dnrops
/
transformers
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix trainer test wrt DeepSpeed + auto_find_bs (#29061) 

* FIx trainer test

* Update tests/trainer/test_trainer.py

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

---------

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
This commit is contained in:
Zach Mueller 2024-02-16 10:04:24 -05:00 committed by GitHub
parent 161fe425c9
commit 636b03244c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
tests/trainer/test_trainer.py
View File

@ -1588,18 +1588,10 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
auto_find_batch_size=True,
deepspeed=deepspeed,
)
trainer = Trainer(model, args, train_dataset=train_dataset, callbacks=[MockCudaOOMCallback()])
trainer.train()
# After `auto_find_batch_size` is ran we should now be at 8
self.assertEqual(trainer._train_batch_size, 8)
# We can then make a new Trainer
trainer = Trainer(model, args, train_dataset=train_dataset)
# Check we are at 16 to start
self.assertEqual(trainer._train_batch_size, 16 * max(trainer.args.n_gpu, 1))
trainer.train(resume_from_checkpoint=True)
# We should be back to 8 again, picking up based upon the last ran Trainer
self.assertEqual(trainer._train_batch_size, 8)
# Note: This can have issues, for now we don't support this functionality
# ref: https://github.com/huggingface/transformers/pull/29057
with self.assertRaises(NotImplementedError):
_ = Trainer(model, args, train_dataset=train_dataset, callbacks=[MockCudaOOMCallback()])
def test_auto_batch_size_with_resume_from_checkpoint(self):
train_dataset = RegressionDataset(length=128)