Add WhisperModel to transformers (#19166)

* simplify loop

* add featur extractor

* add model

* start conversion

* add dropout

* initial commit of test files

* copnversion for all models

* update processor for correct padding

* update feature extraction

* update integration test logits match

* fmnt: off for the logits

* on the fly mel bank

* small nit

* update test

* update tokenizer

* nit feature extraction

* update

* update tokenizer test

* adds logit processor and update tokenizer to get supress tokens

* style

* clean convert

* revert to original modeling tf utils

* Update

* update

* nit

* clean convert file

* update tests and nits

* quality

* slow generation test

* ffn_dim to allow customization

* update readme

* add to toctreee

* start fixing integration tests

* update tests and code

* fix feature extractor

* fix config tests common

* update code to fix tests

* fix feature exctractor

* nit...

README.md

@@ -393,6 +393,7 @@ Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://h
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
 1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
 1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[Whisper](https://huggingface.co/docs/transformers/main/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
 1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
 1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
 1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.

README_ko.md

@@ -343,6 +343,7 @@ Flax, PyTorch, TensorFlow 설치 페이지에서 이들을 conda로 설치하는
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
 1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
 1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[Whisper](https://huggingface.co/docs/transformers/main/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
 1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
 1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li. 
 1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.

README_zh-hans.md

@@ -367,6 +367,7 @@ conda install -c huggingface transformers
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (来自 Facebook AI) 伴随论文 [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) 由 Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino 发布。
 1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (来自 Facebook AI) 伴随论文 [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) 由 Qiantong Xu, Alexei Baevski, Michael Auli 发布。
 1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[Whisper](https://huggingface.co/docs/transformers/main/model_doc/whisper)** (来自 OpenAI) 伴随论文 [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) 由 Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever 发布。
 1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (来自 Microsoft Research) 伴随论文 [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) 由 Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling 发布。
 1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li. 
 1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (来自 Facebook) 伴随论文 [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) 由 Guillaume Lample and Alexis Conneau 发布。

README_zh-hant.md

@@ -379,6 +379,7 @@ conda install -c huggingface transformers
 1. **[Wav2Vec2-Conformer](https://huggingface.co/docs/transformers/model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
 1. **[Wav2Vec2Phoneme](https://huggingface.co/docs/transformers/model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
 1. **[WavLM](https://huggingface.co/docs/transformers/model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[Whisper](https://huggingface.co/docs/transformers/main/model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
 1. **[X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
 1. **[XGLM](https://huggingface.co/docs/transformers/model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li. 
 1. **[XLM](https://huggingface.co/docs/transformers/model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.

docs/source/en/_toctree.yml

@@ -449,6 +449,8 @@
         title: Wav2Vec2Phoneme
       - local: model_doc/wavlm
         title: WavLM
+      - local: model_doc/whisper
+        title: Whisper
       - local: model_doc/xls_r
         title: XLS-R
       - local: model_doc/xlsr_wav2vec2

