Skip to content

Commit

Permalink
docs: add release notes for 0.18.0 (#4102)
Browse files Browse the repository at this point in the history
  • Loading branch information
dzhu authored May 9, 2022
1 parent 5051024 commit 797ceec
Show file tree
Hide file tree
Showing 10 changed files with 72 additions and 83 deletions.
83 changes: 72 additions & 11 deletions docs/release-notes.txt
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,70 @@
Release Notes
###############

**************
Version 0.18
**************

Version 0.18.0
==============

**Release Date:** May 09, 2022

**New Features**

- Add the Core API. The Core API is the first API offered by Determined that allows users to fully
integrate arbitrary models and training loops into the Determined platform. All of the features
offered by the higher-level Trial APIs, such as reporting metrics, pausing and reactivating,
hyperparameter search, and distributed training, are now available to arbitrary models,
frameworks, and training loops, with only light code changes. See the :ref:`user guide
<core-getting-started>` and :ref:`reference documentation <core-reference>` to get started.

- **Breaking Change**: Checkpoints: The Python SDK's ``Checkpoint.download()`` method now writes a
differently formatted ``metadata.json`` file into the checkpoint directory. Previously, the JSON
content in the file contained many system-defined fields, plus a ``metadata`` field that
contained the user-defined metadata for the checkpoint, which was also available as a Python
object as ``Checkpoint.metadata``. Now, ``metadata.json`` contains only the user-defined
metadata, and those metadata appear as top-level keys. Some of the fields which were previously
system-defined are now considered user-defined, even though they are uploaded automatically in
Trial-based training. This decision is in line with the Trial APIs now being optional---that is,
part of userspace---after the release of the Core API.

- Job queue: Add support for dynamic job modification on Kubernetes using the job queue. Users can
now use the WebUI or CLI to change the priority and queue position of jobs in k8s. To update jobs
through the WebUI, go to the Job Queue section, find the target job, and click on the Manage Job
option. To update jobs in the CLI, use the ``det job update`` command. Run ``det job update
--help`` for more information.

**Bug Fixes**

- CLI: API requests executed through the Python bindings have been erroneously using the SSL
"noverify" option since version 0.17.6, making them potentially insecure. The option is now
disabled.

**Deprecated Features**

- The Determined Data Layer has been deprecated and will be removed in a future version. New code
should not begin using it, but we will assist existing users to migrate to using `YogaDL
<https://yogadl.readthedocs.io>`__ directly before removing the feature.

**Removed Features**

- Python API: The old experimental namespace methods for custom reducers in both PyTorchTrial and
EstimatorTrial have been removed. The experimental names were deprecated in 0.15.2 (April 2021)
when custom reducers were promoted to general availability. Any users who have not already
migrated to the non-experimental namespace for custom reducer methods must do so.

- Searcher: Remove the PBT searcher, which was deprecated in version 0.17.6 (January 2022).

- API: Remove the notebook logs endpoint in favor of the new task logs endpoint.

- Python API: Remove the remaining parts of the Native API, which was deprecated in version 0.13.5
(September 2020). The only Native API functions that still remained were
``det.experimental.create()`` and ``det.expeimental.create_trial_instance()``.

- Python API: Remove the ``det.pytorch.reset_parameters()`` function, which was deprecated in
0.12.13 (August 2020).

**************
Version 0.17
**************
Expand Down Expand Up @@ -31,8 +95,6 @@ Version 0.17.15

- Add a translation of DeepSpeed's DCGAN example using the new DeepSpeedTrial API.

----

Version 0.17.14
===============

Expand All @@ -53,12 +115,11 @@ Version 0.17.13

- Support DeepSpeed with a new DeepSpeedTrial API.

- `DeepSpeed <https://www.deepspeed.ai/>`_ is a powerful library for training large scale models.
With the new ``DeepSpeedTrial`` you can combine all the benefits of Determined with the
features available in DeepSpeed like the Zero Redundancy Optimizer and pipeline parallel
training. We also provide an example based on Eleuther AI's `GPT-NeoX
<https://github.com/EleutherAI/gpt-neox/>`_ repo to help you get started training
state-of-the-art language models.
`DeepSpeed <https://www.deepspeed.ai/>`__ is a powerful library for training large scale models.
With the new ``DeepSpeedTrial`` you can combine all the benefits of Determined with the features
available in DeepSpeed like the Zero Redundancy Optimizer and pipeline parallel training. We also
provide an example based on Eleuther AI's `GPT-NeoX <https://github.com/EleutherAI/gpt-neox/>`__
repo to help you get started training state-of-the-art language models.

- CLI: Allow the CLI to accept any unique prefix of a task UUID to refer to the task, rather than
requiring the entire UUID. In some places, Determined only displays the first few characters of a
Expand Down Expand Up @@ -1039,7 +1100,7 @@ Version 0.14.6

- REST API: Add a new endpoint to delete experiments. This endpoint is only enabled for admin users
and deletes all resources associated with an experiment. This includes checkpoint storage,
Tensorboards, trial logs from all backends and metadata such as history and metrics, stored in
TensorBoards, trial logs from all backends and metadata such as history and metrics, stored in
PostgreSQL.

- REST API: Add a new endpoint to fetch aggregated historical resource allocation information.
Expand Down Expand Up @@ -1825,13 +1886,13 @@ Version 0.13.1
recommended to upgrade to version 0.13.1. Users already on version 0.13.0 should upgrade to
version 0.13.1 as usual.

- Tensorboard: Fix a bug that prevents Tensorboards from experiments with old experiment
- TensorBoard: Fix a bug that prevents TensorBoards from experiments with old experiment
configuration versions from being loaded.

- WebUI: Fix an API response decoding issue on React where a null checkpoint resource was unhandled
and could prevent trial detail page from rendering.

- WebUI: Fix an issue where terminated Tensorboard and notebook tasks were rendered as openable.
- WebUI: Fix an issue where terminated TensorBoard and notebook tasks were rendered as openable.

Version 0.13.0
==============
Expand Down
9 changes: 0 additions & 9 deletions docs/release-notes/3938-job-queue-k8s.txt

This file was deleted.

7 changes: 0 additions & 7 deletions docs/release-notes/4055-native.txt

This file was deleted.

7 changes: 0 additions & 7 deletions docs/release-notes/4056-data-layer.txt

This file was deleted.

8 changes: 0 additions & 8 deletions docs/release-notes/4060-custom-reducers.txt

This file was deleted.

5 changes: 0 additions & 5 deletions docs/release-notes/4066-reset-params.txt

This file was deleted.

20 changes: 0 additions & 20 deletions docs/release-notes/4069-core-api.txt

This file was deleted.

6 changes: 0 additions & 6 deletions docs/release-notes/fix-noverify.txt

This file was deleted.

5 changes: 0 additions & 5 deletions docs/release-notes/notebook-logs.txt

This file was deleted.

5 changes: 0 additions & 5 deletions docs/release-notes/remove-pbt.txt

This file was deleted.

0 comments on commit 797ceec

Please sign in to comment.