|
1 | 1 | # CHANGELOG
|
2 | 2 |
|
| 3 | +## torchx-0.1.2 |
| 4 | + |
| 5 | +Milestone: https://github.com/pytorch/torchx/milestones/3 |
| 6 | + |
| 7 | +* PyTorch 1.11 Support |
| 8 | +* Python 3.10 Support |
| 9 | +* `torchx.workspace` |
| 10 | + * TorchX now supports a concept of workspaces. This enables seamless launching |
| 11 | + of jobs using changes present in your local workspace. For Docker based |
| 12 | + schedulers, we automatically build a new docker container on job launch |
| 13 | + making it easier than ever to run experiments. #333 |
| 14 | +* `torchx.schedulers` |
| 15 | + * Ray #329 |
| 16 | + * Newly added Ray scheduler makes it easy to launch jobs on Ray. |
| 17 | + * https://pytorch.medium.com/large-scale-distributed-training-with-torchx-and-ray-1d09a329aacb |
| 18 | + * AWS Batch #381 |
| 19 | + * Newly added AWS Batch scheduler makes it easy to launch jobs in AWS with minimal infrastructure setup. |
| 20 | + * Slurm |
| 21 | + * Slurm jobs will by default launch in the current working directory to match `local_cwd` and workspace behavior. #372 |
| 22 | + * Replicas now have their own log files and can be accessed programmatically. #373 |
| 23 | + * Support for `comment`, `mail-user` and `constraint` fields. #391 |
| 24 | + * Workspace support (prototype) - Slurm jobs can now be launched in isolated experiment directories. #416 |
| 25 | + * Kubernetes |
| 26 | + * Support for running jobs under service accounts. #408 |
| 27 | + * Support for specifying instance types. #433 |
| 28 | + * All Docker-based Schedulers (Kubernetes, Batch, Docker) |
| 29 | + * Added bind mount and volume supports #420, #426 |
| 30 | + * Bug fix: Better shm support for large dataloader #429 |
| 31 | + * Support for `.dockerignore` and custom Dockerfiles #401 |
| 32 | + * Local Scheduler |
| 33 | + * Automatically set `CUDA_VISIBLE_DEVICES` #383 |
| 34 | + * Improved log ordering #366 |
| 35 | +* `torchx.components` |
| 36 | + * `dist.ddp` |
| 37 | + * Rendezvous works out of the box on all schedulers #400 |
| 38 | + * Logs are now prefixed with local ranks #412 |
| 39 | + * Can specify resources via the CLI #395 |
| 40 | + * Can specify environment variables via the CLI #399 |
| 41 | + * HPO |
| 42 | + * Ax runner now lives in the Ax repo https://github.com/facebook/Ax/commit/8e2e68f21155e918996bda0b7d97b5b9ef4e0cba |
| 43 | +* `torchx.cli` |
| 44 | + * `.torchxconfig` |
| 45 | + * You can now specify component argument defaults `.torchxconfig` https://github.com/pytorch/torchx/commit/c37cfd7846d5a0cb527dd19c8c95e881858f8f0a |
| 46 | + * `~/.torchxconfig` can now be used to set user level defaults. #378 |
| 47 | + * `--workspace` can be configured #397 |
| 48 | + * Color change and bug fixes #419 |
| 49 | +* `torchx.runner` |
| 50 | + * Now supports workspace interfaces. #360 |
| 51 | + * Returned lines now preserve whitespace to provide support for progress bars #425 |
| 52 | + * Events are now logged to `torch.monitor` when available. #379 |
| 53 | +* `torchx.notebook` (prototype) |
| 54 | + * Added new workspace interface for developing models and launching jobs via a Jupyter Notebook. #356 |
| 55 | +* Docs |
| 56 | + * Improvements to clarify TorchX usage w/ workspaces and general cleanups. |
| 57 | + * #374, #402, #404, #407, #434 |
| 58 | + |
3 | 59 | ## torchx-0.1.1
|
4 | 60 |
|
5 | 61 | * Milestone: https://github.com/pytorch/torchx/milestone/2
|
|
0 commit comments