docs/source/en/index.mdx

@@ -183,6 +183,7 @@ The documentation is organized into five sections:
 1. **[Wav2Vec2-Conformer](model_doc/wav2vec2-conformer)** (from Facebook AI) released with the paper [FAIRSEQ S2T: Fast Speech-to-Text Modeling with FAIRSEQ](https://arxiv.org/abs/2010.05171) by Changhan Wang, Yun Tang, Xutai Ma, Anne Wu, Sravya Popuri, Dmytro Okhonko, Juan Pino.
 1. **[Wav2Vec2Phoneme](model_doc/wav2vec2_phoneme)** (from Facebook AI) released with the paper [Simple and Effective Zero-shot Cross-lingual Phoneme Recognition](https://arxiv.org/abs/2109.11680) by Qiantong Xu, Alexei Baevski, Michael Auli.
 1. **[WavLM](model_doc/wavlm)** (from Microsoft Research) released with the paper [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://arxiv.org/abs/2110.13900) by Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Furu Wei.
+1. **[Whisper](model_doc/whisper)** (from OpenAI) released with the paper [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
 1. **[X-CLIP](model_doc/xclip)** (from Microsoft Research) released with the paper [Expanding Language-Image Pretrained Models for General Video Recognition](https://arxiv.org/abs/2208.02816) by Bolin Ni, Houwen Peng, Minghao Chen, Songyang Zhang, Gaofeng Meng, Jianlong Fu, Shiming Xiang, Haibin Ling.
 1. **[XGLM](model_doc/xglm)** (From Facebook AI) released with the paper [Few-shot Learning with Multilingual Language Models](https://arxiv.org/abs/2112.10668) by Xi Victoria Lin, Todor Mihaylov, Mikel Artetxe, Tianlu Wang, Shuohui Chen, Daniel Simig, Myle Ott, Naman Goyal, Shruti Bhosale, Jingfei Du, Ramakanth Pasunuru, Sam Shleifer, Punit Singh Koura, Vishrav Chaudhary, Brian O'Horo, Jeff Wang, Luke Zettlemoyer, Zornitsa Kozareva, Mona Diab, Veselin Stoyanov, Xian Li.
 1. **[XLM](model_doc/xlm)** (from Facebook) released together with the paper [Cross-lingual Language Model Pretraining](https://arxiv.org/abs/1901.07291) by Guillaume Lample and Alexis Conneau.
@@ -329,6 +330,7 @@ Flax), PyTorch, and/or TensorFlow.
 |          Wav2Vec2           |       ✅       |       ❌       |       ✅        |         ✅         |      ✅      |
 |     Wav2Vec2-Conformer      |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |            WavLM            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
+|           Whisper           |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 |           X-CLIP            |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 |            XGLM             |       ✅       |       ✅       |       ✅        |         ✅         |      ✅      |
 |             XLM             |       ✅       |       ❌       |       ✅        |         ✅         |      ❌      |

docs/source/en/model_doc/whisper.mdx

+<!--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.
+-->
+
+# Whisper
+
+## Overview
+
+The Whisper model was proposed in [Robust Speech Recognition via Large-Scale Weak Supervision](https://cdn.openai.com/papers/whisper.pdf) by Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, Ilya Sutskever.
+
+The abstract from the paper is the following:
+
+*We study the capabilities of speech processing systems trained simply to predict large amounts of transcripts of audio on the internet. When scaled to 680,000 hours of multilingual and multitask supervision, the resulting models generalize well to standard benchmarks and are often competitive with prior fully supervised results but in a zeroshot transfer setting without the need for any finetuning. When compared to humans, the models approach their accuracy and robustness. We are releasing models and inference code to serve as a foundation for further work on robust speech processing.*
+
+
+Tips:
+
+- The model usually performs well without requiring any finetuning. 
+- The architecture follows a classic encoder-decoder architecture, which means that it relies on the [`~generation_utils.GenerationMixin.generate`] function for inference.
+- One can use [`WhisperProcessor`] to prepare audio for the model, and decode the predicted ID's back into text.
+
+This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ).
+The original code can be found [here](https://github.com/openai/whisper).
+
+
+## WhisperConfig
+
+[[autodoc]] WhisperConfig
+
+## WhisperTokenizer
+
+[[autodoc]] WhisperTokenizer
+    - build_inputs_with_special_tokens
+    - get_special_tokens_mask
+    - create_token_type_ids_from_sequences
+    - save_vocabulary
+
+## WhisperFeatureExtractor
+
+[[autodoc]] WhisperFeatureExtractor
+    - __call__
+
+## WhisperProcessor
+
+[[autodoc]] WhisperProcessor
+    - __call__
+    - from_pretrained
+    - save_pretrained
+    - batch_decode
+    - decode
+
+## WhisperModel
+
+[[autodoc]] WhisperModel
+    - forward
+
+## WhisperForConditionalGeneration
+
+[[autodoc]] WhisperForConditionalGeneration
+    - forward

