Add text2text-generation-with-past test for encoder-decoder model #1338
Conversation
There was a problem hiding this comment.
Left a few nits.
I do not understand everyting, but looks good to me from what I could read.
Thanks @mht-sharma !
| for model_id in model_ids: | ||
| if ( | ||
| model_arch == "encoder-decoder" | ||
| and use_cache is True |
There was a problem hiding this comment.
| and use_cache is True | |
| and use_cache |
There was a problem hiding this comment.
I guess this is the common terminology used in the repo, to explicitly check the value?
There was a problem hiding this comment.
It's me who advocated to use that :D But more and more I agree with Michael, we should not explicitly check is True when the type is bool, we should only for Optional[bool].
There was a problem hiding this comment.
I guess this explicit check is something which is also followed in the transformers, I remember Lewis mentioning this, but yeah I agree
| def _get_model_ids(self, model_arch): | ||
| model_ids = MODEL_NAMES[model_arch] | ||
| if isinstance(model_ids, dict): | ||
| model_ids = list(model_ids.keys()) | ||
| else: | ||
| model_ids = [model_ids] | ||
| return model_ids |
There was a problem hiding this comment.
So what you call model_arch is a "model type" like bert, t5 and so on?
This function returns a list of model ids from the Hub for a given architecture?
There was a problem hiding this comment.
It's a helper function for the test to get the list of model_ids for a given model_arch(model_type) from MODEL_NAMES. Earlier there was one model_id for a model_arch. But with this PR there can be multiple (Format is same as used in onnx export).
| for model_id in model_ids: | ||
| if ( | ||
| model_arch == "encoder-decoder" | ||
| and use_cache is True |
There was a problem hiding this comment.
nit
| and use_cache is True | |
| and use_cache |
| for model_id in model_ids: | ||
| if ( | ||
| model_arch == "encoder-decoder" | ||
| and use_cache is True |
mht-sharma
changed the title
Merging as the test error are not related to PR |
What does this PR do?
Creating a test for an
text2text-generation-with-pasttasks for an encoder-decoder model #851--with-pastwith the modelBefore submitting