Add OPT

docs/source/en/_toctree.yml

@@ -268,6 +268,8 @@
       title: Nyströmformer
     - local: model_doc/openai-gpt
       title: OpenAI GPT
+    - local: model_doc/opt
+      title: OPT
     - local: model_doc/gpt2
       title: OpenAI GPT2
     - local: model_doc/gptj

docs/source/en/index.mdx

@@ -229,6 +229,7 @@ Flax), PyTorch, and/or TensorFlow.
 |        Nystromformer        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |         OpenAI GPT          |       ✅       |       ✅       |       ✅        |         ✅         |      ❌      |
 |        OpenAI GPT-2         |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
+|             OPT             |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           Pegasus           |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |          Perceiver          |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           PLBart            |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |

docs/source/en/model_doc/opt.mdx

@@ -0,0 +1,47 @@
+<!--Copyright 2022 The HuggingFace 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.
+-->
+
+# OPT
+
+## Overview
+
+The OPT model was proposed in [Open Pre-trained Transformer Language Models](https://arxiv.org/pdf/2205.01068) by Meta AI.
+OPT is a series of open-sourced large causal language models which perform similar in performance to GPT3. 
+
+
+The abstract from the paper is the following:
+
+*Large language models, which are often trained for hundreds of thousands of compute days, have shown remarkable capabilities for zero- and few-shot learning. Given their computational cost, these models are difficult to replicate without significant capital. For the few that are available through APIs, no access is granted to the full model weights, making them difficult to study. We present Open Pre-trained Transformers (OPT), a suite of decoder-only pre-trained transformers ranging from 125M to 175B parameters, which we aim to fully and responsibly share with interested researchers. We show that OPT-175B is comparable to GPT-3, while requiring only 1/7th the carbon footprint to develop. We are also releasing our logbook detailing the infrastructure challenges we faced, along with code for experimenting with all of the released models.*
+
+Tips:
+
+If you want to use the `opt-350m`, you need to set the `do_layer_norm_before` parameter to `False`. 
+
+This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ), [Younes Belkada](https://huggingface.co/ybelkada), and [Patrick Von Platen](https://huggingface.co/patrickvonplaten).
+The original code can be found [here](https://github.com/facebookresearch/metaseq).
+
+
+## OPTConfig
+
+[[autodoc]] OPTConfig
+
+## OPTModel
+
+[[autodoc]] OPTModel
+    - forward
+
+
+## OPTForCausalLM
+
+[[autodoc]] OPTForCausalLM
+    - forward
+

src/transformers/__init__.py

@@ -239,6 +239,7 @@
         "NystromformerConfig",
     ],
     "models.openai": ["OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP", "OpenAIGPTConfig", "OpenAIGPTTokenizer"],
+    "models.opt": ["OPTConfig"],
     "models.pegasus": ["PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP", "PegasusConfig", "PegasusTokenizer"],
     "models.perceiver": ["PERCEIVER_PRETRAINED_CONFIG_ARCHIVE_MAP", "PerceiverConfig", "PerceiverTokenizer"],
     "models.phobert": ["PhobertTokenizer"],
@@ -1301,6 +1302,14 @@
             "load_tf_weights_in_openai_gpt",
         ]
     )
+    _import_structure["models.opt"].extend(
+        [
+            "OPT_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "OPTForCausalLM",
+            "OPTModel",
+            "OPTPretrainedModel",
+        ]
+    )
     _import_structure["models.pegasus"].extend(
         ["PegasusForCausalLM", "PegasusForConditionalGeneration", "PegasusModel", "PegasusPreTrainedModel"]
     )
@@ -2351,7 +2360,6 @@
             "FlaxBartPreTrainedModel",
         ]
     )
-
     _import_structure["models.beit"].extend(
         [
             "FlaxBeitForImageClassification",
@@ -2360,6 +2368,7 @@
             "FlaxBeitPreTrainedModel",
         ]
     )
+
     _import_structure["models.bert"].extend(
         [
             "FlaxBertForCausalLM",
@@ -2688,6 +2697,7 @@
     from .models.mt5 import MT5Config
     from .models.nystromformer import NYSTROMFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP, NystromformerConfig
     from .models.openai import OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP, OpenAIGPTConfig, OpenAIGPTTokenizer
+    from .models.opt import OPTConfig
     from .models.pegasus import PEGASUS_PRETRAINED_CONFIG_ARCHIVE_MAP, PegasusConfig, PegasusTokenizer
     from .models.perceiver import PERCEIVER_PRETRAINED_CONFIG_ARCHIVE_MAP, PerceiverConfig, PerceiverTokenizer
     from .models.phobert import PhobertTokenizer
@@ -3588,6 +3598,7 @@
             OpenAIGPTPreTrainedModel,
             load_tf_weights_in_openai_gpt,
         )
+        from .models.opt import OPT_PRETRAINED_MODEL_ARCHIVE_LIST, OPTForCausalLM, OPTModel, OPTPretrainedModel
         from .models.pegasus import (
             PegasusForCausalLM,
             PegasusForConditionalGeneration,

src/transformers/models/__init__.py

@@ -86,6 +86,7 @@
     mt5,
     nystromformer,
     openai,
+    opt,
     pegasus,
     perceiver,
     phobert,

