Update attention_is_all_you_need sample readme

* Update attention_is_all_you_need sample readme

Author: zhangxiang1993

PyTorch/1.13/README.md

@@ -14,27 +14,22 @@ Follow the steps below to get set up with PyTorch on DirectML.
 
 2. Clone this repo.	
 
-3. Install prerequisites
-```
-    pip install torchvision==0.14.0 
-    pip install torch==1.13
-    pip install torch-directml
-```
+3. Install **torch-directml**
 
-4. _(optional)_ Run `pip list`. The following packages should be installed:
-```
-torch                   1.13.0
-torch-directml          0.1.13.*
-torchvision             0.14.0
+>⚠️ Since torch-directml 0.1.13.1.*, **torch** and **torchvision** will be installed as dependencies
+
+```ps
+pip install torch-directml
 ```
 
-5. Create a DML Device and Test
+4. Create a DML Device and Test
+
 ```
 import torch
 import torch_directml
 dml = torch_directml.device()
 ```
->⚠️ Note that device creation has changed in torch-directml 1.13 from previous versions. The torch-directml backend is currently mapped to “PrivateUse1." The new `torch_directml.device()` API is a convenient wrapper for creating your tenors on the correct device. 
+>⚠️ Note that device creation has changed in torch-directml 0.1.13 from previous versions. The torch-directml backend is currently mapped to “PrivateUse1." The new `torch_directml.device()` API is a convenient wrapper for creating your tenors on the correct device.
 
 ## Samples
 

PyTorch/1.13/attention_is_all_you_need/README.md

@@ -5,33 +5,22 @@ This is a PyTorch Directml implementation of the Transformer model in "[Attentio
 
 This sample is extracted from [pytorch benchmark](https://github.com/pytorch/benchmark/tree/main/torchbenchmark/models/attention_is_all_you_need_pytorch), and has been slightly changed to apply torch-directml.
 
-
-# Requirement
-- python  3.8
-- torch 1.13
-- torch_directml 0.1.13
-- torchtext 0.14.0
-- spacy
-- tqdm
-- dill
-- numpy
-
-
 # Usage
 
-## 1) Run install.py to download and preprocess data
+## 1) Install the prerequisites and prepare data
+
 From inside the `attention_is_all_you_need` directory, run the following script:
 ```ps
 python install.py
 ```
 
 ## 2) Train the model
 ```bash
-python train.py -data_pkl m30k_deen_shr.pkl -log m30k_deen_shr -embs_share_weight -proj_share_weight -label_smoothing -save_model trained -b 128 -warmup 128000 -epoch 400 -use_dml
+python train.py -data_pkl .data/m30k_deen_shr.pkl -log m30k_deen_shr -embs_share_weight -proj_share_weight -label_smoothing -save_model trained -b 128 -warmup 128000 -epoch 400 -use_dml
 ```
 
 ## 3) Test the model
 ```bash
-python translate.py -data_pkl m30k_deen_shr.pkl -model trained.chkpt -output prediction.txt -use_dml
+python translate.py -data_pkl .data/m30k_deen_shr.pkl -model trained.chkpt -output prediction.txt -use_dml
 ```
 

PyTorch/1.13/attention_is_all_you_need/install.py

@@ -12,11 +12,10 @@ def spacy_download(language):
 
 def preprocess():
     current_dir = Path(os.path.dirname(os.path.realpath(__file__)))
-    multi30k_data_dir = os.path.join(current_dir.parent.parent, ".data", "multi30k")
-    root = os.path.join(str(Path(__file__).parent), ".data")
-    os.makedirs(root, exist_ok=True)
+    data_root = os.path.join(current_dir, ".data")
+    os.makedirs(data_root, exist_ok=True)
     subprocess.check_call([sys.executable, os.path.join(current_dir, 'preprocess.py'), '-lang_src', 'de', '-lang_trg', 'en', '-share_vocab',
-                           '-save_data', os.path.join(root, 'm30k_deen_shr.pkl')])
+                           '-save_data', os.path.join(data_root, 'm30k_deen_shr.pkl')])
 
 if __name__ == '__main__':
     pip_install_requirements()

PyTorch/1.13/attention_is_all_you_need/requirements.txt

@@ -1,6 +1,6 @@
+torchtext
 dill==0.3.4
 tqdm
 iopath
 numpy
 spacy==2.3.5
-torchtext==0.14.0

PyTorch/README.md

@@ -11,15 +11,11 @@ PyTorch on DirectML is supported on both the latest versions of Windows 10 and t
 ## Pytorch with DirectML Versions
 | torch-directml        | pytorch |
 |-----------------------|-------|
-| [0.1.13.\*](https://pypi.org/project/torch-directml/)                | 1.13  |
+| [0.1.13+](https://pypi.org/project/torch-directml/)                | 1.13+  |
 | [1.8.0a0.\*](https://pypi.org/project/pytorch-directml/) | 1.8   |
 
-## Setup
-* For users of Pytorch-DirectML forked from Pytorch __1.13__, see the setup instructions in the [1.13](./1.13/) folder. 
-* For users of Pytorch-DirectML forked from Pytorch __1.8__, see the setup instructions in the [1.8](./1.8/) folder.
-
 ## Samples
-For users of Pytorch-DirectML forked from Pytorch 1.13, the samples can be found below or in the [1.13](./1.13/) folder: 
+For users of Pytorch-DirectML forked from Pytorch 1.13 or higher, the samples can be found below or in the [1.13](./1.13/) folder:
 * [attenion is all you need- the original transformer model](./1.13/attention_is_all_you_need/)
 * [yolov3- a real-time object detection model](./1.13/yolov3/)
 * [squeezenet - a small image classification model](./1.13/squeezenet)