src/transformers/__init__.py

@@ -390,6 +390,13 @@ _import_structure = {
         "WAVLM_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "WavLMConfig",
     ],
+    "models.whisper": [
+        "WHISPER_PRETRAINED_CONFIG_ARCHIVE_MAP",
+        "WhisperConfig",
+        "WhisperFeatureExtractor",
+        "WhisperProcessor",
+        "WhisperTokenizer",
+    ],
     "models.x_clip": [
         "XCLIP_PRETRAINED_CONFIG_ARCHIVE_MAP",
         "XCLIPConfig",
@@ -1863,6 +1870,14 @@ else:
             "Speech2TextPreTrainedModel",
         ]
     )
+    _import_structure["models.whisper"].extend(
+        [
+            "WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "WhisperForConditionalGeneration",
+            "WhisperModel",
+            "WhisperPreTrainedModel",
+        ]
+    )
     _import_structure["models.speech_to_text_2"].extend(["Speech2Text2ForCausalLM", "Speech2Text2PreTrainedModel"])
     _import_structure["models.splinter"].extend(
         [
@@ -3339,6 +3354,13 @@ if TYPE_CHECKING:
     from .models.wav2vec2_phoneme import Wav2Vec2PhonemeCTCTokenizer
     from .models.wav2vec2_with_lm import Wav2Vec2ProcessorWithLM
     from .models.wavlm import WAVLM_PRETRAINED_CONFIG_ARCHIVE_MAP, WavLMConfig
+    from .models.whisper import (
+        WHISPER_PRETRAINED_CONFIG_ARCHIVE_MAP,
+        WhisperConfig,
+        WhisperFeatureExtractor,
+        WhisperProcessor,
+        WhisperTokenizer,
+    )
     from .models.x_clip import (
         XCLIP_PRETRAINED_CONFIG_ARCHIVE_MAP,
         XCLIPConfig,
@@ -4724,6 +4746,12 @@ if TYPE_CHECKING:
             WavLMModel,
             WavLMPreTrainedModel,
         )
+        from .models.whisper import (
+            WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST,
+            WhisperForConditionalGeneration,
+            WhisperModel,
+            WhisperPreTrainedModel,
+        )
         from .models.x_clip import (
             XCLIP_PRETRAINED_MODEL_ARCHIVE_LIST,
             XCLIPModel,

src/transformers/configuration_utils.py

@@ -299,6 +299,8 @@ class PretrainedConfig(PushToHubMixin):
         self.forced_eos_token_id = kwargs.pop("forced_eos_token_id", None)
         self.remove_invalid_values = kwargs.pop("remove_invalid_values", False)
         self.exponential_decay_length_penalty = kwargs.pop("exponential_decay_length_penalty", None)
+        self.suppress_tokens = kwargs.pop("suppress_tokens", None)
+        self.begin_suppress_tokens = kwargs.pop("begin_suppress_tokens", None)
 
         # Fine-tuning task arguments
         self.architectures = kwargs.pop("architectures", None)

src/transformers/generation_logits_process.py

@@ -702,3 +702,49 @@ class LogitNormalization(LogitsProcessor, LogitsWarper):
     def __call__(self, input_ids: torch.Tensor, scores: torch.Tensor) -> torch.Tensor:
         scores = scores.log_softmax(dim=-1)
         return scores
+
+
+class SuppressTokensAtBeginLogitsProcessor(LogitsProcessor):
+    r"""
+    [`SuppressTokensAtBeginLogitsProcessor`] supresses a list of tokens as soon as the `generate` function starts
+    generating using `begin_index` tokens. This should ensure that the tokens defined by `begin_suppress_tokens` at not
+    sampled at the begining of the generation.
+    """
+
+    def __init__(self, begin_suppress_tokens, begin_index):
+        self.begin_suppress_tokens = list(begin_suppress_tokens)
+        self.begin_index = begin_index
+
+    def __call__(self, input_ids, scores):
+        if input_ids.shape[1] == self.begin_index:
+            scores[:, self.begin_suppress_tokens] = -float("inf")
+
+        return scores
+
+
+class SuppressTokensLogitsProcessor(LogitsProcessor):
+    r"""This processor can be used to suppress a list of tokens. The processor will set their log probs to `-inf` so that they
+    are not sampled."""
+
+    def __init__(self, suppress_tokens):
+        self.suppress_tokens = list(suppress_tokens)
+
+    def __call__(self, input_ids, scores):
+        scores[:, self.suppress_tokens] = -float("inf")
+        return scores
+
+
+class ForceTokensLogitsProcessor(LogitsProcessor):
+    r"""This processor can be used to force a list of tokens. The processor will set their log probs to `inf` so that they
+    are sampled at their corresponding index."""
+
+    def __init__(self, force_token_map):
+        self.force_token_map = dict(force_token_map)
+
+    def __call__(self, input_ids, scores):
+        generation_idx = input_ids.shape[-1]
+        current_token = self.force_token_map.get(generation_idx, None)
+        if current_token is not None:
+            scores[:, :] = -float("inf")
+            scores[:, current_token] = 0
+        return scores

src/transformers/generation_utils.py

@@ -30,6 +30,7 @@ from .generation_logits_process import (
     ExponentialDecayLengthPenalty,
     ForcedBOSTokenLogitsProcessor,
     ForcedEOSTokenLogitsProcessor,
+    ForceTokensLogitsProcessor,
     HammingDiversityLogitsProcessor,
     InfNanRemoveLogitsProcessor,
     LogitNormalization,
@@ -39,6 +40,8 @@ from .generation_logits_process import (
     NoRepeatNGramLogitsProcessor,
     PrefixConstrainedLogitsProcessor,
     RepetitionPenaltyLogitsProcessor,
+    SuppressTokensAtBeginLogitsProcessor,
+    SuppressTokensLogitsProcessor,
     TemperatureLogitsWarper,
     TopKLogitsWarper,
     TopPLogitsWarper,
@@ -691,6 +694,9 @@ class GenerationMixin:
         exponential_decay_length_penalty: Tuple,
         logits_processor: Optional[LogitsProcessorList],
         renormalize_logits: Optional[bool],
+        suppress_tokens: Optional[List[int]] = None,
+        begin_suppress_tokens: Optional[List[int]] = None,
+        forced_decoder_ids: Optional[List[int]] = None,
     ) -> LogitsProcessorList:
         """
         This class returns a [`LogitsProcessorList`] list object that contains all relevant [`LogitsProcessor`]
@@ -725,6 +731,12 @@ class GenerationMixin:
             if exponential_decay_length_penalty is not None
             else self.config.exponential_decay_length_penalty
         )
+        suppress_tokens = suppress_tokens if suppress_tokens is not None else self.config.suppress_tokens
+        begin_suppress_tokens = (
+            begin_suppress_tokens if begin_suppress_tokens is not None else self.config.begin_suppress_tokens
+        )
+        if forced_decoder_ids is None and hasattr(self.config, "forced_decoder_ids"):
+            forced_decoder_ids = self.config.forced_decoder_ids
         # instantiate processors list
 
