 [     args.py:   71]:    batch_size: 8
INFO:root:checkpoints: checkpoints
[INFO] 2022-12-22 14:00:16,090 [     args.py:   71]:    checkpoints: checkpoints
INFO:root:chunk_scheme: IOB
[INFO] 2022-12-22 14:00:16,090 [     args.py:   71]:    chunk_scheme: IOB
INFO:root:decr_every_n_nan_or_inf: 2
[INFO] 2022-12-22 14:00:16,090 [     args.py:   71]:    decr_every_n_nan_or_inf: 2
INFO:root:decr_ratio: 0.8
[INFO] 2022-12-22 14:00:16,091 [     args.py:   71]:    decr_ratio: 0.8
INFO:root:dev_set: None
[INFO] 2022-12-22 14:00:16,091 [     args.py:   71]:    dev_set: None
INFO:root:diagnostic: None
[INFO] 2022-12-22 14:00:16,092 [     args.py:   71]:    diagnostic: None
INFO:root:diagnostic_save: None
[INFO] 2022-12-22 14:00:16,093 [     args.py:   71]:    diagnostic_save: None
INFO:root:do_lower_case: True
[INFO] 2022-12-22 14:00:16,093 [     args.py:   71]:    do_lower_case: True
INFO:root:do_test: True
[INFO] 2022-12-22 14:00:16,094 [     args.py:   71]:    do_test: True
INFO:root:do_train: False
[INFO] 2022-12-22 14:00:16,095 [     args.py:   71]:    do_train: False
INFO:root:do_val: False
[INFO] 2022-12-22 14:00:16,096 [     args.py:   71]:    do_val: False
INFO:root:doc_stride: 128
[INFO] 2022-12-22 14:00:16,096 [     args.py:   71]:    doc_stride: 128
INFO:root:enable_ce: False
[INFO] 2022-12-22 14:00:16,097 [     args.py:   71]:    enable_ce: False
INFO:root:epoch: 2
[INFO] 2022-12-22 14:00:16,098 [     args.py:   71]:    epoch: 2
INFO:root:ernie_config_path: C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/zh_config.json
[INFO] 2022-12-22 14:00:16,098 [     args.py:   71]:    ernie_config_path: C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/zh_config.json
INFO:root:for_cn: True
[INFO] 2022-12-22 14:00:16,099 [     args.py:   71]:    for_cn: True
INFO:root:in_tokens: False
[INFO] 2022-12-22 14:00:16,100 [     args.py:   71]:    in_tokens: False
INFO:root:incr_every_n_steps: 100
[INFO] 2022-12-22 14:00:16,101 [     args.py:   71]:    incr_every_n_steps: 100
INFO:root:incr_ratio: 2.0
[INFO] 2022-12-22 14:00:16,102 [     args.py:   71]:    incr_ratio: 2.0
INFO:root:init_checkpoint: C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/dureader_dual_encoder
[INFO] 2022-12-22 14:00:16,103 [     args.py:   71]:    init_checkpoint: C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/dureader_dual_encoder
INFO:root:init_loss_scaling: 102400
[INFO] 2022-12-22 14:00:16,104 [     args.py:   71]:    init_loss_scaling: 102400
INFO:root:init_pretraining_params: None
[INFO] 2022-12-22 14:00:16,105 [     args.py:   71]:    init_pretraining_params: None
INFO:root:is_classify: True
[INFO] 2022-12-22 14:00:16,108 [     args.py:   71]:    is_classify: True
INFO:root:is_distributed: False
[INFO] 2022-12-22 14:00:16,108 [     args.py:   71]:    is_distributed: False
INFO:root:is_regression: False
[INFO] 2022-12-22 14:00:16,109 [     args.py:   71]:    is_regression: False
INFO:root:label_map_config: None
