SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fix max_length criteria when using inputs_embeds (#28994)

Browse Source 
* fix max_length for inputs_embeds

* make style

* Update src/transformers/generation/utils.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* Static Cache: load models with MQA or GQA (#28975)

* fix

* fix tests

* fix tests

* Update src/transformers/generation/utils.py

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

* more fixes

* make style

---------

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
This commit is contained in:
Raushan Turganbay
2024-02-16 16:25:12 +05:00
committed by GitHub
parent 8876ce8a5f
commit aee11fe427
2 changed files with 26 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/generation/test_utils.py
View File

@@ -1963,7 +1963,7 @@ class GenerationTesterMixin:
)
self.assertListEqual(
outputs_from_embeds[:, inputs_embeds.shape[1] :].tolist(),
outputs_from_embeds_wo_ids[:, 1:].tolist(),
outputs_from_embeds_wo_ids.tolist(),
)
def test_generate_continue_from_past_key_values(self):
@@ -2730,6 +2730,20 @@ class GenerationIntegrationTests(unittest.TestCase, GenerationIntegrationTestsMi
**model_kwargs,
)
def test_max_length_if_input_embeds(self):
# PT-only test: TF doesn't have StoppingCriteria
article = "Today a dragon flew over Paris."
model = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-gpt2").to(torch_device)
tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-gpt2")
input_ids = tokenizer(article, return_tensors="pt").input_ids.to(torch_device)
inputs_embeds = model.get_input_embeddings()(input_ids)
max_length = 20
input_len = input_ids.shape[-1]
out_gen = model.generate(input_ids=input_ids, max_length=max_length)
out_gen_embeds = model.generate(inputs_embeds=inputs_embeds, max_length=max_length)
self.assertEqual(out_gen.shape[-1], input_len + out_gen_embeds.shape[-1])
def test_custom_stopping_criteria_overload_error(self):
# PT-only test: TF doesn't have StoppingCriteria
article = """Justin Timberlake and Jessica Biel, welcome to parenthood."""