What's new

Added 🎉

• Added StepGraph and Executor abstractions.
• Added a basic PyTorch training step registered as "torch::train", along with other registrable
  components, such as Model, DataLoader, Sampler, DataCollator, Optimizer, and LRScheduler.
• Added DatasetRemixStep in tango.steps.
• Added module tango.common.sequences.
• Added DatasetDict class in tango.common.dataset_dict.
• Added 🤗 Datasets integration.
• Added command-line options to set log level or disable logging completely.

Changed ⚠️

• Step.work_dir, Step.unique_id, Step.dependencies, and Step.recursive_dependencies
  are now a properties instead of methods.
• tango run command will acquire a lock on the directory to avoid race conditions.
• Integrations can now be installed with pip install tango[INTEGRATION_NAME]. For example,
  pip install tango[torch].
• Added method Registrable.search_modules() for automatically finding and importing the modules
  where a given name might be registered.
• FromParams.from_params() and Registrable.resolve_class_name will now call Registrable.search_modules() to automatically import modules where the type might be defined.
  Thus for classes that are defined and registered within any tango.* submodules it is not necessary to explicitly import them.

Fixed ✅

• Step implementations can now take arbitrary **kwargs in their run() methods.

Commits

3616040 add step graph and executor abstractions, support for distributed training in TorchTrainStep (#14)
0fa7d5e add torch components and simple train Step (#12)
b408b8a auto search and import modules in from params (#10)
2c376f4 Update feature_request.md
ff4fe11 Add 🤗 Datasets integration (#8)
2210c62 add DatasetRemixStep (#7)
9dffab3 document installing with integrations
5e225cf more work on integrations (#6)
f45b98f rename workflow
971906c rename workflow file
d33f57f add extras to setup.py (#5)
cdc4393 add source code to docs
c51f2a0 acquire lock on directory during run