         # the following idea is largely copied from this PR: https://github.com/huggingface/transformers/pull/5420/files
@@ -762,6 +774,16 @@ class GenerationMixin:
             processors.append(
                 ExponentialDecayLengthPenalty(exponential_decay_length_penalty, eos_token_id, input_ids_seq_length)
             )
+        if suppress_tokens is not None:
+            processors.append(SuppressTokensLogitsProcessor(suppress_tokens))
+        if begin_suppress_tokens is not None:
+            begin_index = input_ids_seq_length
+            begin_index = begin_index if (input_ids_seq_length > 1 or forced_bos_token_id is None) else begin_index + 1
+            if forced_decoder_ids is not None:
+                begin_index += forced_decoder_ids[-1][0]  # generation starts after the last token that is forced
+            processors.append(SuppressTokensAtBeginLogitsProcessor(begin_suppress_tokens, begin_index))
+        if forced_decoder_ids is not None:
+            processors.append(ForceTokensLogitsProcessor(forced_decoder_ids))
         processors = self._merge_criteria_processor_list(processors, logits_processor)
         # `LogitNormalization` should always be the last logit processor, when present
         if renormalize_logits is True:
@@ -932,6 +954,9 @@ class GenerationMixin:
         remove_invalid_values: Optional[bool] = None,
         synced_gpus: Optional[bool] = False,
         exponential_decay_length_penalty: Optional[Tuple[Union[int, float]]] = None,
+        suppress_tokens: Optional[List[int]] = None,
+        begin_suppress_tokens: Optional[List[int]] = None,
+        forced_decoder_ids: Optional[List[int]] = None,
         **model_kwargs,
     ) -> Union[GreedySearchOutput, SampleOutput, BeamSearchOutput, BeamSampleOutput, torch.LongTensor]:
         r"""
@@ -1090,6 +1115,14 @@ class GenerationMixin:
                 This Tuple adds an exponentially increasing length penalty, after a certain amount of tokens have been
                 generated. The tuple shall consist of: `(start_index, decay_factor)` where `start_index` indicates
                 where penalty starts and `decay_factor` represents the factor of exponential decay
+            suppress_tokens  (`List[int]`, *optional*, defaults to `model.config.suppress_tokens`):
+                A list of tokens that will be supressed at generation. The `SupressTokens` logit processor will set
+                their log probs to `-inf` so that they are not sampled.
+            begin_suppress_tokens  (`List[int]`, *optional*, defaults to `model.config.begin_suppress_tokens`):
+                A list of tokens that will be supressed at the begining of the generation. The `SupressBeginTokens`
+                logit processor will set their log probs to `-inf` so that they are not sampled.
+            forced_decoder_ids (`List[int]`, *optional*, defaults to `model.config.forced_decoder_ids`):
+                A list of tokens that will be forced as beginning tokens, before sampling.
 
             model_kwargs:
                 Additional model specific kwargs will be forwarded to the `forward` function of the model. If the model
@@ -1337,6 +1370,9 @@ class GenerationMixin:
             exponential_decay_length_penalty=exponential_decay_length_penalty,
             logits_processor=logits_processor,
             renormalize_logits=renormalize_logits,
+            suppress_tokens=suppress_tokens,
+            begin_suppress_tokens=begin_suppress_tokens,
+            forced_decoder_ids=forced_decoder_ids,
         )
 
         # 8. prepare stopping criteria

src/transformers/models/__init__.py

@@ -159,6 +159,7 @@ from . import (
     wav2vec2_phoneme,
     wav2vec2_with_lm,
     wavlm,
+    whisper,
     x_clip,
     xglm,
     xlm,

src/transformers/models/auto/configuration_auto.py

@@ -152,6 +152,7 @@ CONFIG_MAPPING_NAMES = OrderedDict(
         ("wav2vec2", "Wav2Vec2Config"),
         ("wav2vec2-conformer", "Wav2Vec2ConformerConfig"),
         ("wavlm", "WavLMConfig"),
+        ("whisper", "WhisperConfig"),
         ("xclip", "XCLIPConfig"),
         ("xglm", "XGLMConfig"),
         ("xlm", "XLMConfig"),
@@ -276,6 +277,7 @@ CONFIG_ARCHIVE_MAP_MAPPING_NAMES = OrderedDict(
         ("vit_msn", "VIT_MSN_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("wav2vec2", "WAV_2_VEC_2_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("wav2vec2-conformer", "WAV2VEC2_CONFORMER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
+        ("whisper", "WHISPER_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("xclip", "X_CLIP_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("xglm", "XGLM_PRETRAINED_CONFIG_ARCHIVE_MAP"),
         ("xlm", "XLM_PRETRAINED_CONFIG_ARCHIVE_MAP"),
@@ -434,6 +436,7 @@ MODEL_NAMES_MAPPING = OrderedDict(
         ("wav2vec2-conformer", "Wav2Vec2-Conformer"),
         ("wav2vec2_phoneme", "Wav2Vec2Phoneme"),
         ("wavlm", "WavLM"),
+        ("whisper", "Whisper"),
         ("xclip", "X-CLIP"),
         ("xglm", "XGLM"),
         ("xlm", "XLM"),

src/transformers/models/auto/feature_extraction_auto.py

@@ -77,6 +77,7 @@ FEATURE_EXTRACTOR_MAPPING_NAMES = OrderedDict(
         ("vit_msn", "ViTFeatureExtractor"),
         ("wav2vec2", "Wav2Vec2FeatureExtractor"),
         ("wav2vec2-conformer", "Wav2Vec2FeatureExtractor"),
+        ("whisper", "WhisperFeatureExtractor"),
         ("xclip", "CLIPFeatureExtractor"),
         ("yolos", "YolosFeatureExtractor"),
     ]

