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

Add output in a dictionary for TF generate method (#12139)

Browse Source 
* Add output args to greedy search

* Fix critical typo + make style quality

* Handle generate_beam_search

* Add dict_specific tests and fix the placement of encoder outputs

* Add  specific outputs

* Update doc

* Fix typo

* Adjust handling encoder_outputs + Fix generating for T5

* Fix generate for RAG

* Fix handling ouptut_attentions when target_mapping is not None

Take care of situations when target_mapping is provided
as there are 2-tuple of attentions

Change from:
if inputs["output_attentions"]:
    attentions = tuple(tf.transpose(t, perm(2, 3, 0, 1)) for t in attentions)

to:
if inputs["output_attentions"]:
    if inputs["target_mapping"] is not None:
        # when target_mapping is provided, there are 2-tuple of attentions
         attentions = tuple(
             tuple(tf.transpose(attn_stream, perm=(2, 3, 0, 1)) for attn_stream in t) for t in attentions
        )
    else:
        attentions = tuple(tf.transpose(t, perm=(2, 3, 0, 1)) for t in attentions)

* Rename kwargs to model_kwargs

* make style quality

* Move imports in test_modeling_tf_common.py

Move ModelOutput-related imports in test_modeling_tf_common.py
into the `is_tf_available():` statement.

* Rewrite nested if-statements

* Fix added tests
This commit is contained in:
Daniel Stancl
2021-06-23 11:52:11 +02:00
committed by GitHub
parent d4be498441
commit 26a2e36595
5 changed files with 672 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

74
tests/test_modeling_tf_common.py
View File

@@ -61,6 +61,16 @@ if is_tf_available():
TFSharedEmbeddings,
tf_top_k_top_p_filtering,
)
from transformers.generation_tf_utils import (
TFBeamSampleDecoderOnlyOutput,
TFBeamSampleEncoderDecoderOutput,
TFBeamSearchDecoderOnlyOutput,
TFBeamSearchEncoderDecoderOutput,
TFGreedySearchDecoderOnlyOutput,
TFGreedySearchEncoderDecoderOutput,
TFSampleDecoderOnlyOutput,
TFSampleEncoderDecoderOutput,
)
if _tf_gpu_memory_limit is not None:
gpus = tf.config.list_physical_devices("GPU")
@@ -1100,6 +1110,37 @@ class TFModelTesterMixin:
generated_ids = output_tokens[:, input_ids.shape[-1] :]
self.assertFalse(self._check_match_tokens(generated_ids.numpy().tolist(), bad_words_ids))
def test_lm_head_model_no_beam_search_generate_dict_outputs(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_ids = inputs_dict.get("input_ids", None)
# iterate over all generative models
for model_class in self.all_generative_model_classes:
model = model_class(config)
output_greedy = model.generate(
input_ids,
do_sample=False,
output_scores=True,
output_hidden_states=True,
output_attentions=True,
return_dict_in_generate=True,
)
output_sample = model.generate(
input_ids,
do_sample=True,
output_scores=True,
output_hidden_states=True,
output_attentions=True,
return_dict_in_generate=True,
)
if model.config.is_encoder_decoder:
self.assertIsInstance(output_greedy, TFGreedySearchEncoderDecoderOutput)
self.assertIsInstance(output_sample, TFSampleEncoderDecoderOutput)
else:
self.assertIsInstance(output_greedy, TFGreedySearchDecoderOnlyOutput)
self.assertIsInstance(output_sample, TFSampleDecoderOnlyOutput)
def test_lm_head_model_random_beam_search_generate(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_ids = inputs_dict.get("input_ids", None)
@@ -1140,6 +1181,39 @@ class TFModelTesterMixin:
generated_ids = output_tokens[:, input_ids.shape[-1] :]
self.assertFalse(self._check_match_tokens(generated_ids.numpy().tolist(), bad_words_ids))
def test_lm_head_model_beam_search_generate_dict_outputs(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
input_ids = inputs_dict.get("input_ids", None)
# iterate over all generative models
for model_class in self.all_generative_model_classes:
model = model_class(config)
output_beam_search = model.generate(
input_ids,
num_beams=2,
do_sample=False,
output_scores=True,
output_hidden_states=True,
output_attentions=True,
return_dict_in_generate=True,
)
output_beam_sample = model.generate(
input_ids,
num_beams=2,
do_sample=True,
output_scores=True,
output_hidden_states=True,
output_attentions=True,
return_dict_in_generate=True,
)
if model.config.is_encoder_decoder:
self.assertIsInstance(output_beam_search, TFBeamSearchEncoderDecoderOutput)
self.assertIsInstance(output_beam_sample, TFBeamSampleEncoderDecoderOutput)
else:
self.assertIsInstance(output_beam_search, TFBeamSearchDecoderOnlyOutput)
self.assertIsInstance(output_beam_sample, TFBeamSampleDecoderOnlyOutput)
def test_loss_computation(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes: