# Pytorch❤️Keras
English | [简体中文](README.md)
The torchkeras library is a simple tool for training neural network in pytorch jusk in a keras style. 😋😋
## 1, Introduction
With torchkeras, You need not to write your training loop with many lines of code, all you need to do is just
like these two steps as below:
(i) create your network and wrap it and the loss_fn together with torchkeras.KerasModel like this:
`model = torchkeras.KerasModel(net,loss_fn=nn.BCEWithLogitsLoss())`.
(ii) fit your model with the training data and validate data.
The main code of use torchkeras is like below.
```python
import torch
import torchkeras
model = torchkeras.KerasModel(net,
loss_fn = nn.BCEWithLogitsLoss(),
optimizer= torch.optim.Adam(net.parameters(),lr = 0.001),
metrics_dict = {"acc":torchmetrics.Accuracy(task='binary')}
)
dfhistory=model.fit(train_data=dl_train,
val_data=dl_val,
epochs=20,
patience=3,
ckpt_path='checkpoint',
monitor="val_acc",
mode="max",
plot=True
)
```
Besides，You can use torchkeras.VLog to get the dynamic training visualization any where as you like ~
```python
import time
import math,random
from torchkeras import VLog
epochs = 10
batchs = 30
#0, init vlog
vlog = VLog(epochs, monitor_metric='val_loss', monitor_mode='min')
#1, log_start
vlog.log_start()
for epoch in range(epochs):
#train
for step in range(batchs):
#2, log_step (for training step)
vlog.log_step({'train_loss':100-2.5*epoch+math.sin(2*step/batchs)})
time.sleep(0.05)
#eval
for step in range(20):
#3, log_step (for eval step)
vlog.log_step({'val_loss':100-2*epoch+math.sin(2*step/batchs)},training=False)
time.sleep(0.05)
#4, log_epoch
vlog.log_epoch({'val_loss':100 - 2*epoch+2*random.random()-1,
'train_loss':100-2.5*epoch+2*random.random()-1})
# 5, log_end
vlog.log_end()
```
**This project seems somehow powerful, but the source code is very simple.**
**Actually, only about 200 lines of Python code.**
**If you want to understand or modify some details of this project, feel free to read and change the source code!!!**
```python
```
## 2, Features
The main features supported by torchkeras are listed below.
Versions when these features are introduced and the libraries which they used or inspired from are given.
|features| supported from version | used or inspired by library |
|:----|:-------------------:|:--------------|
|✅ training progress bar | 3.0.0 | use tqdm,inspired by keras|
|✅ training metrics | 3.0.0 | inspired by pytorch_lightning |
|✅ notebook visualization in traning | 3.8.0 |inspired by fastai |
|✅ early stopping | 3.0.0 | inspired by keras |
|✅ gpu training | 3.0.0 |use accelerate|
|✅ multi-gpus training(ddp) | 3.6.0 | use accelerate|
|✅ fp16/bf16 training| 3.6.0 | use accelerate|
|✅ tensorboard callback | 3.7.0 |use tensorboard |
|✅ wandb callback | 3.7.0 |use wandb |
|✅ VLog | 3.9.5 | use matplotlib|
```python
```
### 3, Basic Examples
You can follow these full examples to get started with torchkeras.
|example| read notebook code | run example in kaggle|
|:----|:-------------------------|:-----------:|
|①kerasmodel basic 🔥🔥| [**torchkeras.KerasModel example**](./01，kerasmodel_example.ipynb) | <br><div></a><a href="https://www.kaggle.com/lyhue1991/kerasmodel-example"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a></div><br> |
|②kerasmodel wandb 🔥🔥🔥|[**torchkeras.KerasModel with wandb demo**](./02，kerasmodel_wandb_demo.ipynb) | <br><div></a><a href="https://www.kaggle.com/lyhue1991/kerasmodel-wandb-example"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a></div><br> |
|③kerasmodel tunning 🔥🔥🔥|[**torchkeras.KerasModel with wandb sweep demo**](./03，kerasmodel_tuning_demo.ipynb) | <br><div></a><a href="https://www.kaggle.com/lyhue1991/torchkeras-loves-wandb-sweep"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a></div><br> |
|④kerasmodel tensorboard | [**torchkeras.KerasModel with tensorboard example**](./04，kerasmodel_tensorboard_demo.ipynb) | |
|⑤kerasmodel ddp/tpu | [**torchkeras.KerasModel ddp tpu examples**](https://www.kaggle.com/code/lyhue1991/torchkeras-ddp-tpu-examples) |<br><div></a><a href="https://www.kaggle.com/lyhue1991/torchkeras-ddp-tpu-examples"><img src="https://kaggle.com/static/images/open-in-kaggle.svg" alt="Open In Kaggle"></a></div><br> |
|⑥ VLog for lightgbm/ultralytics/transformers🔥🔥🔥| [**VLog example**](./10，vlog_example.ipynb) | |
### 4, Advanced Examples
In some using cases, because of the differences of the model input types, you need to rewrite the StepRunner of
KerasModel. Here are some examples.