[INFO] 2022-12-22 14:00:16,110 [     args.py:   71]:    label_map_config: None
INFO:root:learning_rate: 1e-05
[INFO] 2022-12-22 14:00:16,110 [     args.py:   71]:    learning_rate: 1e-05
INFO:root:log_folder: de_log
[INFO] 2022-12-22 14:00:16,111 [     args.py:   71]:    log_folder: de_log
INFO:root:lr_scheduler: linear_warmup_decay
[INFO] 2022-12-22 14:00:16,112 [     args.py:   71]:    lr_scheduler: linear_warmup_decay
INFO:root:max_answer_length: 100
[INFO] 2022-12-22 14:00:16,112 [     args.py:   71]:    max_answer_length: 100
INFO:root:max_query_length: 64
[INFO] 2022-12-22 14:00:16,113 [     args.py:   71]:    max_query_length: 64
INFO:root:max_seq_len: 512
[INFO] 2022-12-22 14:00:16,114 [     args.py:   71]:    max_seq_len: 512
INFO:root:metric: simple_accuracy
[INFO] 2022-12-22 14:00:16,115 [     args.py:   71]:    metric: simple_accuracy
INFO:root:metrics: True
[INFO] 2022-12-22 14:00:16,115 [     args.py:   71]:    metrics: True
INFO:root:model_name: zh_dureader_de
[INFO] 2022-12-22 14:00:16,116 [     args.py:   71]:    model_name: zh_dureader_de
INFO:root:n_best_size: 20
[INFO] 2022-12-22 14:00:16,116 [     args.py:   71]:    n_best_size: 20
INFO:root:num_iteration_per_drop_scope: 10
[INFO] 2022-12-22 14:00:16,117 [     args.py:   71]:    num_iteration_per_drop_scope: 10
INFO:root:num_labels: 2
[INFO] 2022-12-22 14:00:16,118 [     args.py:   71]:    num_labels: 2
INFO:root:output_file_name: None
[INFO] 2022-12-22 14:00:16,119 [     args.py:   71]:    output_file_name: None
INFO:root:output_item: 3
[INFO] 2022-12-22 14:00:16,120 [     args.py:   71]:    output_item: 3
INFO:root:p_max_seq_len: 384
[INFO] 2022-12-22 14:00:16,120 [     args.py:   71]:    p_max_seq_len: 384
INFO:root:predict_batch_size: None
[INFO] 2022-12-22 14:00:16,123 [     args.py:   71]:    predict_batch_size: None
INFO:root:q_max_seq_len: 32
[INFO] 2022-12-22 14:00:16,123 [     args.py:   71]:    q_max_seq_len: 32
INFO:root:random_seed: None
[INFO] 2022-12-22 14:00:16,124 [     args.py:   71]:    random_seed: None
INFO:root:save_model_path: de_models
[INFO] 2022-12-22 14:00:16,124 [     args.py:   71]:    save_model_path: de_models
INFO:root:save_steps: 10
[INFO] 2022-12-22 14:00:16,125 [     args.py:   71]:    save_steps: 10
INFO:root:share_parameter: 0
[INFO] 2022-12-22 14:00:16,126 [     args.py:   71]:    share_parameter: 0
INFO:root:shuffle: True
[INFO] 2022-12-22 14:00:16,126 [     args.py:   71]:    shuffle: True
INFO:root:skip_steps: 100
[INFO] 2022-12-22 14:00:16,127 [     args.py:   71]:    skip_steps: 100
INFO:root:task_id: 0
[INFO] 2022-12-22 14:00:16,128 [     args.py:   71]:    task_id: 0
INFO:root:test_data_cnt: 1110000
[INFO] 2022-12-22 14:00:16,129 [     args.py:   71]:    test_data_cnt: 1110000
INFO:root:test_save: ./checkpoints/test_result
[INFO] 2022-12-22 14:00:16,130 [     args.py:   71]:    test_save: ./checkpoints/test_result
INFO:root:test_set: None
[INFO] 2022-12-22 14:00:16,131 [     args.py:   71]:    test_set: None
INFO:root:tokenizer: FullTokenizer
