Release: v4.2.2

* fix tf led

* remove loop file

.github/workflows/release-conda.yml

@@ -37,7 +37,8 @@ jobs:
       - name: Build conda packages
         run: |
           conda info
-          conda build .github/conda
+          conda list
+          conda-build .github/conda
 
       - name: Upload to Anaconda
-        run: anaconda upload `conda build .github/conda --output` --force
+        run: anaconda upload `conda-build .github/conda --output` --force

setup.py

@@ -248,7 +248,7 @@ install_requires = [
 
 setup(
     name="transformers",
-    version="4.2.1", # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    version="4.2.2", # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
     author="Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Sam Shleifer, Patrick von Platen, Sylvain Gugger, Google AI Language Team Authors, Open AI team Authors, Facebook AI Authors, Carnegie Mellon University Authors",
     author_email="thomas@huggingface.co",
     description="State-of-the-art Natural Language Processing for TensorFlow 2.0 and PyTorch",

src/transformers/__init__.py

@@ -22,7 +22,7 @@
 # to defer the actual importing for when the objects are requested. This way `import transformers` provides the names
 # in the namespace without actually importing anything (and especially none of the backends).
 
-__version__ = "4.2.1"
+__version__ = "4.2.2"
 
 # Work around to update TensorFlow's absl.logging threshold which alters the
 # default Python logging output behavior when present.

src/transformers/models/albert/convert_albert_original_tf_checkpoint_to_pytorch.py

@@ -19,8 +19,8 @@ import argparse
 
 import torch
 
-from ...utils import logging
-from . import AlbertConfig, AlbertForPreTraining, load_tf_weights_in_albert
+from transformers import AlbertConfig, AlbertForPreTraining, load_tf_weights_in_albert
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/bart/convert_bart_original_pytorch_checkpoint_to_pytorch.py

@@ -23,9 +23,15 @@ import fairseq
 import torch
 from packaging import version
 
-from ...utils import logging
-from . import BartConfig, BartForConditionalGeneration, BartForSequenceClassification, BartModel, BartTokenizer
-from .modeling_bart import _make_linear_from_emb
+from transformers import (
+    BartConfig,
+    BartForConditionalGeneration,
+    BartForSequenceClassification,
+    BartModel,
+    BartTokenizer,
+)
+from transformers.models.bart.modeling_bart import _make_linear_from_emb
+from transformers.utils import logging
 
 
 FAIRSEQ_MODELS = ["bart.large", "bart.large.mnli", "bart.large.cnn", "bart_xsum/model.pt"]

src/transformers/models/bert/convert_bert_original_tf2_checkpoint_to_pytorch.py

@@ -28,8 +28,8 @@ import re
 import tensorflow as tf
 import torch
 
-from ...utils import logging
-from . import BertConfig, BertModel
+from transformers import BertConfig, BertModel
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/bert/convert_bert_original_tf_checkpoint_to_pytorch.py

@@ -19,8 +19,8 @@ import argparse
 
 import torch
 
-from ...utils import logging
-from . import BertConfig, BertForPreTraining, load_tf_weights_in_bert
+from transformers import BertConfig, BertForPreTraining, load_tf_weights_in_bert
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/bert/convert_bert_pytorch_checkpoint_to_original_tf.py

@@ -22,7 +22,7 @@ import numpy as np
 import tensorflow as tf
 import torch
 
-from . import BertModel
+from transformers import BertModel
 
 
 def convert_pytorch_checkpoint_to_tf(model: BertModel, ckpt_dir: str, model_name: str):

src/transformers/models/blenderbot/convert_blenderbot_original_pytorch_checkpoint_to_pytorch.py

@@ -18,8 +18,8 @@ import argparse
 
 import torch
 
-from ...models.bart import BartConfig, BartForConditionalGeneration
-from ...utils import logging
+from transformers import BartConfig, BartForConditionalGeneration
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/dialogpt/convert_dialogpt_original_pytorch_checkpoint_to_pytorch.py

@@ -17,7 +17,7 @@ import os
 
 import torch
 
-from ...file_utils import WEIGHTS_NAME
+from transformers.file_utils import WEIGHTS_NAME
 
 
 DIALOGPT_MODELS = ["small", "medium", "large"]

src/transformers/models/dpr/convert_dpr_original_checkpoint_to_pytorch.py

@@ -19,8 +19,7 @@ from pathlib import Path
 import torch
 from torch.serialization import default_restore_location
 
-from ...models.bert import BertConfig
-from . import DPRConfig, DPRContextEncoder, DPRQuestionEncoder, DPRReader
+from .transformers import BertConfig, DPRConfig, DPRContextEncoder, DPRQuestionEncoder, DPRReader
 
