Generate: support for left-padding on GPTNeoX and Llama (#22382)
This commit is contained in:
@@ -237,7 +237,7 @@ class GPTNeoXModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
|
||||
@require_torch
|
||||
class GPTNeoXLanguageGenerationTest(unittest.TestCase):
|
||||
@slow
|
||||
def test_lm_generate_codegen(self):
|
||||
def test_lm_generate_gptneox(self):
|
||||
tokenizer = AutoTokenizer.from_pretrained("EleutherAI/pythia-410m-deduped")
|
||||
for checkpointing in [True, False]:
|
||||
model = GPTNeoXForCausalLM.from_pretrained("EleutherAI/pythia-410m-deduped")
|
||||
|
||||
Reference in New Issue
Block a user