Rooba
/
lightning
mirror of https://github.com/Lightning-AI/lightning.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
7,969 Commits 59 Branches 217 Tags 356 MiB
Commit Graph

68 Commits

Author SHA1 Message Date
Carlos Mocholí 02074f16c7
Fix PyTorch versions in Lite CI (#15338) 
* replace oldest in lite

* Fix PyTorch versions in Lite CI

* This will be moved to install pkg workflow in the mirror PR

* 1.13 fixes

* Windows fix

* sorting

Co-authored-by: otaj <ota@lightning.ai>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
 2022-10-26 15:09:08 -04:00
Adrian Wälchli 38a9e69543
Extend the detection of interactive mode (#15293) 
* extend interactive mode detection
* update test names
* changelog
* test
 2022-10-26 15:24:11 +00:00
Adrian Wälchli 0f9156374d
Mark internal Lite APIs as protected (#15307) 
* mark internal lite apis as protected
* formatting
* docs update

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
 2022-10-26 12:51:50 +00:00
otaj 76e462a0be
Do not lose references of trainer in test (#15272) 
* Fix reference error

* Skip flaky hanging test

* .

Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-10-25 09:23:15 -04:00
Dan Dale 27585a9bcf
Fix and refactor `test_deepspeed_engine_is_steppable` test (#15251) 
Co-authored-by: Rohit Gupta <rohitgr1998@gmail.com>
Co-authored-by: Adrian Wälchli <aedu.waelchli@gmail.com>
 2022-10-23 21:25:36 +00:00
Carlos Mocholí 961e395677
Resolve collectives test issues (#15195) 
Co-authored-by: otaj <ota@lightning.ai>
 2022-10-21 01:08:38 +00:00
Carlos Mocholí b866dc3a6a
Collective's PREMUL_SUM support with PyTorch 1.13 (#15201) 
* Collective's PREMUL_SUM support with PyTorch 1.13
* Fix test
* Skip under 1.13
 2022-10-20 12:36:06 +00:00
Carlos Mocholí bf458701de
Avoid underscore suffix in filenames (#15189) 2022-10-20 07:39:19 -04:00
otaj 741462f373
[LAI] Make lite tests safe for combined package (#15204) 
Make lite tests safe for combined package

Co-authored-by: Justus Schock <12886177+justusschock@users.noreply.github.com>
 2022-10-20 09:10:39 +00:00
Adrian Wälchli 576757fd79
Validate SRUN variables when launching in SLURM (#15011) 2022-10-19 21:42:11 +00:00
Adrian Wälchli 045c2f5715
Efficient gradient accumulation in LightningLite (#14966) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-10-19 19:55:12 +00:00
Jirka Borovec d0b092fda8
Lite: setting extras & fix CI (#15192) 
* extras
* test.txt
* doctest
* Apply suggestions from code review
* Fix imports
* Oops

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-10-19 19:05:23 +00:00
Carlos Mocholí 24c26f7db2
Standardize Lite's filenames (#15058) 2022-10-19 14:09:41 +02:00
Carlos Mocholí 0e18266023
Fix collective tests with PyTorch 1.13 (#15167) 2022-10-18 14:31:48 +02:00
Justus Schock 27965cc36b
Fix locally failing lite tests (#15137) 2022-10-18 09:49:14 +00:00
Adrian Wälchli ed891e5049
Force NVML-based CUDA check in PyTorch 1.14+ (#15110) 2022-10-13 13:10:29 -04:00
Carlos Mocholí da25d1d30d
Remove unused Lite code (#15000) 
* Remove unused Lite code
* Remove duplicate import
* Group variable
* Fix monkeypatch
 2022-10-10 22:16:56 +00:00
Carlos Mocholí c334b7766c
Remove old testing artifacts (#15052) 2022-10-10 17:34:18 +00:00
Carlos Mocholí d15bd1520e
[Lite] precision_plugin -> precision (#15001) 
Co-authored-by: Adrian Wälchli <aedu.waelchli@gmail.com>
 2022-10-10 15:00:32 +00:00
Carlos Mocholí 0b04aa879f
Resolve interactions between CUDA tests (#15042) 2022-10-09 06:20:40 -04:00
Adrian Wälchli c76a95ea12
More tests for TPU accelerator in Lite (#14960) 
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
 2022-10-08 15:42:21 +00:00
Carlos Mocholí 62ca073a41
Introduce base collective and main subclasses (#15016) 
Co-authored-by: otaj <ota@lightning.ai>
 2022-10-07 19:53:19 +00:00
Dan Dale 3b75c52869
Support ddp_fork strategy with native AMP by attempting NVML-based CUDA availability assessment (#14984) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
 2022-10-05 18:52:06 -04:00
Dan Dale ab1eb6531e
Fix fork tests failing in environments with CUDA available (#14982) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-10-05 00:02:55 +00:00
Carlos Mocholí 7ef87464dd
Refactor XLA and TPU checks across codebase (#14550) 2022-10-04 22:54:14 +00:00
Carlos Mocholí 3028fd287d
Fix TPU test CI (#14926) 
* Fix TPU test CI

* +x first

* Lite first to uncovert errors faster

* Fixes

* One more

* Simplify XLALauncher wrapping to avoid pickle error

* debug

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Debug commit successful. Trying local definitions

* Require tpu for mock test

* ValueError: The number of devices must be either 1 or 8, got 4 instead

* Fix mock test

* Simplify call, rely on defaults

* Skip OSError for now. Maybe upgrading will help

* Simplify launch tests, move some to lite

* Stricter typing

* RuntimeError: Accessing the XLA device before processes have spawned is not allowed.

* Revert "RuntimeError: Accessing the XLA device before processes have spawned is not allowed."

This reverts commit f65107ebf3.

* Alternative boring solution to the reverted commit

* Fix failing test on CUDA machine

* Workarounds

* Try latest mkl

* Revert "Try latest mkl"

This reverts commit d06813aa67.

* Wrong exception

* xfail

* Mypy

* Comment change

* Spawn launch refactor

* Accept that we cannot lazy init now

* Fix mypy and launch test failures

* The base dockerfile already includes mkl-2022.1.0 - what if we use it?

* try a different mkl version

* Revert mkl version changes

Co-authored-by: awaelchli <aedu.waelchli@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Akihiro Nitta <nitta@akihironitta.com>
 2022-10-03 09:13:33 -04:00
Adrian Wälchli d7af8ce2a5
Simplify root node resolution for SLURM environment (#14912) 
Co-authored-by: Seppo Enarvi <seppo.git@marjaniemi.com>
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-09-30 15:40:43 +00:00
Adrian Wälchli cd9247a782
Introduce primitives for input/output dtype conversion in Lite Precision (#14792) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
Co-authored-by: Justus Schock <12886177+justusschock@users.noreply.github.com>
Co-authored-by: otaj <6065855+otaj@users.noreply.github.com>
 2022-09-30 15:29:03 +00:00
Carlos Mocholí 6256a318d7
Refactor launching tests to use our launchers (#14954) 2022-09-30 09:57:18 +02:00
Atharva Phatak fdcb5cc90b
Hydra changes to lightning-lite (#14950) 
Co-authored-by: awaelchli <aedu.waelchli@gmail.com>
 2022-09-29 21:59:35 -04:00
Adrian Wälchli 498cb60417
Fairscale integration tests for Lite (#14921) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-09-29 17:46:49 +00:00
Adrian Wälchli 5b446aec4d
DeepSpeed integration tests for Lite (#14901) 2022-09-29 16:39:32 +00:00
Adrian Wälchli ea5e817973
Better error message when trying to re-initialize CUDA in forked subprocess (#14709) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-09-28 05:07:33 -04:00
Carlos Mocholí 9fc4ff3278
Move logic to error out on deprecation warnings into conftest (#14902) 2022-09-27 17:49:25 +02:00
Adrian Wälchli d572a7e2ec
Fix double precision support in Lite (#14827) 2022-09-27 08:38:20 +00:00
Adrian Wälchli d7404c775a
Integration tests for Precision in Lite (#14815) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
Co-authored-by: Justus Schock <12886177+justusschock@users.noreply.github.com>
 2022-09-26 18:50:11 +00:00
Adrian Wälchli dc1dc0df36
Attempt to query device count via NVML (#14631) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-09-22 09:57:13 +00:00
otaj 5ee2b86c44
Tests for fixed TypeError (#14821) 
* tests for 14809
* Apply suggestions from code review

Co-authored-by: Justus Schock <12886177+justusschock@users.noreply.github.com>
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
 2022-09-22 09:04:27 +02:00
Carlos Mocholí 7e803ba53e
Clean-up dtype management (#14823) 
Co-authored-by: Justus Schock <12886177+justusschock@users.noreply.github.com>
 2022-09-22 00:07:36 +00:00
Adrian Wälchli 3f0fec591d
Update device attribute in Lite's module wrapper (#14822) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
 2022-09-21 19:06:10 +00:00
Carlos Mocholí abc805f9ef
Remove the model argument from Lite's `optimizer_step` via structural typing (#14810) 
Co-authored-by: awaelchli <aedu.waelchli@gmail.com>
 2022-09-21 19:28:45 +02:00
awaelchli c0ff7a1b77 Add backward-compatibility for LightningLite in PL (#14735) 2022-09-20 13:31:56 +02:00
awaelchli e3e71670e6 Move src/pytorch_lightning/lite to src/lightning_lite (#14735) 2022-09-20 13:31:56 +02:00
Carlos Mocholí e9c571d39f
Move accelerator-specific parsing functions with their accelerators (#14753) 
Co-authored-by: awaelchli <aedu.waelchli@gmail.com>
 2022-09-18 22:48:45 +00:00
Adrian Wälchli 1092265140
Remove check `num_slurm_tasks` in Lite (#14761) 2022-09-18 14:01:49 -04:00
Adrian Wälchli 35c65b0287
Fix test suite when running on MPS-enabled hardware (#14708) 2022-09-16 19:21:36 +00:00
Adrian Wälchli 47f0d336f1
Standalone Lite: Update LightningLite (#14726) 2022-09-16 17:25:27 +00:00
Adrian Wälchli 619e76f22d
Remove silent behavior when `num_slurm_tasks` does not correspond to number of processes in Trainer (#14300) 
* simplify logic
* remove hpc
* update
* add changelog
* more tests
* update test

Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
Co-authored-by: Jirka <jirka.borovec@seznam.cz>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
 2022-09-16 11:00:09 +00:00
Adrian Wälchli 38d89713a5
Standalone Lite: Connector (#14692) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
Co-authored-by: Kaushik B <45285388+kaushikb11@users.noreply.github.com>
 2022-09-15 14:14:51 +00:00
Adrian Wälchli d3dcd68852
Standalone Lite: DDP Spawn Strategy Family (#14675) 
Co-authored-by: Carlos Mocholí <carlossmocholi@gmail.com>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
 2022-09-15 10:51:12 +00:00