 
 CheckpointState = collections.namedtuple(

src/transformers/models/electra/convert_electra_original_tf_checkpoint_to_pytorch.py

@@ -19,8 +19,8 @@ import argparse
 
 import torch
 
-from ...utils import logging
-from . import ElectraConfig, ElectraForMaskedLM, ElectraForPreTraining, load_tf_weights_in_electra
+from transformers import ElectraConfig, ElectraForMaskedLM, ElectraForPreTraining, load_tf_weights_in_electra
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/fsmt/convert_fsmt_original_pytorch_checkpoint_to_pytorch.py

@@ -31,10 +31,11 @@ import torch
 from fairseq import hub_utils
 from fairseq.data.dictionary import Dictionary
 
-from ...file_utils import WEIGHTS_NAME
-from ...tokenization_utils_base import TOKENIZER_CONFIG_FILE
-from ...utils import logging
-from . import VOCAB_FILES_NAMES, FSMTConfig, FSMTForConditionalGeneration
+from transfomers.models.fsmt.tokenization_fsmt import VOCAB_FILES_NAMES
+from transformers import FSMTConfig, FSMTForConditionalGeneration
+from transformers.file_utils import WEIGHTS_NAME
+from transformers.tokenization_utils_base import TOKENIZER_CONFIG_FILE
+from transformers.utils import logging
 
 
 logging.set_verbosity_warning()

src/transformers/models/funnel/__init__.py

@@ -23,6 +23,7 @@ from ...file_utils import _BaseLazyModule, is_tf_available, is_tokenizers_availa
 
 _import_structure = {
     "configuration_funnel": ["FUNNEL_PRETRAINED_CONFIG_ARCHIVE_MAP", "FunnelConfig"],
+    "convert_funnel_original_tf_checkpoint_to_pytorch": [],
     "tokenization_funnel": ["FunnelTokenizer"],
 }
 

src/transformers/models/funnel/convert_funnel_original_tf_checkpoint_to_pytorch.py

@@ -16,14 +16,14 @@
 
 
 import argparse
-import logging
 
 import torch
 
-from . import FunnelConfig, FunnelForPreTraining, load_tf_weights_in_funnel
+from transformers import FunnelConfig, FunnelForPreTraining, load_tf_weights_in_funnel
+from transformers.utils import logging
 
 
-logging.basicConfig(level=logging.INFO)
+logging.set_verbosity_info()
 
 
 def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, config_file, pytorch_dump_path):

src/transformers/models/gpt2/convert_gpt2_original_tf_checkpoint_to_pytorch.py

@@ -19,9 +19,9 @@ import argparse
 
 import torch
 
-from ...file_utils import CONFIG_NAME, WEIGHTS_NAME
-from ...utils import logging
-from . import GPT2Config, GPT2Model, load_tf_weights_in_gpt2
+from transformers import GPT2Config, GPT2Model, load_tf_weights_in_gpt2
+from transformers.file_utils import CONFIG_NAME, WEIGHTS_NAME
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/led/modeling_tf_led.py

@@ -1862,7 +1862,6 @@ class TFLEDDecoder(tf.keras.layers.Layer):
         hidden_states = inputs["inputs_embeds"]
 
         # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
-        combined_attention_mask = None
         if input_shape[-1] > 1:
             combined_attention_mask = _make_causal_mask(input_shape, past_key_values_length=past_key_values_length)
         else:
@@ -1870,20 +1869,9 @@ class TFLEDDecoder(tf.keras.layers.Layer):
                 tf.ones((input_shape[0], input_shape[1] + past_key_values_length)), tgt_len=input_shape[-1]
             )
 
-        if inputs["attention_mask"] is None and inputs["input_ids"] is not None and input_shape[-1] > 1:
-            inputs["attention_mask"] = tf.cast(
-                tf.math.not_equal(inputs["input_ids"], self.config.pad_token_id), inputs["input_ids"].dtype
-            )
-            inputs["attention_mask"] = tf.concat(
-                [
-                    tf.ones((input_shape[0], past_key_values_length), dtype=inputs["attention_mask"].dtype),
-                    inputs["attention_mask"],
-                ],
-                axis=-1,
-            )
-        else:
-            inputs["attention_mask"] = tf.ones(
-                (input_shape[0], input_shape[1] + past_key_values_length), dtype=tf.int32
+        if inputs["attention_mask"] is not None and input_shape[-1] > 1:
+            combined_attention_mask = combined_attention_mask + _expand_mask(
+                inputs["attention_mask"], tgt_len=input_shape[-1]
             )
 
         if inputs["encoder_hidden_states"] is not None and inputs["encoder_attention_mask"] is not None:

src/transformers/models/longformer/convert_longformer_original_pytorch_lightning_to_pytorch.py

