Merge pull request #99 from grok-ai/develop

0.4.0

.github/workflows/test_suite.yml

@@ -17,6 +17,7 @@ env:
   CONDA_ENV_FILE: 'env.yaml'
   CONDA_ENV_NAME: 'project-test'
   COOKIECUTTER_PROJECT_NAME: 'project-test'
+  HUGGING_FACE_HUB_TOKEN: ${{secrets.HUGGING_FACE_HUB_TOKEN}}
 
 jobs:
   build:

README.md

@@ -73,6 +73,7 @@ Avoid writing boilerplate code to integrate:
 
 - [PyTorch Lightning](https://github.com/PyTorchLightning/pytorch-lightning), lightweight PyTorch wrapper for high-performance AI research.
 - [Hydra](https://github.com/facebookresearch/hydra), a framework for elegantly configuring complex applications.
+- [Hugging Face Datasets](https://huggingface.co/docs/datasets/index),a library for easily accessing and sharing datasets.
 - [Weights and Biases](https://wandb.ai/home), organize and analyze machine learning experiments. *(educational account available)*
 - [Streamlit](https://streamlit.io/), turns data scripts into shareable web apps in minutes.
 - [MkDocs](https://www.mkdocs.org/) and [Material for MkDocs](https://squidfunk.github.io/mkdocs-material/), a fast, simple and downright gorgeous static site generator.

cookiecutter.json

@@ -9,5 +9,5 @@
     "repository_url": "https://github.com/{{ cookiecutter.github_user }}/{{ cookiecutter.project_name.strip().lower().replace(' ', '-') }}",
     "conda_env_name": "{{ cookiecutter.project_name.strip().lower().replace(' ', '-') }}",
     "python_version": "3.11",
-    "__version": "0.3.1"
+    "__version": "0.4.0"
 }

docs/index.md

@@ -36,6 +36,7 @@ and to avoid writing boilerplate code to integrate:
 
 - [PyTorch Lightning](https://github.com/PyTorchLightning/pytorch-lightning), lightweight PyTorch wrapper for high-performance AI research.
 - [Hydra](https://github.com/facebookresearch/hydra), a framework for elegantly configuring complex applications.
+- [Hugging Face Datasets](https://huggingface.co/docs/datasets/index),a library for easily accessing and sharing datasets.
 - [Weights and Biases](https://wandb.ai/home), organize and analyze machine learning experiments. *(educational account available)*
 - [Streamlit](https://streamlit.io/), turns data scripts into shareable web apps in minutes.
 - [MkDocs](https://www.mkdocs.org/) and [Material for MkDocs](https://squidfunk.github.io/mkdocs-material/), a fast, simple and downright gorgeous static site generator.

{{ cookiecutter.repository_name }}/.env.template

@@ -1 +1,10 @@
-# While .env is a local file full of secrets, this can be public and ease the setup of known env variables.
+# .env.template is a template for .env file that can be versioned.
+
+# Set to 1 to show full stack trace on error, 0 to hide it
+HYDRA_FULL_ERROR=1
+
+# Configure where huggingface_hub will locally store data.
+HF_HOME="~/.cache/huggingface"
+
+# Configure the User Access Token to authenticate to the Hub
+# HUGGING_FACE_HUB_TOKEN=

{{ cookiecutter.repository_name }}/.github/workflows/publish.yml

@@ -9,8 +9,9 @@ env:
   CACHE_NUMBER: 0  # increase to reset cache manually
   CONDA_ENV_FILE: './env.yaml'
   CONDA_ENV_NAME: '{{ cookiecutter.conda_env_name }}'
-
 {% raw %}
+  HUGGING_FACE_HUB_TOKEN: ${{secrets.HUGGING_FACE_HUB_TOKEN}}
+
 jobs:
   build:
     strategy:

{{ cookiecutter.repository_name }}/.github/workflows/test_suite.yml

@@ -16,8 +16,9 @@ env:
   CACHE_NUMBER: 1  # increase to reset cache manually
   CONDA_ENV_FILE: './env.yaml'
   CONDA_ENV_NAME: '{{ cookiecutter.conda_env_name }}'
-
 {% raw %}
+  HUGGING_FACE_HUB_TOKEN: ${{secrets.HUGGING_FACE_HUB_TOKEN}}
+
 jobs:
   build:
 

{{ cookiecutter.repository_name }}/conf/default.yaml

@@ -5,6 +5,10 @@ core:
   version: 0.0.1
   tags: null
 
+conventions:
+  x_key: 'x'
+  y_key: 'y'
+
 defaults:
   - hydra: default
   - nn: default

{{ cookiecutter.repository_name }}/conf/nn/data/dataset/vision/mnist.yaml

@@ -0,0 +1,22 @@
+# This class defines which dataset to use,
+# and also how to split in train/[val]/test.
+_target_: {{ cookiecutter.package_name }}.utils.hf_io.load_hf_dataset
+name: "mnist"
+ref: "mnist"
+train_split: train
+# val_split: val
+val_percentage: 0.1
+test_split: test
+label_key: label
+data_key: image
+num_classes: 10
+input_shape: [1, 28, 28]
+standard_x_key: ${conventions.x_key}
+standard_y_key: ${conventions.y_key}
+transforms:
+  _target_: {{ cookiecutter.package_name }}.utils.hf_io.HFTransform
+  key: ${conventions.x_key}
+  transform:
+    _target_: torchvision.transforms.Compose
+    transforms:
+      - _target_: torchvision.transforms.ToTensor

{{ cookiecutter.repository_name }}/conf/nn/data/default.yaml

@@ -0,0 +1,28 @@
+_target_: {{ cookiecutter.package_name }}.data.datamodule.MyDataModule
+
+val_images_fixed_idxs: [7371, 3963, 2861, 1701, 3172,
+                        1749, 7023, 1606, 6481, 1377,
+                        6003, 3593, 3410, 3399, 7277,
+                        5337, 968, 8206, 288, 1968,
+                        5677, 9156, 8139, 7660, 7089,
+                        1893, 3845, 2084, 1944, 3375,
+                        4848, 8704, 6038, 2183, 7422,
+                        2682, 6878, 6127, 2941, 5823,
+                        9129, 1798, 6477, 9264, 476,
+                        3007, 4992, 1428, 9901, 5388]
+
+accelerator: ${train.trainer.accelerator}
+
+num_workers:
+  train: 4
+  val: 2
+  test: 0
+
+batch_size:
+  train: 512
+  val: 128
+  test: 16
+
+defaults:
+  - _self_
+  - dataset: vision/mnist # pick one of the yamls in nn/data/

{{ cookiecutter.repository_name }}/conf/nn/default.yaml

@@ -1,47 +1,23 @@
-data:
-  _target_: {{ cookiecutter.package_name }}.data.datamodule.MyDataModule
-
-  datasets:
-    train:
-      _target_: {{ cookiecutter.package_name }}.data.dataset.MyDataset
-
-#    val:
-#      - _target_: {{ cookiecutter.package_name }}.data.dataset.MyDataset
-
-    test:
-      - _target_: {{ cookiecutter.package_name }}.data.dataset.MyDataset
-
-  accelerator: ${train.trainer.accelerator}
-
-  num_workers:
-    train: 8
-    val: 4
-    test: 4
-
-  batch_size:
-    train: 32
-    val: 16
-    test: 16
-
-  # example
-  val_percentage: 0.1
+data: ???
 
 module:
-  _target_: {{ cookiecutter.package_name }}.pl_modules.pl_module.MyLightningModule
-
   optimizer:
-    #  Adam-oriented deep learning
     _target_: torch.optim.Adam
-    #  These are all default parameters for the Adam optimizer
-    lr: 0.001
+    lr: 1e-3
     betas: [ 0.9, 0.999 ]
     eps: 1e-08
     weight_decay: 0
 
