add model resources for CPMAnt (new) (#20906)

* resolve conflicts

* rebase and make style

* test

* test

* test

* rebase and make style

* rebase and make style

* tests

* tests

* rewrite some functions

* rebase and make style

* fix load_tf_weights_in_cpmant

* reformat some unrelated files

* upgrade quality

* fix some bugs & docstring

* add models and tests

* solve conflicts

* resolve conflicts

* resolve conflicts

* resolve conflicts

* resolve conflicts

* tests

* resolve conflicts

* resolve conflicts

* fix load_tf_weights_in_cpmant

* reformat some unrelated files

* upgrade quality

* fix some bugs & docstring

* save resolution

* make style

* delete redefinition code

* reformat function

* reformat

* resolve conflicts

* resolve conflicts

* resolve conflicts

* resolve conflicts

* resolve conflicts

* tests

* resolve conflicts

* resolve conflicts

* fix load_tf_weights_in_cpmant

* reformat some unrelated files

* upgrade quality

* resolve conflicts

* resolve conflicts

* resolve conflicts

* resolve conflicts

* resolve conflicts

* fix load_tf_weights_in_cpmant

* reformat some unrelated files

* upgrade quality

* resolve conflicts

* make style

* fix bugs and refactor

* modify docstrings and make style

* unify import format in __init__.py

* fix import-altclp bug

* fix copies to update index.md

* fix unused config parameters

* fix unused config parameters

* fix unused config parameters

* update README_ja.md

* dummy commit for unit test

* fix attention mask

* add CPMAntTokenizer&-Fast to auto-mapping

* drop redundant changes in README_ko

* fix  defaults in docstring

* fix use_cache and some docstring

* add missing args in tokenizer

* modify tester inheritance

* add is_jieba_available

* fix some bugs

* make style and fix-copies

* add doctests

* skip integration tests

* add is_jieba_available

* fix bugs in common tests

* adjust docstrings and make style

* add argument docstring

* adjust code to some specifications

* make style and fix-copies

* add fast tokenization test

* dummy commit for unit test

* dummy commit for unit test

* dummy commit for unit test

* normalize some comments and names

* Bert->CPMAnt

* camel names and drop redundant codes

* make style and fix-coies

* add CpmTokenizerFast _import_structure

* drop cpmanttokenizerfast in model_doc

* fix some problems

* fix CPMAnt tokenization for common test

* make style and fixup

* fix copies and fixup

* fix bugs in tokenization test

* dummy commit for connection failure in unittest

* fix copies

* drop trailing comma

* fix decorator in tests

* dummy commit for connection failure in unittest

---------
Co-authored-by: Gong Baitao <gongbaitao11@gmail.com>

src/transformers/testing_utils.py

@@ -58,6 +58,7 @@ from .utils import (
     is_flax_available,
     is_ftfy_available,
     is_ipex_available,
+    is_jieba_available,
     is_jumanpp_available,
     is_keras_nlp_available,
     is_librosa_available,
@@ -277,6 +278,13 @@ def require_rjieba(test_case):
     return unittest.skipUnless(is_rjieba_available(), "test requires rjieba")(test_case)
 
 
+def require_jieba(test_case):
+    """
+    Decorator marking a test that requires jieba. These tests are skipped when jieba isn't installed.
+    """
+    return unittest.skipUnless(is_jieba_available(), "test requires jieba")(test_case)
+
+
 def require_tf2onnx(test_case):
     return unittest.skipUnless(is_tf2onnx_available(), "test requires tf2onnx")(test_case)
 

src/transformers/utils/__init__.py

@@ -110,6 +110,7 @@ from .import_utils import (
     is_ftfy_available,
     is_in_notebook,
     is_ipex_available,
+    is_jieba_available,
     is_jumanpp_available,
     is_kenlm_available,
     is_keras_nlp_available,

src/transformers/utils/dummy_pt_objects.py

@@ -1841,6 +1841,30 @@ class ConvNextV2PreTrainedModel(metaclass=DummyObject):
         requires_backends(self, ["torch"])
 
 
+CPMANT_PRETRAINED_MODEL_ARCHIVE_LIST = None
+
+
+class CpmAntForCausalLM(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class CpmAntModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
+class CpmAntPreTrainedModel(metaclass=DummyObject):
+    _backends = ["torch"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch"])
+
+
 CTRL_PRETRAINED_MODEL_ARCHIVE_LIST = None
 