@@ -20,7 +20,7 @@ import argparse
 import pytorch_lightning as pl
 import torch
 
-from . import LongformerForQuestionAnswering, LongformerModel
+from transformers import LongformerForQuestionAnswering, LongformerModel
 
 
 class LightningModel(pl.LightningModule):

src/transformers/models/lxmert/convert_lxmert_original_tf_checkpoint_to_pytorch.py

@@ -16,14 +16,14 @@
 
 
 import argparse
-import logging
 
 import torch
 
-from . import LxmertConfig, LxmertForPreTraining, load_tf_weights_in_lxmert
+from transformers import LxmertConfig, LxmertForPreTraining, load_tf_weights_in_lxmert
+from transformers.utils import logging
 
 
-logging.basicConfig(level=logging.INFO)
+logging.set_verbosity_info()
 
 
 def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, config_file, pytorch_dump_path):

src/transformers/models/marian/convert_marian_tatoeba_to_pytorch.py

@@ -17,7 +17,7 @@ import os
 from pathlib import Path
 from typing import List, Tuple
 
-from .convert_marian_to_pytorch import (
+from transformers.models.marian.convert_marian_to_pytorch import (
     FRONT_MATTER_TEMPLATE,
     _parse_readme,
     convert_all_sentencepiece_models,

src/transformers/models/marian/convert_marian_to_pytorch.py

@@ -26,8 +26,8 @@ import numpy as np
 import torch
 from tqdm import tqdm
 
-from ...hf_api import HfApi
-from . import MarianConfig, MarianMTModel, MarianTokenizer
+from transformers import MarianConfig, MarianMTModel, MarianTokenizer
+from transformers.hf_api import HfApi
 
 
 def remove_suffix(text: str, suffix: str):

src/transformers/models/mbart/convert_mbart_original_checkpoint_to_pytorch.py

@@ -16,9 +16,8 @@ import argparse
 
 import torch
 
-from ..bart import BartForConditionalGeneration
-from ..bart.convert_bart_original_pytorch_checkpoint_to_pytorch import remove_ignore_keys_
-from . import MBartConfig
+from transformers import BartForConditionalGeneration, MBartConfig
+from transformers.models.bart.convert_bart_original_pytorch_checkpoint_to_pytorch import remove_ignore_keys_
 
 
 def convert_fairseq_mbart_checkpoint_from_disk(checkpoint_path, hf_config_path="facebook/mbart-large-en-ro"):

src/transformers/models/mobilebert/convert_mobilebert_original_tf_checkpoint_to_pytorch.py

@@ -16,8 +16,8 @@ import argparse
 
 import torch
 
-from ...utils import logging
-from . import MobileBertConfig, MobileBertForPreTraining, load_tf_weights_in_mobilebert
+from transformers import MobileBertConfig, MobileBertForPreTraining, load_tf_weights_in_mobilebert
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/openai/convert_openai_original_tf_checkpoint_to_pytorch.py

@@ -19,9 +19,9 @@ import argparse
 
 import torch
 
-from ...file_utils import CONFIG_NAME, WEIGHTS_NAME
-from ...utils import logging
-from . import OpenAIGPTConfig, OpenAIGPTModel, load_tf_weights_in_openai_gpt
+from transformers import OpenAIGPTConfig, OpenAIGPTModel, load_tf_weights_in_openai_gpt
+from transformers.file_utils import CONFIG_NAME, WEIGHTS_NAME
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/pegasus/convert_pegasus_tf_to_pytorch.py

@@ -22,8 +22,8 @@ import tensorflow as tf
 import torch
 from tqdm import tqdm
 
-from . import PegasusConfig, PegasusForConditionalGeneration, PegasusTokenizer
-from .configuration_pegasus import DEFAULTS, task_specific_params
+from transformers import PegasusConfig, PegasusForConditionalGeneration, PegasusTokenizer
+from transformers.models.pegasus.configuration_pegasus import DEFAULTS, task_specific_params
 
 
 PATTERNS = [

src/transformers/models/prophetnet/convert_prophetnet_original_pytorch_checkpoint_to_pytorch.py

@@ -19,6 +19,8 @@ import argparse
 
 import torch
 
+from transformers import ProphetNetForConditionalGeneration, XLMProphetNetForConditionalGeneration, logging
+
 # transformers_old should correspond to branch `save_old_prophetnet_model_structure` here
 # original prophetnet_checkpoints are saved under `patrickvonplaten/..._old` respectively
 from transformers_old.modeling_prophetnet import (
@@ -28,8 +30,6 @@ from transformers_old.modeling_xlm_prophetnet import (
     XLMProphetNetForConditionalGeneration as XLMProphetNetForConditionalGenerationOld,
 )
 