src/transformers/models/auto/configuration_auto.py

@@ -97,6 +97,7 @@
         ("megatron-bert", "MegatronBertConfig"),
         ("mpnet", "MPNetConfig"),
         ("bart", "BartConfig"),
+        ("opt", "OPTConfig"),
         ("blenderbot", "BlenderbotConfig"),
         ("reformer", "ReformerConfig"),
         ("longformer", "LongformerConfig"),
@@ -188,6 +189,7 @@
         ("blenderbot-small", "BLENDERBOT_SMALL_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("bert", "BERT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("bart", "BART_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("opt", "OPT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("blenderbot", "BLENDERBOT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("mbart", "MBART_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("openai-gpt", "OPENAI_GPT_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -298,6 +300,7 @@
         ("mbart", "mBART"),
         ("megatron-bert", "MegatronBert"),
         ("bart", "BART"),
+        ("opt", "OPT"),
         ("reformer", "Reformer"),
         ("longformer", "Longformer"),
         ("roberta", "RoBERTa"),

src/transformers/models/auto/modeling_auto.py

@@ -92,6 +92,7 @@
         ("xlm-roberta-xl", "XLMRobertaXLModel"),
         ("xlm-roberta", "XLMRobertaModel"),
         ("bart", "BartModel"),
+        ("opt", "OPTModel"),
         ("longformer", "LongformerModel"),
         ("roberta", "RobertaModel"),
         ("data2vec-text", "Data2VecTextModel"),
@@ -259,6 +260,7 @@
         ("xlm-prophetnet", "XLMProphetNetForCausalLM"),
         ("prophetnet", "ProphetNetForCausalLM"),
         ("bart", "BartForCausalLM"),
+        ("opt", "OPTForCausalLM"),
         ("mbart", "MBartForCausalLM"),
         ("pegasus", "PegasusForCausalLM"),
         ("marian", "MarianForCausalLM"),

src/transformers/models/auto/tokenization_auto.py

@@ -107,6 +107,7 @@
             ("blenderbot", ("BlenderbotTokenizer", "BlenderbotTokenizerFast")),
             ("tapex", ("TapexTokenizer", None)),
             ("bart", ("BartTokenizer", "BartTokenizerFast")),
+            ("opt", ("GPT2Tokenizer", None)),
             ("longformer", ("LongformerTokenizer", "LongformerTokenizerFast" if is_tokenizers_available() else None)),
             ("roberta", ("RobertaTokenizer", "RobertaTokenizerFast" if is_tokenizers_available() else None)),
             (

src/transformers/models/gpt2/tokenization_gpt2.py

@@ -162,20 +162,26 @@ def __init__(
         unk_token="<|endoftext|>",
         bos_token="<|endoftext|>",
         eos_token="<|endoftext|>",
+        pad_token=None,
         add_prefix_space=False,
+        add_bos_token=False,
         **kwargs
     ):
         bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
         eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
         unk_token = AddedToken(unk_token, lstrip=False, rstrip=False) if isinstance(unk_token, str) else unk_token
+        pad_token = AddedToken(pad_token, lstrip=False, rstrip=False) if isinstance(pad_token, str) else pad_token
         super().__init__(
             errors=errors,
             unk_token=unk_token,
             bos_token=bos_token,
             eos_token=eos_token,
+            pad_token=pad_token,
             add_prefix_space=add_prefix_space,
+            add_bos_token=add_bos_token,
             **kwargs,
         )
+        self.add_bos_token = add_bos_token
 
         with open(vocab_file, encoding="utf-8") as vocab_handle:
             self.encoder = json.load(vocab_handle)
@@ -242,6 +248,19 @@ def bpe(self, token):
         self.cache[token] = word
         return word
 
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        if self.add_bos_token:
+            bos_token_ids = [self.bos_token_id]
+        else:
+            bos_token_ids = []
+
+        output = bos_token_ids + token_ids_0
+
+        if token_ids_1 is None:
+            return output
+
+        return output + bos_token_ids + token_ids_1
+
     def _tokenize(self, text):
         """Tokenize a string."""
         bpe_tokens = []

src/transformers/models/opt/__init__.py

@@ -0,0 +1,46 @@
+# flake8: noqa
+# There's no way to ignore "F401 '...' imported but unused" warnings in this
+# module, but to preserve other warnings. So, don't check this module at all.
+
+# Copyright 2022 The HuggingFace 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.
+from typing import TYPE_CHECKING
+
+from ...utils import _LazyModule, is_tokenizers_available, is_torch_available
+
+
+_import_structure = {
+    "configuration_opt": ["OPT_PRETRAINED_CONFIG_ARCHIVE_MAP", "OPTConfig"],
+}
+
+
+if is_torch_available():
+    _import_structure["modeling_opt"] = [
+        "OPT_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "OPTForCausalLM",
+        "OPTModel",
+        "OPTPretrainedModel",
+    ]
+
+
+if TYPE_CHECKING:
+    from .configuration_opt import OPT_PRETRAINED_CONFIG_ARCHIVE_MAP, OPTConfig
+
+    if is_torch_available():
+        from .modeling_opt import OPT_PRETRAINED_MODEL_ARCHIVE_LIST, OPTForCausalLM, OPTModel, OPTPretrainedModel
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)