-  lr_scheduler:
-    _target_: torch.optim.lr_scheduler.CosineAnnealingWarmRestarts
-    T_0: 10
-    T_mult: 2
-    eta_min: 0 # min value for the lr
-    last_epoch: -1
-    verbose: False
+#  lr_scheduler:
+#    _target_: torch.optim.lr_scheduler.CosineAnnealingWarmRestarts
+#    T_0: 20
+#    T_mult: 1
+#    eta_min: 0
+#    last_epoch: -1
+#    verbose: False
+
+
+defaults:
+  - _self_
+  - data: default
+  - module: default

{{ cookiecutter.repository_name }}/conf/nn/module/default.yaml

@@ -0,0 +1,7 @@
+_target_: {{ cookiecutter.package_name }}.pl_modules.pl_module.MyLightningModule
+x_key: ${conventions.x_key}
+y_key: ${conventions.y_key}
+
+defaults:
+  - _self_
+  - model: cnn

{{ cookiecutter.repository_name }}/conf/nn/module/model/cnn.yaml

@@ -0,0 +1,2 @@
+_target_: {{ cookiecutter.package_name }}.modules.module.CNN
+input_shape: ${nn.data.dataset.input_shape}

{{ cookiecutter.repository_name }}/conf/train/default.yaml

@@ -17,30 +17,30 @@ trainer:
 
 restore:
   ckpt_or_run_path: null
-  mode: continue # null, finetune, hotstart, continue
+  mode: null # null, finetune, hotstart, continue
 
 monitor:
   metric: 'loss/val'
   mode: 'min'
 
 callbacks:
-  - _target_: pytorch_lightning.callbacks.EarlyStopping
+  - _target_: lightning.pytorch.callbacks.EarlyStopping
     patience: 42
     verbose: False
     monitor: ${train.monitor.metric}
     mode: ${train.monitor.mode}
 
-  - _target_: pytorch_lightning.callbacks.ModelCheckpoint
+  - _target_: lightning.pytorch.callbacks.ModelCheckpoint
     save_top_k: 1
     verbose: False
     monitor: ${train.monitor.metric}
     mode: ${train.monitor.mode}
 
-  - _target_: pytorch_lightning.callbacks.LearningRateMonitor
+  - _target_: lightning.pytorch.callbacks.LearningRateMonitor
     logging_interval: "step"
     log_momentum: False
 
-  - _target_: pytorch_lightning.callbacks.progress.tqdm_progress.TQDMProgressBar
+  - _target_: lightning.pytorch.callbacks.progress.tqdm_progress.TQDMProgressBar
     refresh_rate: 20
 
 logging:
@@ -49,7 +49,7 @@ logging:
     source: true
 
   logger:
-    _target_: pytorch_lightning.loggers.WandbLogger
+    _target_: lightning.pytorch.loggers.WandbLogger
 
     project: ${core.project_name}
     entity: null

{{ cookiecutter.repository_name }}/setup.cfg

@@ -15,7 +15,8 @@ package_dir=
     =src
 packages=find:
 install_requires =
-    nn-template-core==0.3.*
+    nn-template-core==0.4.*
+    anypy==0.0.*
 
     # Add project specific dependencies
     # Stuff easy to break with updates

{{ cookiecutter.repository_name }}/src/{{ cookiecutter.package_name }}/__init__.py

@@ -6,7 +6,7 @@
 # thus the logging configuration defined in the __init__.py must be called before
 # the lightning import otherwise it has no effect.
 # See https://github.com/PyTorchLightning/pytorch-lightning/issues/1503
-lightning_logger = logging.getLogger("pytorch_lightning")
+lightning_logger = logging.getLogger("lightning.pytorch")
 # Remove all handlers associated with the lightning logger.
 for handler in lightning_logger.handlers[:]:
     lightning_logger.removeHandler(handler)