src/transformers/models/auto/modeling_auto.py

@@ -146,6 +146,7 @@ MODEL_MAPPING_NAMES = OrderedDict(
         ("wav2vec2", "Wav2Vec2Model"),
         ("wav2vec2-conformer", "Wav2Vec2ConformerModel"),
         ("wavlm", "WavLMModel"),
+        ("whisper", "WhisperModel"),
         ("xclip", "XCLIPModel"),
         ("xglm", "XGLMModel"),
         ("xlm", "XLMModel"),
@@ -273,6 +274,7 @@ MODEL_WITH_LM_HEAD_MAPPING_NAMES = OrderedDict(
         ("tapas", "TapasForMaskedLM"),
         ("transfo-xl", "TransfoXLLMHeadModel"),
         ("wav2vec2", "Wav2Vec2ForMaskedLM"),
+        ("whisper", "WhisperForConditionalGeneration"),
         ("xlm", "XLMWithLMHeadModel"),
         ("xlm-roberta", "XLMRobertaForMaskedLM"),
         ("xlm-roberta-xl", "XLMRobertaXLForMaskedLM"),
@@ -500,6 +502,7 @@ MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING_NAMES = OrderedDict(
     [
         ("speech-encoder-decoder", "SpeechEncoderDecoderModel"),
         ("speech_to_text", "Speech2TextForConditionalGeneration"),
+        ("whisper", "WhisperForConditionalGeneration"),
     ]
 )
 

src/transformers/models/auto/processing_auto.py

@@ -61,6 +61,7 @@ PROCESSOR_MAPPING_NAMES = OrderedDict(
         ("wav2vec2-conformer", "Wav2Vec2Processor"),
         ("wav2vec2_with_lm", "Wav2Vec2ProcessorWithLM"),
         ("wavlm", "Wav2Vec2Processor"),
+        ("whisper", "WhisperProcessor"),
         ("xclip", "CLIPProcessor"),
     ]
 )

src/transformers/models/auto/tokenization_auto.py

@@ -255,6 +255,7 @@ else:
             ("wav2vec2", ("Wav2Vec2CTCTokenizer", None)),
             ("wav2vec2-conformer", ("Wav2Vec2CTCTokenizer", None)),
             ("wav2vec2_phoneme", ("Wav2Vec2PhonemeCTCTokenizer", None)),
+            ("whisper", ("WhisperTokenizer" if is_sentencepiece_available() else None, None)),
             ("xclip", ("CLIPTokenizer", "CLIPTokenizerFast" if is_tokenizers_available() else None)),
             (
                 "xglm",

src/transformers/models/whisper/__init__.py

+# 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 OptionalDependencyNotAvailable, _LazyModule, is_torch_available
+
+
+_import_structure = {
+    "configuration_whisper": ["WHISPER_PRETRAINED_CONFIG_ARCHIVE_MAP", "WhisperConfig"],
+    "feature_extraction_whisper": ["WhisperFeatureExtractor"],
+    "processing_whisper": ["WhisperProcessor"],
+    "tokenization_whisper": ["WhisperTokenizer"],
+}
+
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_whisper"] = [
+        "WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST",
+        "WhisperForConditionalGeneration",
+        "WhisperModel",
+        "WhisperPreTrainedModel",
+    ]
+
+
+if TYPE_CHECKING:
+    from .configuration_whisper import WHISPER_PRETRAINED_CONFIG_ARCHIVE_MAP, WhisperConfig
+    from .feature_extraction_whisper import WhisperFeatureExtractor
+    from .processing_whisper import WhisperProcessor
+    from .tokenization_whisper import WhisperTokenizer
+
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_whisper import (
+            WHISPER_PRETRAINED_MODEL_ARCHIVE_LIST,
+            WhisperForConditionalGeneration,
+            WhisperModel,
+            WhisperPreTrainedModel,
+        )
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(__name__, globals()["__file__"], _import_structure, module_spec=__spec__)