 

src/transformers/utils/import_utils.py

@@ -274,6 +274,13 @@ try:
 except importlib_metadata.PackageNotFoundError:
     _decord_availale = False
 
+_jieba_available = importlib.util.find_spec("jieba") is not None
+try:
+    _jieba_version = importlib_metadata.version("jieba")
+    logger.debug(f"Successfully imported jieba version {_jieba_version}")
+except importlib_metadata.PackageNotFoundError:
+    _jieba_available = False
+
 # This is the version of torch required to run torch.fx features and torch.onnx with dictionary inputs.
 TORCH_FX_REQUIRED_VERSION = version.parse("1.10")
 
@@ -740,6 +747,10 @@ def is_cython_available():
     return importlib.util.find_spec("pyximport") is not None
 
 
+def is_jieba_available():
+    return _jieba_available
+
+
 # docstyle-ignore
 DATASETS_IMPORT_ERROR = """
 {0} requires the 🤗 Datasets library but it was not found in your environment. You can install it with:
@@ -997,6 +1008,11 @@ CYTHON_IMPORT_ERROR = """
 Cython`. Please note that you may need to restart your runtime after installation.
 """
 
+JIEBA_IMPORT_ERROR = """
+{0} requires the jieba library but it was not found in your environment. You can install it with pip: `pip install
+jieba`. Please note that you may need to restart your runtime after installation.
+"""
+
 BACKENDS_MAPPING = OrderedDict(
     [
         ("bs4", (is_bs4_available, BS4_IMPORT_ERROR)),
@@ -1029,6 +1045,7 @@ BACKENDS_MAPPING = OrderedDict(
         ("oneccl_bind_pt", (is_ccl_available, CCL_IMPORT_ERROR)),
         ("decord", (is_decord_available, DECORD_IMPORT_ERROR)),
         ("cython", (is_cython_available, CYTHON_IMPORT_ERROR)),
+        ("jieba", (is_jieba_available, JIEBA_IMPORT_ERROR)),
     ]
 )
 