|example|model library |notebook |
|:----|:-----------|:-----------:|
||||
|**RL**|||
|ReinforcementLearning——Q-Learning🔥🔥|- |[Q-learning](./examples/Q-learning.ipynb)|
|ReinforcementLearning——DQN|- |[DQN](./examples/DQN.ipynb)|
||||
|**CV**|||
|ImageClassification——Resnet| - | [Resnet](./examples/ResNet.ipynb) |
|ImageSegmentation——UNet| - | [UNet](./examples/UNet.ipynb) |
|ObjectDetection——SSD| - | [SSD](./examples/SSD.ipynb) |
|OCR——CRNN 🔥🔥| - | [CRNN-CTC](./examples/CRNN_CTC.ipynb) |
|ImageClassification——SwinTransformer|timm| [Swin](./examples/SwinTransformer——timm.ipynb)|
|ObjectDetection——FasterRCNN| torchvision | [FasterRCNN](./examples/FasterRCNN——vision.ipynb) |
|ImageSegmentation——DeepLabV3++ | segmentation_models_pytorch | [Deeplabv3++](./examples/Deeplabv3plus——smp.ipynb) |
|InstanceSegmentation——MaskRCNN | detectron2 | [MaskRCNN](./examples/MaskRCNN——detectron2.ipynb) |
|ObjectDetection——YOLOv8 🔥🔥🔥| ultralytics | [YOLOv8](./examples/YOLOV8_Detect——ultralytics.ipynb) |
|InstanceSegmentation——YOLOv8 🔥🔥🔥| ultralytics | [YOLOv8](./examples/YOLOV8_Segment——ultralytics.ipynb) |
||||
|**NLP**|||
|Seq2Seq——Transformer🔥🔥| - | [Transformer](./examples/Dive_into_Transformer.ipynb) |
|TextGeneration——Llama🔥| - | [Llama](./examples/Dive_into_Llama.ipynb) |
|TextClassification——BERT | transformers | [BERT](./examples/BERT——transformers.ipynb) |
|TokenClassification——BERT | transformers | [BERT_NER](./examples/BERT_NER——transformers.ipynb) |
|FinetuneLLM——ChatGLM2_LoRA 🔥🔥🔥| transformers,peft | [ChatGLM2_LoRA](./examples/ChatGLM2_LoRA——transformers.ipynb) |
|FinetuneLLM——ChatGLM2_AdaLoRA 🔥| transformers,peft | [ChatGLM2_AdaLoRA](./examples/ChatGLM2_AdaLoRA——transformers.ipynb) |
|FinetuneLLM——ChatGLM2_QLoRA🔥 | transformers | [ChatGLM2_QLoRA_Kaggle](./examples/ChatGLM2_QLoRA_Kaggle——transformers.ipynb) |
|FinetuneLLM——BaiChuan13B_QLoRA🔥 | transformers | [BaiChuan13B_QLoRA](./examples/BaiChuan13B_QLoRA——transformers.ipynb) |
|FinetuneLLM——BaiChuan13B_NER 🔥🔥🔥| transformers | [BaiChuan13B_NER](./examples/BaiChuan13B_NER——transformers.ipynb) |
|FinetuneLLM——BaiChuan13B_MultiRounds 🔥| transformers | [BaiChuan13B_MultiRounds](./examples/BaiChuan13B_MultiRounds——transformers.ipynb) |
|FinetuneLLM——Qwen7B_MultiRounds 🔥🔥🔥| transformers | [Qwen7B_MultiRounds](./examples/Qwen7B_MultiRounds——transformers.ipynb) |
|FinetuneLLM——BaiChuan2_13B 🔥| transformers | [BaiChuan2_13B](./examples/BaiChuan2_13B——transformers.ipynb) |
**If you want to understand or modify some details of this project, feel free to read and change the source code!!!**
Any other questions, you can contact the author form the wechat official account below:
**算法美食屋**
Raw data
{
"_id": null,
"home_page": "https://github.com/lyhue1991/torchkeras",
"name": "torchkeras",
"maintainer": "",
"docs_url": null,
"requires_python": ">=3.5",
"maintainer_email": "",
"keywords": "vlog,deep-learning,DL,pytorch,torch,keras",
"author": "PythonAiRoad, Laugh",
"author_email": "lyhue1991@163.com",
"download_url": "https://files.pythonhosted.org/packages/0c/e3/eb84af79839490b383c43a6f2a3fa8c3cf6eb47ed5dd295625bcdd920878/torchkeras-3.9.6.tar.gz",
"platform": null,
