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

Remove static pretrained maps from the library's internals (#29112)

Browse Source 
* [test_all] Remove static pretrained maps from the library's internals

* Deprecate archive maps instead of removing them

* Revert init changes

* [test_all] Deprecate instead of removing

* [test_all] PVT v2 support

* [test_all] Tests should all pass

* [test_all] Style

* Address review comments

* Update src/transformers/models/deprecated/_archive_maps.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Update src/transformers/models/deprecated/_archive_maps.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* [test_all] trigger tests

* [test_all] LLAVA

* [test_all] Bad rebase

---------

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
This commit is contained in:
Lysandre Debut
2024-03-25 10:33:38 +01:00
committed by GitHub
parent 76a33a1092
commit 39114c0383
842 changed files with 4608 additions and 8613 deletions
Download Patch File Download Diff File Expand all files Collapse all files

188
tests/models/auto/test_modeling_auto.py
View File

@@ -85,10 +85,6 @@ if is_torch_available():
MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING,
MODEL_MAPPING,
)
from transformers.models.bert.modeling_bert import BERT_PRETRAINED_MODEL_ARCHIVE_LIST
from transformers.models.gpt2.modeling_gpt2 import GPT2_PRETRAINED_MODEL_ARCHIVE_LIST
from transformers.models.t5.modeling_t5 import T5_PRETRAINED_MODEL_ARCHIVE_LIST
from transformers.models.tapas.modeling_tapas import TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST
@require_torch
@@ -98,138 +94,134 @@ class AutoModelTest(unittest.TestCase):
@slow
def test_model_from_pretrained(self):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModel.from_pretrained(model_name)
model, loading_info = AutoModel.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertModel)
model = AutoModel.from_pretrained(model_name)
model, loading_info = AutoModel.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertModel)
self.assertEqual(len(loading_info["missing_keys"]), 0)
# When using PyTorch checkpoint, the expected value is `8`. With `safetensors` checkpoint (if it is
# installed), the expected value becomes `7`.
EXPECTED_NUM_OF_UNEXPECTED_KEYS = 7 if is_safetensors_available() else 8
self.assertEqual(len(loading_info["unexpected_keys"]), EXPECTED_NUM_OF_UNEXPECTED_KEYS)
self.assertEqual(len(loading_info["mismatched_keys"]), 0)
self.assertEqual(len(loading_info["error_msgs"]), 0)
self.assertEqual(len(loading_info["missing_keys"]), 0)
# When using PyTorch checkpoint, the expected value is `8`. With `safetensors` checkpoint (if it is
# installed), the expected value becomes `7`.
EXPECTED_NUM_OF_UNEXPECTED_KEYS = 7 if is_safetensors_available() else 8
self.assertEqual(len(loading_info["unexpected_keys"]), EXPECTED_NUM_OF_UNEXPECTED_KEYS)
self.assertEqual(len(loading_info["mismatched_keys"]), 0)
self.assertEqual(len(loading_info["error_msgs"]), 0)
@slow
def test_model_for_pretraining_from_pretrained(self):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModelForPreTraining.from_pretrained(model_name)
model, loading_info = AutoModelForPreTraining.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForPreTraining)
# Only one value should not be initialized and in the missing keys.
for key, value in loading_info.items():
self.assertEqual(len(value), 0)
model = AutoModelForPreTraining.from_pretrained(model_name)
model, loading_info = AutoModelForPreTraining.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForPreTraining)
# Only one value should not be initialized and in the missing keys.
for key, value in loading_info.items():
self.assertEqual(len(value), 0)
@slow
def test_lmhead_model_from_pretrained(self):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModelWithLMHead.from_pretrained(model_name)
model, loading_info = AutoModelWithLMHead.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForMaskedLM)
model = AutoModelWithLMHead.from_pretrained(model_name)
model, loading_info = AutoModelWithLMHead.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForMaskedLM)
@slow
def test_model_for_causal_lm(self):
for model_name in GPT2_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, GPT2Config)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, GPT2Config)
model = AutoModelForCausalLM.from_pretrained(model_name)
model, loading_info = AutoModelForCausalLM.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, GPT2LMHeadModel)
model = AutoModelForCausalLM.from_pretrained(model_name)
model, loading_info = AutoModelForCausalLM.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, GPT2LMHeadModel)
@slow
def test_model_for_masked_lm(self):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModelForMaskedLM.from_pretrained(model_name)
model, loading_info = AutoModelForMaskedLM.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForMaskedLM)
model = AutoModelForMaskedLM.from_pretrained(model_name)
model, loading_info = AutoModelForMaskedLM.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForMaskedLM)
@slow
def test_model_for_encoder_decoder_lm(self):
for model_name in T5_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, T5Config)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, T5Config)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
model, loading_info = AutoModelForSeq2SeqLM.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, T5ForConditionalGeneration)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
model, loading_info = AutoModelForSeq2SeqLM.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, T5ForConditionalGeneration)
@slow
def test_sequence_classification_model_from_pretrained(self):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModelForSequenceClassification.from_pretrained(model_name)
model, loading_info = AutoModelForSequenceClassification.from_pretrained(
model_name, output_loading_info=True
)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForSequenceClassification)
model = AutoModelForSequenceClassification.from_pretrained(model_name)
model, loading_info = AutoModelForSequenceClassification.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForSequenceClassification)
@slow
def test_question_answering_model_from_pretrained(self):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
model, loading_info = AutoModelForQuestionAnswering.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForQuestionAnswering)
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
model, loading_info = AutoModelForQuestionAnswering.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForQuestionAnswering)
@slow
def test_table_question_answering_model_from_pretrained(self):
for model_name in TAPAS_PRETRAINED_MODEL_ARCHIVE_LIST[5:6]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, TapasConfig)
model_name = "google/tapas-base"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, TapasConfig)
model = AutoModelForTableQuestionAnswering.from_pretrained(model_name)
model, loading_info = AutoModelForTableQuestionAnswering.from_pretrained(
model_name, output_loading_info=True
)
self.assertIsNotNone(model)
self.assertIsInstance(model, TapasForQuestionAnswering)
model = AutoModelForTableQuestionAnswering.from_pretrained(model_name)
model, loading_info = AutoModelForTableQuestionAnswering.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, TapasForQuestionAnswering)
@slow
def test_token_classification_model_from_pretrained(self):
for model_name in BERT_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model_name = "google-bert/bert-base-uncased"
config = AutoConfig.from_pretrained(model_name)
self.assertIsNotNone(config)
self.assertIsInstance(config, BertConfig)
model = AutoModelForTokenClassification.from_pretrained(model_name)
model, loading_info = AutoModelForTokenClassification.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForTokenClassification)
model = AutoModelForTokenClassification.from_pretrained(model_name)
model, loading_info = AutoModelForTokenClassification.from_pretrained(model_name, output_loading_info=True)
self.assertIsNotNone(model)
self.assertIsInstance(model, BertForTokenClassification)
@slow
def test_auto_backbone_timm_model_from_pretrained(self):