tests/models/cpmant/test_modeling_cpmant.py

+# coding=utf-8
+# Copyright 2022 The OpenBMB Team and The HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" Testing suite for the PyTorch CPMAnt model. """
+
+import unittest
+
+from transformers.testing_utils import is_torch_available, require_torch, tooslow
+
+from ...generation.test_utils import torch_device
+from ...test_configuration_common import ConfigTester
+from ...test_modeling_common import ModelTesterMixin, ids_tensor
+
+
+if is_torch_available():
+    import torch
+
+    from transformers import (
+        CpmAntConfig,
+        CpmAntForCausalLM,
+        CpmAntModel,
+        CpmAntTokenizer,
+    )
+
+
+@require_torch
+class CpmAntModelTester:
+    def __init__(
+        self,
+        parent,
+        batch_size=2,
+        seq_length=8,
+        is_training=True,
+        use_token_type_ids=False,
+        use_input_mask=False,
+        use_labels=False,
+        use_mc_token_ids=False,
+        vocab_size=99,
+        hidden_size=32,
+        num_hidden_layers=3,
+        num_attention_heads=4,
+        intermediate_size=37,
+        num_buckets=32,
+        max_distance=128,
+        prompt_length=8,
+        prompt_types=8,
+        segment_types=8,
+        init_std=1.0,
+        return_dict=True,
+    ):
+        self.parent = parent
+        self.batch_size = batch_size
+        self.seq_length = seq_length
+        self.is_training = is_training
+        self.use_token_type_ids = use_token_type_ids
+        self.use_input_mask = use_input_mask
+        self.use_labels = use_labels
+        self.use_mc_token_ids = use_mc_token_ids
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.intermediate_size = intermediate_size
+        self.num_buckets = num_buckets
+        self.max_distance = max_distance
+        self.prompt_length = prompt_length
+        self.prompt_types = prompt_types
+        self.segment_types = segment_types
+        self.init_std = init_std
+        self.return_dict = return_dict
+
+    def prepare_config_and_inputs(self):
+        input_ids = {}
+        input_ids["input_ids"] = ids_tensor([self.batch_size, self.seq_length], self.vocab_size).type(torch.int32)
+        input_ids["use_cache"] = False
+
+        config = self.get_config()
+
+        return (config, input_ids)
+
+    def get_config(self):
+        return CpmAntConfig(
+            vocab_size=self.vocab_size,
+            hidden_size=self.hidden_size,
+            num_hidden_layers=self.num_hidden_layers,
+            num_attention_heads=self.num_attention_heads,
+            dim_ff=self.intermediate_size,
+            position_bias_num_buckets=self.num_buckets,
+            position_bias_max_distance=self.max_distance,
+            prompt_types=self.prompt_types,
+            prompt_length=self.prompt_length,
+            segment_types=self.segment_types,
+            use_cache=True,
+            init_std=self.init_std,
+            return_dict=self.return_dict,
+        )
+
+    def create_and_check_cpmant_model(self, config, input_ids, *args):
+        model = CpmAntModel(config=config)
+        model.to(torch_device)
+        model.eval()
+
+        hidden_states = model(**input_ids).last_hidden_state
+
+        self.parent.assertEqual(hidden_states.shape, (self.batch_size, self.seq_length, config.hidden_size))
+
+    def create_and_check_lm_head_model(self, config, input_ids, *args):
+        model = CpmAntForCausalLM(config)
+        model.to(torch_device)
+        input_ids["input_ids"] = input_ids["input_ids"].to(torch_device)
+        model.eval()
+
+        model_output = model(**input_ids)
+        self.parent.assertEqual(
+            model_output.logits.shape,
+            (self.batch_size, self.seq_length, config.vocab_size + config.prompt_types * config.prompt_length),
+        )
+
+    def prepare_config_and_inputs_for_common(self):
+        config, inputs_dict = self.prepare_config_and_inputs()
+        return config, inputs_dict
+
+
+@require_torch
+class CpmAntModelTest(ModelTesterMixin, unittest.TestCase):
+    all_model_classes = (CpmAntModel, CpmAntForCausalLM) if is_torch_available() else ()
+
+    test_pruning = False
+    test_missing_keys = False
+    test_mismatched_shapes = False
+    test_head_masking = False
+    test_resize_embeddings = False
+
+    def setUp(self):
+        self.model_tester = CpmAntModelTester(self)
+        self.config_tester = ConfigTester(self, config_class=CpmAntConfig)
+
+    def test_config(self):
+        self.config_tester.create_and_test_config_common_properties()
+        self.config_tester.create_and_test_config_to_json_string()
+        self.config_tester.create_and_test_config_to_json_file()
+        self.config_tester.create_and_test_config_from_and_save_pretrained()
+        self.config_tester.check_config_can_be_init_without_params()
+        self.config_tester.check_config_arguments_init()
+
+    def test_inputs_embeds(self):
+        unittest.skip("CPMAnt doesn't support input_embeds.")(self.test_inputs_embeds)
+
+    def test_retain_grad_hidden_states_attentions(self):
+        unittest.skip(
+            "CPMAnt doesn't support retain grad in hidden_states or attentions, because prompt management will peel off the output.hidden_states from graph.\
+                 So is attentions. We strongly recommand you use loss to tune model."
+        )(self.test_retain_grad_hidden_states_attentions)
+
+    def test_cpmant_model(self):
+        config, inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_cpmant_model(config, inputs)
+
+    def test_cpmant_lm_head_model(self):
+        config, inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_lm_head_model(config, inputs)
+
+
+@require_torch
+class CpmAntModelIntegrationTest(unittest.TestCase):
+    @tooslow
+    def test_inference_masked_lm(self):
+        texts = "今天天气真好！"
+        model_path = "openbmb/cpm-ant-10b"
+        model = CpmAntModel.from_pretrained(model_path)
+        tokenizer = CpmAntTokenizer.from_pretrained(model_path)
+        inputs = tokenizer(texts, return_tensors="pt")
+        hidden_states = model(**inputs).last_hidden_state
+
+        expected_slice = torch.tensor(
+            [[[6.1708, 5.9244, 1.0835], [6.5207, 6.2893, -11.3324], [-1.0107, -0.0576, -5.9577]]],
+        )
+        self.assertTrue(torch.allclose(hidden_states[:, :3, :3], expected_slice, atol=1e-2))
+
+
+@require_torch
+class CpmAntForCausalLMlIntegrationTest(unittest.TestCase):
+    @tooslow
+    def test_inference_casual(self):
+        texts = "今天天气真好！"
+        model_path = "openbmb/cpm-ant-10b"
+        model = CpmAntForCausalLM.from_pretrained(model_path)
+        tokenizer = CpmAntTokenizer.from_pretrained(model_path)
+        inputs = tokenizer(texts, return_tensors="pt")
+        hidden_states = model(**inputs).logits
+
+        expected_slice = torch.tensor(
+            [[[-6.4267, -6.4083, -6.3958], [-5.8802, -5.9447, -5.7811], [-5.3896, -5.4820, -5.4295]]],
+        )
+        self.assertTrue(torch.allclose(hidden_states[:, :3, :3], expected_slice, atol=1e-2))
+
+    @tooslow
+    def test_simple_generation(self):
+        model_path = "openbmb/cpm-ant-10b"
+        model = CpmAntForCausalLM.from_pretrained(model_path)
+        tokenizer = CpmAntTokenizer.from_pretrained(model_path)
+        texts = "今天天气不错，"
+        expected_output = "今天天气不错，阳光明媚，我和妈妈一起去超市买东西。\n在超市里，我看到了一个很好玩的玩具，它的名字叫“机器人”。它有一个圆圆的脑袋，两只圆圆的眼睛，还有一个圆圆的"
+        model_inputs = tokenizer(texts, return_tensors="pt")
+        token_ids = model.generate(**model_inputs)
+        output_texts = tokenizer.batch_decode(token_ids)
+        self.assertEqual(expected_output, output_texts)
+
+    @tooslow
+    def test_batch_generation(self):
+        model_path = "openbmb/cpm-ant-10b"
+        model = CpmAntForCausalLM.from_pretrained(model_path)
+        tokenizer = CpmAntTokenizer.from_pretrained(model_path)
+        texts = ["今天天气不错，", "新年快乐，万事如意！"]
+        expected_output = [
+            "今天天气不错，阳光明媚，我和妈妈一起去超市买东西。\n在超市里，我看到了一个很好玩的玩具，它的名字叫“机器人”。它有一个圆圆的脑袋，两只圆圆的眼睛，还有一个圆圆的",
+            "新年快乐，万事如意！在这辞旧迎新的美好时刻，我谨代表《农村新技术》杂志社全体同仁，向一直以来关心、支持《农村新技术》杂志发展的各级领导、各界朋友和广大读者致以最诚挚的",
+        ]
+        model_inputs = tokenizer(texts, return_tensors="pt", padding=True)
+        token_ids = model.generate(**model_inputs)
+        output_texts = tokenizer.batch_decode(token_ids)
+        self.assertEqual(expected_output, output_texts)