"description": "\n# Pytorch\u2764\ufe0fKeras\n\nEnglish | [\u7b80\u4f53\u4e2d\u6587](README.md)\n\n\nThe torchkeras library is a simple tool for training neural network in pytorch jusk in a keras style. \ud83d\ude0b\ud83d\ude0b\n\n\n## 1, Introduction\n\n\nWith torchkeras, You need not to write your training loop with many lines of code, all you need to do is just \n\nlike these two steps as below:\n\n(i) create your network and wrap it and the loss_fn together with torchkeras.KerasModel like this: \n`model = torchkeras.KerasModel(net,loss_fn=nn.BCEWithLogitsLoss())`.\n\n(ii) fit your model with the training data and validate data.\n\n\nThe main code of use torchkeras is like below.\n\n```python\nimport torch \nimport torchkeras\n\nmodel = torchkeras.KerasModel(net,\n loss_fn = nn.BCEWithLogitsLoss(),\n optimizer= torch.optim.Adam(net.parameters(),lr = 0.001),\n metrics_dict = {\"acc\":torchmetrics.Accuracy(task='binary')}\n )\ndfhistory=model.fit(train_data=dl_train, \n val_data=dl_val, \n epochs=20, \n patience=3, \n ckpt_path='checkpoint',\n monitor=\"val_acc\",\n mode=\"max\",\n plot=True\n )\n\n```\n\n\n\n\nBesides\uff0cYou can use torchkeras.VLog to get the dynamic training visualization any where as you like ~\n\n```python\nimport time\nimport math,random\nfrom torchkeras import VLog\n\nepochs = 10\nbatchs = 30\n\n#0, init vlog\nvlog = VLog(epochs, monitor_metric='val_loss', monitor_mode='min') \n\n#1, log_start \nvlog.log_start() \n\nfor epoch in range(epochs):\n \n #train\n for step in range(batchs):\n \n #2, log_step (for training step)\n vlog.log_step({'train_loss':100-2.5*epoch+math.sin(2*step/batchs)}) \n time.sleep(0.05)\n \n #eval \n for step in range(20):\n \n #3, log_step (for eval step)\n vlog.log_step({'val_loss':100-2*epoch+math.sin(2*step/batchs)},training=False)\n time.sleep(0.05)\n \n #4, log_epoch\n vlog.log_epoch({'val_loss':100 - 2*epoch+2*random.random()-1,\n 'train_loss':100-2.5*epoch+2*random.random()-1}) \n\n# 5, log_end\nvlog.log_end()\n\n```\n\n\n**This project seems somehow powerful, but the source code is very simple.**\n\n**Actually, only about 200 lines of Python code.**\n\n**If you want to understand or modify some details of this project, feel free to read and change the source code!!!**\n\n```python\n\n```\n\n## 2, Features \n\n\nThe main features supported by torchkeras are listed below.\n\nVersions when these features are introduced and the libraries which they used or inspired from are given.\n\n\n\n|features| supported from version | used or inspired by library |\n|:----|:-------------------:|:--------------|\n|\u2705 training progress bar | 3.0.0 | use tqdm,inspired by keras|\n|\u2705 training metrics | 3.0.0 | inspired by pytorch_lightning |\n|\u2705 notebook visualization in traning | 3.8.0 |inspired by fastai |\n|\u2705 early stopping | 3.0.0 | inspired by keras |\n|\u2705 gpu training | 3.0.0 |use accelerate|\n|\u2705 multi-gpus training(ddp) | 3.6.0 | use accelerate|\n|\u2705 fp16/bf16 training| 3.6.0 | use accelerate|\n|\u2705 tensorboard callback | 3.7.0 |use tensorboard |\n|\u2705 wandb callback | 3.7.0 |use wandb |\n|\u2705 VLog | 3.9.5 | use matplotlib|\n\n\n```python\n\n```\n\n### 3, Basic Examples \n\nYou can follow these full examples to get started with torchkeras.