Determined - 深度学习训练平台

Changelog

d9a2cdc fix: avoid saving pytorch model architecture (#594)
071b3eb docs: release notes for 0.12.5 (#595)
be835fd feat: use str instead of pathlib.Path in checkpoint callbacks
4b1c8ca fix: enable logging with --local --test mode (#589)
0978df0 docs: add documentation for EsatimatorTrial callbacks
2e1eb39 feat: add callbacks to EstimatorTrial
52d50c9 fix: fix off-by-one error in master logs
fc27be0 feat: auto focus the username field on page load (#580)
9475664 docs: tweak advice on mounting file systems with cloud deployments
49ae381 docs: use "distributed training" to mean any kind of multi-GPU training
9e5506d feat: support hooks in EvalSpec when using TF Estimators
f975690 docs: add documentation for pytorch callbacks
9ab8984 feat: add timeout to tensorboard startup
354c744 fix: fix startup-hook.sh for tensorboard-entrypoint.sh
69999cd docs: evaluation functions should return JSON-serializable metrics
e0ec9f1 feat: support PyTorch on_validation_end callbacks with multi-GPU
bd569b6 feat: update to nccl 2.6.4 and fix multi-machine dtrain (#564)
1b79ec1 feat: add meta-learning example using protonets (#527)
b003bc7 docs: re-organize model definitions docs
7b82073 docs: package-based install documentation improvements
b0f075d feat: bump YogaDL verion to 0.1.1
ae8e192 feat: make TF 2.2 the default TF2 version
253fd74 feat: support TF 2.2
4f99bb9 feat: add multiple lr schedulers example
085114e fix: pass s3 endpoint url to tensorboard process
2cf584c fix: set default http method name to GET (#516)
a7a3221 feat: add on_checkpoint_end PyTorch callback
5fa952e feat: add user with password and user without password to tests
5c8bf4c fix: enable OSX interactive session during det user create
9c08cc6 feat: support arbitrary login redirect routes (#522)
47ec6cc feat: checkpoint load uses trial to retrieve checkpoints
f14f769 feat: add model code to checkpoints
8623fea feat: add ability to load a trial class locally
f1b30a4 feat: support --csv, --json to slot list and agent list in CLI
9a707ab docs: add CONTRIBUTING guidelines
bdf83f8 fix: don't support AMP w/ aggregation_frequency > 1
13078b7 docs: update Users docs for auto-login removal [DET-2992] (#532)
25ffffe fix: re-sync package-lock.json to package.json for react
87dc544 docs: various fixes
208ee11 feat: add imagenet NAS architecture (#378)
6c185a2 feat: add container count to "det agent list"
ebf3628 feat: make Link component to be based on HTML anchor tag
2300788 docs: release notes for v0.12.4 (#520)
7668874 fix: update login docs link
74b0fb4 feat: add support for presenting the authentication token
b659f72 docs: update reference docs to include PyTorchTrialContext
4c55a90 feat: add a PyTorchContext with ability to access model, optimizer, lr scheduler
f2726c6 feat: add generic callback support to PyTorch trials

Docker images

• docker pull determinedai/determined-master:0.12.5
• docker pull determinedai/determined-master:35e75b5
• docker pull determinedai/determined-master:35e75b5c2fa2241f2ecdccbdc58634d107234377