[INFO] 2022-12-22 14:00:16,131 [     args.py:   71]:    tokenizer: FullTokenizer
INFO:root:train_data_size: 0
[INFO] 2022-12-22 14:00:16,132 [     args.py:   71]:    train_data_size: 0
INFO:root:train_set: ./data/dual.train.tsv
[INFO] 2022-12-22 14:00:16,133 [     args.py:   71]:    train_set: ./data/dual.train.tsv
INFO:root:use_cross_batch: False
[INFO] 2022-12-22 14:00:16,134 [     args.py:   71]:    use_cross_batch: False
INFO:root:use_cuda: True
[INFO] 2022-12-22 14:00:16,135 [     args.py:   71]:    use_cuda: True
INFO:root:use_dynamic_loss_scaling: True
[INFO] 2022-12-22 14:00:16,136 [     args.py:   71]:    use_dynamic_loss_scaling: True
INFO:root:use_fast_executor: True
[INFO] 2022-12-22 14:00:16,139 [     args.py:   71]:    use_fast_executor: True
INFO:root:use_lamb: False
[INFO] 2022-12-22 14:00:16,140 [     args.py:   71]:    use_lamb: False
INFO:root:use_mix_precision: False
[INFO] 2022-12-22 14:00:16,141 [     args.py:   71]:    use_mix_precision: False
INFO:root:use_multi_gpu_test: False
[INFO] 2022-12-22 14:00:16,142 [     args.py:   71]:    use_multi_gpu_test: False
INFO:root:use_recompute: False
[INFO] 2022-12-22 14:00:16,143 [     args.py:   71]:    use_recompute: False
INFO:root:validation_steps: 1000
[INFO] 2022-12-22 14:00:16,143 [     args.py:   71]:    validation_steps: 1000
INFO:root:verbose: False
[INFO] 2022-12-22 14:00:16,144 [     args.py:   71]:    verbose: False
INFO:root:vocab_path: C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/zh_vocab.txt
[INFO] 2022-12-22 14:00:16,145 [     args.py:   71]:    vocab_path: C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/zh_vocab.txt
INFO:root:warmup_proportion: 0.1
[INFO] 2022-12-22 14:00:16,146 [     args.py:   71]:    warmup_proportion: 0.1
INFO:root:weight_decay: 0.01
[INFO] 2022-12-22 14:00:16,146 [     args.py:   71]:    weight_decay: 0.01
INFO:root:------------------------------------------------
[INFO] 2022-12-22 14:00:16,147 [     args.py:   72]:    ------------------------------------------------
INFO:root:Device count: 1
[INFO] 2022-12-22 14:00:16,165 [dual_encoder.py:  291]: Device count: 1
INFO:root:Num train examples: 112
[INFO] 2022-12-22 14:00:16,166 [dual_encoder.py:  292]: Num train examples: 112
INFO:root:Max train steps: 28
[INFO] 2022-12-22 14:00:16,167 [dual_encoder.py:  293]: Max train steps: 28
INFO:root:Num warmup steps: 2
[INFO] 2022-12-22 14:00:16,168 [dual_encoder.py:  294]: Num warmup steps: 2
INFO:root:Learning rate: 0.000010
[INFO] 2022-12-22 14:00:16,170 [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] 2022-12-22 14:00:16,171 [       io.py:  719]: 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 C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/dureader_dual_encoder.
[INFO] 2022-12-22 14:00:26,350 [     init.py:   73]:    Load pretraining parameters from C:\Users\lincheng.wen/.rocketqa/zh_dureader_de/dureader_dual_encoder.

[yslu-TW]

#80
應該跟我遇到同個問題,要去調整程式,可參考我的回覆

[kaihe]

每1000 step 存储一次模型，例子中的训练数据量太少，每等到1000 step就结束了，所以没有输出。