\n\n\n|example| read notebook code | run example in kaggle| \n|:----|:-------------------------|:-----------:|\n|\u2460kerasmodel basic \ud83d\udd25\ud83d\udd25| [**torchkeras.KerasModel example**](./01\uff0ckerasmodel_example.ipynb) | <br><div></a><a href=\"https://www.kaggle.com/lyhue1991/kerasmodel-example\"><img src=\"https://kaggle.com/static/images/open-in-kaggle.svg\" alt=\"Open In Kaggle\"></a></div><br> |\n|\u2461kerasmodel wandb \ud83d\udd25\ud83d\udd25\ud83d\udd25|[**torchkeras.KerasModel with wandb demo**](./02\uff0ckerasmodel_wandb_demo.ipynb) | <br><div></a><a href=\"https://www.kaggle.com/lyhue1991/kerasmodel-wandb-example\"><img src=\"https://kaggle.com/static/images/open-in-kaggle.svg\" alt=\"Open In Kaggle\"></a></div><br> |\n|\u2462kerasmodel tunning \ud83d\udd25\ud83d\udd25\ud83d\udd25|[**torchkeras.KerasModel with wandb sweep demo**](./03\uff0ckerasmodel_tuning_demo.ipynb) | <br><div></a><a href=\"https://www.kaggle.com/lyhue1991/torchkeras-loves-wandb-sweep\"><img src=\"https://kaggle.com/static/images/open-in-kaggle.svg\" alt=\"Open In Kaggle\"></a></div><br> |\n|\u2463kerasmodel tensorboard | [**torchkeras.KerasModel with tensorboard example**](./04\uff0ckerasmodel_tensorboard_demo.ipynb) | |\n|\u2464kerasmodel ddp/tpu | [**torchkeras.KerasModel ddp tpu examples**](https://www.kaggle.com/code/lyhue1991/torchkeras-ddp-tpu-examples) |<br><div></a><a href=\"https://www.kaggle.com/lyhue1991/torchkeras-ddp-tpu-examples\"><img src=\"https://kaggle.com/static/images/open-in-kaggle.svg\" alt=\"Open In Kaggle\"></a></div><br> |\n|\u2465 VLog for lightgbm/ultralytics/transformers\ud83d\udd25\ud83d\udd25\ud83d\udd25| [**VLog example**](./10\uff0cvlog_example.ipynb) | |\n\n\n### 4, Advanced Examples \n\nIn some using cases, because of the differences of the model input types, you need to rewrite the StepRunner of \nKerasModel. Here are some examples.\n\n|example|model library |notebook |\n|:----|:-----------|:-----------:|\n||||\n|**RL**|||\n|ReinforcementLearning\u2014\u2014Q-Learning\ud83d\udd25\ud83d\udd25|- |[Q-learning](./examples/Q-learning.ipynb)|\n|ReinforcementLearning\u2014\u2014DQN|- |[DQN](./examples/DQN.ipynb)|\n||||\n|**CV**|||\n|ImageClassification\u2014\u2014Resnet| - | [Resnet](./examples/ResNet.ipynb) |\n|ImageSegmentation\u2014\u2014UNet| - | [UNet](./examples/UNet.ipynb) |\n|ObjectDetection\u2014\u2014SSD| - | [SSD](./examples/SSD.ipynb) |\n|OCR\u2014\u2014CRNN \ud83d\udd25\ud83d\udd25| - | [CRNN-CTC](./examples/CRNN_CTC.ipynb) |\n|ImageClassification\u2014\u2014SwinTransformer|timm| [Swin](./examples/SwinTransformer\u2014\u2014timm.ipynb)|\n|ObjectDetection\u2014\u2014FasterRCNN| torchvision | [FasterRCNN](./examples/FasterRCNN\u2014\u2014vision.ipynb) | \n|ImageSegmentation\u2014\u2014DeepLabV3++ | segmentation_models_pytorch | [Deeplabv3++](./examples/Deeplabv3plus\u2014\u2014smp.ipynb) |\n|InstanceSegmentation\u2014\u2014MaskRCNN | detectron2 | [MaskRCNN](./examples/MaskRCNN\u2014\u2014detectron2.ipynb) |\n|ObjectDetection\u2014\u2014YOLOv8 \ud83d\udd25\ud83d\udd25\ud83d\udd25| ultralytics | [YOLOv8](./examples/YOLOV8_Detect\u2014\u2014ultralytics.ipynb) |\n|InstanceSegmentation\u2014\u2014YOLOv8 \ud83d\udd25\ud83d\udd25\ud83d\udd25| ultralytics | [YOLOv8](./examples/YOLOV8_Segment\u2014\u2014ultralytics.ipynb) |\n||||\n|**NLP**|||\n|Seq2Seq\u2014\u2014Transformer\ud83d\udd25\ud83d\udd25| - | [Transformer](./examples/Dive_into_Transformer.ipynb) |\n|TextGeneration\u2014\u2014Llama\ud83d\udd25| - | [Llama](./examples/Dive_into_Llama.ipynb) |\n|TextClassification\u2014\u2014BERT | transformers | [BERT](./examples/BERT\u2014\u2014transformers.ipynb) |\n|TokenClassification\u2014\u2014BERT | transformers | [BERT_NER](./examples/BERT_NER\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014ChatGLM2_LoRA \ud83d\udd25\ud83d\udd25\ud83d\udd25| transformers,peft | [ChatGLM2_LoRA](./examples/ChatGLM2_LoRA\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014ChatGLM2_AdaLoRA \ud83d\udd25| transformers,peft | [ChatGLM2_AdaLoRA](./examples/ChatGLM2_AdaLoRA\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014ChatGLM2_QLoRA\ud83d\udd25 | transformers | [ChatGLM2_QLoRA_Kaggle](./examples/ChatGLM2_QLoRA_Kaggle\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014BaiChuan13B_QLoRA\ud83d\udd25 | transformers | [BaiChuan13B_QLoRA](./examples/BaiChuan13B_QLoRA\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014BaiChuan13B_NER \ud83d\udd25\ud83d\udd25\ud83d\udd25| transformers | [BaiChuan13B_NER](./examples/BaiChuan13B_NER\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014BaiChuan13B_MultiRounds \ud83d\udd25| transformers | [BaiChuan13B_MultiRounds](./examples/BaiChuan13B_MultiRounds\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014Qwen7B_MultiRounds \ud83d\udd25\ud83d\udd25\ud83d\udd25| transformers | [Qwen7B_MultiRounds](./examples/Qwen7B_MultiRounds\u2014\u2014transformers.ipynb) |\n|FinetuneLLM\u2014\u2014BaiChuan2_13B \ud83d\udd25| transformers | [BaiChuan2_13B](./examples/BaiChuan2_13B\u2014\u2014transformers.ipynb) |\n\n\n**If you want to understand or modify some details of this project, feel free to read and change the source code!!!**\n\nAny other questions, you can contact the author form the wechat official account below:\n\n**\u7b97\u6cd5\u7f8e\u98df\u5c4b** \n\n\n\n\n",
"bugtrack_url": null,
"license": "",
"summary": "pytorch\u2764\ufe0fkeras",
"version": "3.9.6",
"project_urls": {
"Homepage": "https://github.com/lyhue1991/torchkeras"
},
"split_keywords": [
"vlog",
"deep-learning",
"dl",
"pytorch",
"torch",
"keras"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "c530ae8a58ff24dfa7a9ae29e45a9e2d985a2c8b9c851207124018294026f696",
"md5": "ddcf8566d71f260cda2ee707632df765",
"sha256": "e64a6f5ec62cca0cd55ca12e51f2504904e57aabad2c907e127fa9e1bd173d78"
},
"downloads": -1,
"filename": "torchkeras-3.9.6-py3-none-any.whl",
"has_sig": false,
"md5_digest": "ddcf8566d71f260cda2ee707632df765",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.5",
"size": 6536895,
"upload_time": "2024-02-18T09:25:16",
"upload_time_iso_8601": "2024-02-18T09:25:16.350124Z",
"url": "https://files.pythonhosted.org/packages/c5/30/ae8a58ff24dfa7a9ae29e45a9e2d985a2c8b9c851207124018294026f696/torchkeras-3.9.6-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "0ce3eb84af79839490b383c43a6f2a3fa8c3cf6eb47ed5dd295625bcdd920878",
"md5": "d58af6a47caab1afe56484a519a59cc7",
"sha256": "8961ceaab5e35e9cc4b7390aac4a927ee9b1ea50bdfdd1c8ec5c8438a4f3c67e"
},
"downloads": -1,
"filename": "torchkeras-3.9.6.tar.gz",
"has_sig": false,
"md5_digest": "d58af6a47caab1afe56484a519a59cc7",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.5",
"size": 6513281,
"upload_time": "2024-02-18T09:25:36",
"upload_time_iso_8601": "2024-02-18T09:25:36.234781Z",
"url": "https://files.pythonhosted.org/packages/0c/e3/eb84af79839490b383c43a6f2a3fa8c3cf6eb47ed5dd295625bcdd920878/torchkeras-3.9.6.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2024-02-18 09:25:36",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "lyhue1991",
"github_project": "torchkeras",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "torchkeras"
}
JSON
