enable misc test cases on XPU #38852
Conversation
Signed-off-by: YAO Matrix <matrix.yao@intel.com>
| @@ -3940,7 +3941,7 @@ def test_torchdynamo_memory(self): | |||
| from torch import _dynamo as torchdynamo | |||
|
|
|||
| class CustomTrainer(Trainer): | |||
| def compute_loss(self, model, inputs, return_outputs=False): | |||
| def compute_loss(self, model, inputs, num_items_in_batch=None, return_outputs=False): | |||
There was a problem hiding this comment.
fix the similar issue as #36331
Signed-off-by: YAO Matrix <matrix.yao@intel.com>
Signed-off-by: YAO Matrix <matrix.yao@intel.com>
| @@ -623,7 +623,7 @@ def test_simple_generate(self): | |||
| ( | |||
| "xpu", | |||
| 3, | |||
| ): "<|begin_of_text|>Hey how are you doing on this lovely evening? I hope you are all doing well. Today I", | |||
| ): "<|begin_of_text|>Hey how are you doing on this lovely evening? I hope you are all doing well. I am", | |||
There was a problem hiding this comment.
tweak ground truth on stock PyTorch
There was a problem hiding this comment.
If you don't mind, could you change
(
"cuda",
8,
)
to
("cuda", 8)
etc.
no big deal though 🙏
There was a problem hiding this comment.
BTW, could you share what is stock PyTorch? You probably mentioned once but I can't remember.
There was a problem hiding this comment.
oh, sorry for confusing on vocabulary. stock PyTorch in our vocabulary means "PyTorch released by Meta w/o any extensions or modification", it's what users can get by pip install torch --index-url xxx
There was a problem hiding this comment.
If you don't mind, could you change
( "cuda", 8, )to
("cuda", 8)etc.
no big deal though 🙏
done, thx.
| @@ -623,7 +623,7 @@ def test_simple_generate(self): | |||
| ( | |||
| "xpu", | |||
| 3, | |||
| ): "<|begin_of_text|>Hey how are you doing on this lovely evening? I hope you are all doing well. Today I", | |||
| ): "<|begin_of_text|>Hey how are you doing on this lovely evening? I hope you are all doing well. I am", | |||
There was a problem hiding this comment.
If you don't mind, could you change
(
"cuda",
8,
)
to
("cuda", 8)
etc.
no big deal though 🙏
| @@ -623,7 +623,7 @@ def test_simple_generate(self): | |||
| ( | |||
| "xpu", | |||
| 3, | |||
| ): "<|begin_of_text|>Hey how are you doing on this lovely evening? I hope you are all doing well. Today I", | |||
| ): "<|begin_of_text|>Hey how are you doing on this lovely evening? I hope you are all doing well. I am", | |||
There was a problem hiding this comment.
BTW, could you share what is stock PyTorch? You probably mentioned once but I can't remember.
| @@ -1623,9 +1623,10 @@ def is_any_loss_nan_or_inf(log_history): | |||
| self.assertFalse(is_any_loss_nan_or_inf(log_history_filter)) | |||
|
|
|||
| def test_train_and_eval_dataloaders(self): | |||
| if torch_device in ["cuda", "xpu"]: | |||
| if torch_device in ["cuda"]: | |||
There was a problem hiding this comment.
I am ok, but since you added that before, it was a mistake back then?
There was a problem hiding this comment.
yes, it was a mistake by me. In that time I just tested in a single card, so may missed. In my recent test in multi-card env, i find it fail and debug to the DP issue... That's why i put a comment in else statement, to remind me and other xpu guys to avoid later mistakes in case we forgot the history, :)
| } | ||
| ) | ||
| # fmt: on |
There was a problem hiding this comment.
it could be a simple # fmt: skip after the ending ). For the next time :-) though, if you find it's easier to add.
@ydshieh, pls help review, thx very much.