-from . import ProphetNetForConditionalGeneration, XLMProphetNetForConditionalGeneration, logging
-
 
 logger = logging.get_logger(__name__)
 logging.set_verbosity_info()

src/transformers/models/reformer/convert_reformer_trax_checkpoint_to_pytorch.py

@@ -21,8 +21,8 @@ import pickle
 import numpy as np
 import torch
 
-from ...utils import logging
-from . import ReformerConfig, ReformerModelWithLMHead
+from transformers import ReformerConfig, ReformerModelWithLMHead
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/roberta/convert_roberta_original_pytorch_checkpoint_to_pytorch.py

@@ -24,9 +24,15 @@ from fairseq.models.roberta import RobertaModel as FairseqRobertaModel
 from fairseq.modules import TransformerSentenceEncoderLayer
 from packaging import version
 
-from ...models.bert.modeling_bert import BertIntermediate, BertLayer, BertOutput, BertSelfAttention, BertSelfOutput
-from ...utils import logging
-from .modeling_roberta import RobertaConfig, RobertaForMaskedLM, RobertaForSequenceClassification
+from transformers import RobertaConfig, RobertaForMaskedLM, RobertaForSequenceClassification
+from transformers.models.bert.modeling_bert import (
+    BertIntermediate,
+    BertLayer,
+    BertOutput,
+    BertSelfAttention,
+    BertSelfOutput,
+)
+from transformers.utils import logging
 
 
 if version.parse(fairseq.__version__) < version.parse("0.9.0"):

src/transformers/models/t5/convert_t5_original_tf_checkpoint_to_pytorch.py

@@ -17,8 +17,8 @@
 
 import argparse
 
-from ...utils import logging
-from . import T5Config, T5ForConditionalGeneration, load_tf_weights_in_t5
+from transformers import T5Config, T5ForConditionalGeneration, load_tf_weights_in_t5
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/tapas/convert_tapas_original_tf_checkpoint_to_pytorch.py

@@ -17,8 +17,7 @@
 
 import argparse
 
-from ...utils import logging
-from . import (
+from transformers import (
     TapasConfig,
     TapasForMaskedLM,
     TapasForQuestionAnswering,
@@ -27,6 +26,7 @@ from . import (
     TapasTokenizer,
     load_tf_weights_in_tapas,
 )
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/transfo_xl/convert_transfo_xl_original_tf_checkpoint_to_pytorch.py

@@ -22,11 +22,11 @@ import sys
 
 import torch
 
-from ...file_utils import CONFIG_NAME, WEIGHTS_NAME
-from ...utils import logging
-from . import TransfoXLConfig, TransfoXLLMHeadModel, load_tf_weights_in_transfo_xl
-from . import tokenization_transfo_xl as data_utils
-from .tokenization_transfo_xl import CORPUS_NAME, VOCAB_FILES_NAMES
+from transformers import TransfoXLConfig, TransfoXLLMHeadModel, load_tf_weights_in_transfo_xl
+from transformers.file_utils import CONFIG_NAME, WEIGHTS_NAME
+from transformers.models.transfo_xl import tokenization_transfo_xl as data_utils
+from transformers.models.transfo_xl.tokenization_transfo_xl import CORPUS_NAME, VOCAB_FILES_NAMES
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/xlm/convert_xlm_original_pytorch_checkpoint_to_pytorch.py

@@ -21,9 +21,9 @@ import json
 import numpy
 import torch
 
-from ...file_utils import CONFIG_NAME, WEIGHTS_NAME
-from ...utils import logging
-from .tokenization_xlm import VOCAB_FILES_NAMES
+from transformers.file_utils import CONFIG_NAME, WEIGHTS_NAME
+from transformers.models.xlm.tokenization_xlm import VOCAB_FILES_NAMES
+from transformers.utils import logging
 
 
 logging.set_verbosity_info()

src/transformers/models/xlnet/convert_xlnet_original_tf_checkpoint_to_pytorch.py

@@ -20,15 +20,15 @@ import os
 
 import torch
 
-from ...file_utils import CONFIG_NAME, WEIGHTS_NAME
-from ...utils import logging
-from . import (
+from transformers import (
     XLNetConfig,
     XLNetForQuestionAnswering,
     XLNetForSequenceClassification,
     XLNetLMHeadModel,
     load_tf_weights_in_xlnet,
 )
+from transformers.file_utils import CONFIG_NAME, WEIGHTS_NAME
+from transformers.utils import logging
 
 
 GLUE_TASKS_NUM_LABELS = {