Changelog

65b1e51 chore: bump version: 0.12.3rc3 -> 0.12.3
4ddfab4 fix: add logic in agent to invoke docker credentials helpers
dcd4d86 docs: add 0.12.3 release notes
c6ca487 fix: fix loading logic for tensorpacktrial when backbone is set
f2e5702 chore: bump version: 0.12.3rc2 -> 0.12.3rc3
17818fc fix: remove cd from startup-hook.sh in bert_glue_pytorch
d6f429e fix: fix make -C agent clean
ef74363 chore: bump version: 0.12.3rc1 -> 0.12.3rc2
ffc4544 feat: remove container_path from shared_fs configuration documentation
dab6de7 fix: wrong postgres query in restart
22e7be9 chore: better logs --tail default
87bcc70 fix: fix FasterRCNN example
4aeeca1 fix: install determined before using it
f502935 fix: update imports in nas
9d5df9d fix: add specificity to css rules to avoid override sub styles
85c6c1e fix: add missing order for steps in det t describe
a8ee76e chore: fix determined-deploy publish
52723b0 feat: reintroduce trial endpoints
6deafa9 fix: add script to run inside agent's container
a352a21 docs: reorganize main install page
2743056 chore: bump version: 0.12.3rc0 -> 0.12.3rc1
50f452c docs: add a topic guide for effective distributed training
2e6c826 docs: cleanup docstrings
e94e865 chore: bump version: 0.12.3.dev0 -> 0.12.3rc0
e3de26c docs: improve docs around context_dir argument
42b3223 feat: improve experiment configuration handling for LOCAL mode
6e29449 feat: add --local support to cli
5873205 chore: update company name in license files
66da4e9 docs: fix typo in quickstart
7aea9c8 docs: document Docker daemon socket bind mount requirement for agents
fc591c1 fix: display progress bar for image pulls
04edcca docs: add HP and dtrain to quick-start
154e3af chore: change mode of trial entrypoint to 744
a900609 chore: make PyTorch checkpoint code path open to all
97d127e fix: properly restore stopping experiments
defe9ec fix: incorrect shared_fs path for tensorboard if storage_path is set
cbfe85c docs: do some miscellaneous copy editing
ff6f7aa docs: use better code block markup in CLI reference page
1f67d8f docs: fix link to docs on Chinese AWS site
67c6668 docs: edit doc page on users
df2b3b8 docs: remove outdated reference to single-file model definitions
4175ea9 docs: fix unintentionally separated lists
edad1a7 docs: fix unintentional definition lists
23c444f docs: add attributions for master and agent
2203869 feat: add restart policy for fault-tolerance
673007a feat: re-revert add Determined object for consolidating authentication
daf6244 feat: usability messages for det-deploy aws
e7ece2a fix: correctly initialize LRSchedulers for multi-gpu training
c35cb1b fix: replace get_lr() with get_last_lr() in LRScheduler
62beacf fix: correctly call step() for LRScheduler when using epochs
c3cadc3 fix: revert add Determined object for consolidating authentication
b8ddc3d feat: add Determined object for consolidating authentication
1d5e3fe feat: make pending commands killable
c82925b feat: exit log tailing based on state and time rather than polls
18ae0e5 docs: add experimental warnings to native tutorial
bbd15ad chore: delegate auth failure handling to error handler
1daa6e6 docs: release notes for version 0.12.2
b953623 fix: clean up logic around commands
2f97381 docs: add a Native API tutorial
d84de8d chore: use pypi to install yogadl
54a3f48 chore: bump version: 0.12.2 -> 0.12.3.dev0
f8a81c0 feat: introduce REST API builder and refactor existing APIs
f0b996c fix: update modal to support flexible height and auto scrolling
e530ad7 chore: add PyTorch object detection example
89c530f feat: sort user selection option to keep the authenticated user on top
2979949 docs: update docs to move native apis to experimental
e9ee930 chore: update native examples to use experimental namespace
df3e0ea chore: fix experiment config comment typos in examples
6f99e07 chore: fix broken and outdated reference links in TF Keras example
bc9ce53 feat: recover notebook state on agent failure
52fb275 chore: shift naming of test and submit to local and cluster
b38bdef fix: grab back control after loading native implementation
69af009 chore: support .dev and rc tags in determined version
1c4be08 chore: bump version: 0.12.1 -> 0.12.2
be3b295 chore: disambiguate logs in workload_sequencer
961798c chore: standardize official example experiment configurations
dfd54ce docs: update README.md for recent docs reorg
515afe4 chore: bump environment images
6667a87 fix: respect shared_fs storage_path configuration in tensorboard
65c3b2e docs: minor fixes for quick start
acabe33 docs: minor fixes for tutorials
bbab37d docs: revert quick-start to use tarball
8a42748 feat: expose container failure reason in trial logs
d5792ad docs: tweak tutorial text around downloading model code
26a1e0b fix: fix data config for MNIST PyTorch distributed example
0c02269 chore: update environments
1eee210 fix: add missing arguments to GraphQL schema update command
d12ceb6 fix: don't automatically import tf with determined
8798035 chore: update generated GraphQL files for function rename
6ecb84e fix: move a new SQL migration after all previous ones numerically
1d014b8 fix: use per-slot batch size in tf keras Native
514a2c6 chore: move calculating batch sizes to EnvContext
c1c5f0b fix: use a default session when initializing TFKerasTrialController
64c9a32 chore: add more logging to harness
9a70a74 feat: add experiment class for querying top checkpoints
2af7f2a feat: add experiment level best checkpoint by metric
4435d7d chore: mount master config file rather etc root
8fb768d fix: add det command to PATH
dcea476 fix: update old references to "mnist_tf_keras"
b44abe6 docs: update CONTRIBUTING.md
dbefe18 chore: add JetBrains IDE config folder to .gitignore
6573bac chore: bump and limit Cypress version to 4.3.x
d1c3e2b fix: fixture-down before fixture-up
4e52149 docs: update local deploy with new commands
60c108e fix: use correct hvd size function
775f24e docs: add documentation for experimental contexts
5c26ca7 docs: add examples using data layer
65c09ff feat: support data layer in harness
9fcefee feat: bind mount data layer paths
cc1a5eb feat: add data layer config
c21aab8 docs: add command to speed up experiments
17ccd19 docs: fix tutorials typos