src/transformers/models/whisper/configuration_whisper.py

+# coding=utf-8
+# Copyright 2022 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.
+""" Whisper model configuration"""
+
+from ...configuration_utils import PretrainedConfig
+from ...utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+WHISPER_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "openai/whisper-base": "https://huggingface.co/openai/whisper-base/resolve/main/config.json",
+}
+
+# fmt: off
+NON_SPEECH_TOKENS = [
+    1, 2, 6, 7, 8, 9, 10, 12, 14, 25,
+    26, 27, 28, 29, 31, 58, 59, 60, 61, 62,
+    63, 90, 91, 92, 93, 357, 366, 438, 532, 685,
+    705, 796, 930, 1058, 1220, 1267, 1279, 1303, 1343, 1377,
+    1391, 1635, 1782, 1875, 2162, 2361, 2488, 3467, 4008, 4211,
+    4600, 4808, 5299, 5855, 6329, 7203, 9609, 9959, 10563, 10786,
+    11420, 11709, 11907, 13163, 13697, 13700, 14808, 15306, 16410, 16791,
+    17992, 19203, 19510, 20724, 22305, 22935, 27007, 30109, 30420, 33409,
+    34949, 40283, 40493, 40549, 47282, 49146, 50257, 50359, 50360, 50361
+]
+NON_SPEECH_TOKENS_MULTI = [
+    1, 2, 6, 7, 8, 9, 10, 12, 14, 25,
+    26, 27, 28, 29, 31, 58, 59, 60, 61, 62,
+    63, 90, 91, 92, 93, 359, 503, 522, 542, 873,
+    893, 902, 918, 922, 931, 1350, 1853, 1982, 2460, 2627,
+    3246, 3253, 3268, 3536, 3846, 3961, 4183, 4667, 6585, 6647,
+    7273, 9061, 9383, 10428, 10929, 11938, 12033, 12331, 12562, 13793,
+    14157, 14635, 15265, 15618, 16553, 16604, 18362, 18956, 20075, 21675,
+    22520, 26130, 26161, 26435, 28279, 29464, 31650, 32302, 32470, 36865,
+    42863, 47425, 49870, 50254, 50258, 50360, 50361, 50362
+]
+# fmt: on
+
+
+class WhisperConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`WhisperModel`]. It is used to instantiate a
+    Whisper model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of the Whisper
+    [openai/whisper-tiny](https://huggingface.co/openai/whisper-tiny) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 51865):
+            Vocabulary size of the Whisper model. Defines the number of different tokens that can be represented by the
+            `decoder_input_ids` passed when calling [`WhisperModel`]
+        num_mel_bins (`int`, *optional*, defaults to 80):
+            Number of mel features used per input features. Should correspond to the value used in the
+            `WhisperProcessor` class.
+        encoder_layers (`int`, *optional*, defaults to 6):
+            Number of encoder layers.
+        decoder_layers (`int`, *optional*, defaults to 6):
+            Number of decoder layers.
+        encoder_attention_heads (`int`, *optional*, defaults to 4):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        decoder_attention_heads (`int`, *optional*, defaults to 4):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        encoder_ffn_dim (`int`, *optional*, defaults to 1536):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in encoder.
+        decoder_ffn_dim (`int`, *optional*, defaults to 1536):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in decoder.
+        encoder_layerdrop (`float`, *optional*, defaults to 0.0):
+            The LayerDrop probability for the encoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556)
+            for more details.
+        decoder_layerdrop (`float`, *optional*, defaults to 0.0):
+            The LayerDrop probability for the decoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556)
+            for more details.
+        decoder_start_token_id (`int`, *optional*, defaults to 50257):
+            Corresponds to the "<|startoftranscript|>" token, which is automatically used when no `decoder_input_ids`
+            are provided to the `generate` function. It is used to guide the model`s generation process depending on
+            the task.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models).
+        is_encoder_decoder (`bool`, *optional*, defaults to `True`):
+            Whether the model is used as an encoder/decoder or not.
+        activation_function (`str`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"silu"` and `"gelu_new"` are supported.
+        d_model (`int`, *optional*, defaults to 256):
+            Dimensionality of the layers.
+        dropout (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        activation_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for activations inside the fully connected layer.
