Skip to content

Tongjilibo/torch4keras

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

torch4keras

licence GitHub release PyPI PyPI - Downloads GitHub stars GitHub Issues contributions welcome

Documentation | Bert4torch | Examples | Source code | build_MiniLLM_from_scratch

1. 下载安装

安装稳定版

pip install torch4keras

安装最新版

pip install git+https://github.com/Tongjilibo/torch4keras.git

2. 功能

  • 简述:抽象出来的Trainer,适用于一般神经网络的训练,仅需关注网络结构代码

  • 特色:进度条展示训练过程,自定义metric,自带Evaluator, Checkpoint, Tensorboard, Logger等Callback,也可自定义Callback

  • 初衷:前期功能是作为bert4torchrec4torch的Trainer

  • 训练:

    2022-10-28 23:16:10 - Start Training
    2022-10-28 23:16:10 - Epoch: 1/5
    5000/5000 [==============================] - 13s 3ms/step - loss: 0.1351 - acc: 0.9601
    Evaluate: 100%|██████████████████████████████████████████████████| 2500/2500 [00:03<00:00, 798.09it/s] 
    test_acc: 0.98045. best_test_acc: 0.98045
    
    2022-10-28 23:16:27 - Epoch: 2/5
    5000/5000 [==============================] - 13s 3ms/step - loss: 0.0465 - acc: 0.9862
    Evaluate: 100%|██████████████████████████████████████████████████| 2500/2500 [00:03<00:00, 635.78it/s] 
    test_acc: 0.98280. best_test_acc: 0.98280
    
    2022-10-28 23:16:44 - Epoch: 3/5
    5000/5000 [==============================] - 15s 3ms/step - loss: 0.0284 - acc: 0.9915
    Evaluate: 100%|██████████████████████████████████████████████████| 2500/2500 [00:03<00:00, 673.60it/s] 
    test_acc: 0.98365. best_test_acc: 0.98365
    
    2022-10-28 23:17:03 - Epoch: 4/5
    5000/5000 [==============================] - 15s 3ms/step - loss: 0.0179 - acc: 0.9948
    Evaluate: 100%|██████████████████████████████████████████████████| 2500/2500 [00:03<00:00, 692.34it/s] 
    test_acc: 0.98265. best_test_acc: 0.98365
    
    2022-10-28 23:17:21 - Epoch: 5/5
    5000/5000 [==============================] - 14s 3ms/step - loss: 0.0129 - acc: 0.9958
    Evaluate: 100%|██████████████████████████████████████████████████| 2500/2500 [00:03<00:00, 701.77it/s] 
    test_acc: 0.98585. best_test_acc: 0.98585
    
    2022-10-28 23:17:37 - Finish Training
    

3. 快速上手

4. 版本历史

更新日期 版本 版本说明
20240418 v0.2.2 增加YamlConfig和IniConfig, 优化deepspeed的使用,修复_prepare_inputs的bug,修复断点续训SmoothMetricsCallback起点错误的bug, Trainer断点续训记录batch数
20240317 v0.2.1.post2 训练异常时保存权重,避免空数据集error,默认指标使用滑窗平滑,BaseModelDDP修改的更易用,mapping允许是函数更易用,Checkpoint增加save_on_train_end,增加SystemCallback,修改run_callback=False的bug, 适配build_MiniLLM_from_scratch, 修复ddp中mix_precision和torch重名的bug
20240221 v0.2.0 fit中修改.train()逻辑较少耗时

更多版本

5. 更新历史:

更多历史