tests/models/cpmant/test_tokenization_cpmant.py

+# coding=utf-8
+# Copyright 2022 The OpenBMB Team and The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import unittest
+
+from transformers.models.cpmant.tokenization_cpmant import VOCAB_FILES_NAMES, CpmAntTokenizer
+from transformers.testing_utils import require_jieba, tooslow
+
+from ...test_tokenization_common import TokenizerTesterMixin
+
+
+@require_jieba
+class CPMAntTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
+    tokenizer_class = CpmAntTokenizer
+    test_rust_tokenizer = False
+
+    def setUp(self):
+        super().setUp()
+
+        vocab_tokens = [
+            "<d>",
+            "</d>",
+            "<s>",
+            "</s>",
+            "</_>",
+            "<unk>",
+            "<pad>",
+            "</n>",
+            "我",
+            "是",
+            "C",
+            "P",
+            "M",
+            "A",
+            "n",
+            "t",
+        ]
+        self.vocab_file = os.path.join(self.tmpdirname, VOCAB_FILES_NAMES["vocab_file"])
+        with open(self.vocab_file, "w", encoding="utf-8") as vocab_writer:
+            vocab_writer.write("".join([x + "\n" for x in vocab_tokens]))
+
+    @tooslow
+    def test_pre_tokenization(self):
+        tokenizer = CpmAntTokenizer.from_pretrained("openbmb/cpm-ant-10b")
+        texts = "今天天气真好！"
+        jieba_tokens = ["今天", "天气", "真", "好", "！"]
+        tokens = tokenizer.tokenize(texts)
+        self.assertListEqual(tokens, jieba_tokens)
+        normalized_text = "今天天气真好！"
+        input_tokens = [tokenizer.bos_token] + tokens
+
+        input_jieba_tokens = [6, 9802, 14962, 2082, 831, 244]
+        self.assertListEqual(tokenizer.convert_tokens_to_ids(input_tokens), input_jieba_tokens)
+
+        reconstructed_text = tokenizer.decode(input_jieba_tokens)
+        self.assertEqual(reconstructed_text, normalized_text)