+        init_std (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        scale_embedding (`bool`, *optional*, defaults to False):
+            Scale embeddings by diving by sqrt(d_model).
+        max_source_positions (`int`, *optional*, defaults to 1500):
+            The maximum sequence length of log-mel filter-bank features that this model might ever be used with.
+        max_target_positions (`int`, *optional*, defaults to 448):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
+        pad_token_id (`int`, *optional*, defaults to 50256):
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 50256):
+            Begin of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 50257):
+            End of stream token id.
+        tie_word_embeddings (`bool`, *optional*, defaults to `True`):
+            Whether to tie input and output embeddings.
+        suppress_tokens (`List[int]`, *optional*):
+            A list containing the non-speech tokens that will be used by the logit processor in the `generate`
+            function. NON_SPEECH_TOKENS and NON_SPEECH_TOKENS_MULTI each correspond to the `english-only` and the
+            `multilingual` model.
+        begin_suppress_tokens (`List[int]`, *optional*, defaults to `[220,50256]`):
+            A list containing tokens that will be supressed at the beginning of the sampling process. Initialized as
+            the token for `" "` (`blank_token_id`) and the `eos_token_id`
+
+
+    Example:
+
+    ```python
+    >>> from transformers import WhisperModel, WhisperConfig
+
+    >>> # Initializing a Whisper tiny style configuration
+    >>> configuration = WhisperConfig()
+
+    >>> # Initializing a model from the tiny style configuration
+    >>> model = WhisperModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+    model_type = "whisper"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {"num_attention_heads": "encoder_attention_heads", "hidden_size": "d_model"}
+
+    def __init__(
+        self,
+        vocab_size=51865,
+        num_mel_bins=80,
+        encoder_layers=6,
+        encoder_attention_heads=4,
+        decoder_layers=6,
+        decoder_attention_heads=4,
+        decoder_ffn_dim=1536,
+        encoder_ffn_dim=1536,
+        encoder_layerdrop=0.0,
+        decoder_layerdrop=0.0,
+        decoder_start_token_id=50257,
+        use_cache=True,
+        is_encoder_decoder=True,
+        activation_function="gelu",
+        d_model=256,
+        dropout=0.0,
+        attention_dropout=0.0,
+        activation_dropout=0.0,
+        init_std=0.02,
+        scale_embedding=False,
+        max_source_positions=1500,
+        max_target_positions=448,
+        pad_token_id=50256,
+        bos_token_id=50257,
+        eos_token_id=50256,
+        tie_word_embeddings=True,
+        suppress_tokens=None,
+        begin_suppress_tokens=[220, 50256],
+        **kwargs
+    ):
+        self.vocab_size = vocab_size
+        self.num_mel_bins = num_mel_bins
+        self.d_model = d_model
+        self.encoder_layers = encoder_layers
+        self.encoder_attention_heads = encoder_attention_heads
+        self.decoder_layers = decoder_layers
+        self.decoder_attention_heads = decoder_attention_heads
+        self.decoder_ffn_dim = decoder_ffn_dim
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.dropout = dropout
+        self.attention_dropout = attention_dropout
+        self.activation_dropout = activation_dropout
+        self.activation_function = activation_function
+        self.init_std = init_std
+        self.encoder_layerdrop = encoder_layerdrop
+        self.decoder_layerdrop = decoder_layerdrop
+        self.use_cache = use_cache
+        self.num_hidden_layers = encoder_layers
+        self.scale_embedding = scale_embedding  # scale factor will be sqrt(d_model) if True
+        self.tie_word_embeddings = tie_word_embeddings
+        self.max_source_positions = max_source_positions
+        self.max_target_positions = max_target_positions
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            is_encoder_decoder=is_encoder_decoder,
+            decoder_start_token_id=decoder_start_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            suppress_tokens=suppress_tokens,
+            begin_suppress_tokens=begin_suppress_tokens,
+            **kwargs,
+        )