Whisper
Overview
The Whisper model was proposed in Robust Speech Recognition via Large-Scale Weak Supervision 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.
This model was contributed by Arthur Zucker. The Tensorflow version of this model was contributed by amyeroberts. The original code can be found here.
Usage 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 generate() function for inference.
One can use WhisperProcessor to prepare audio for the model, and decode the predicted ID’s back into text.
To convert the model and the processor, we recommend using the following:
python src/transformers/models/whisper/convert_openai_to_hf.py --checkpoint_path "" --pytorch_dump_folder_path "Arthur/whisper-3" --convert_preprocessor True
The script will automatically determine all necessary parameters from the OpenAI checkpoint. A tiktoken
library needs to be installed
to perform the conversion of the OpenAI tokenizer to the tokenizers
version.
Inference
Here is a step-by-step guide to transcribing an audio sample using a pre-trained Whisper model:
>>> from datasets import load_dataset
>>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
>>> # Select an audio file and read it:
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> audio_sample = ds[0]["audio"]
>>> # Load the Whisper model in Hugging Face format:
>>> processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
>>> # Use the model and processor to transcribe the audio:
>>> input_features = processor(
... audio_sample["array"], sampling_rate=audio_sample["sampling_rate"], return_tensors="pt"
... ).input_features
>>> # Generate token ids
>>> predicted_ids = model.generate(input_features)
>>> # Decode token ids to text
>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
>>> transcription[0]
' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
Whisper is compatible with the following optimisations:
- PyTorch Scaled Dot Product Attention (SDPA): flash attention and memory-efficient attention kernels. Enabled by default for
torch>=2.1.1
. - Flash Attention 2: improved implementation of flash attention through better parallelism and work partitioning.
- torch.compile: JIT-compile the forward pass to dispatch to efficient fused kernels.
As an example, the following codesnippet enables SDPA and torch.compile
for up to 5x faster inference:
>>> from datasets import load_dataset
>>> from transformers import WhisperProcessor, WhisperForConditionalGeneration
>>> # Select an audio file and read it:
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> audio_sample = ds[0]["audio"]
>>> # Load the Whisper model with SDPA attention
>>> processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en", attn_implementation="sdpa")
>>> # Enable static cache and compile the forward pass
>>> model.generation_config.cache_implementation = "static"
>>> model.forward = torch.compile(model.forward, mode="reduce-overhead", fullgraph=True)
>>> # Use the model and processor to transcribe the audio:
>>> input_features = processor(
... audio_sample["array"], sampling_rate=audio_sample["sampling_rate"], return_tensors="pt"
... ).input_features
>>> # Compile the forward pass
>>> _ = model.generate(input_features)
>>> # Generate token ids using compiled graph (fast!)
>>> predicted_ids = model.generate(input_features)
>>> # Decode token ids to text
>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
>>> transcription[0]
' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
For more details on each optimisation, refer to the documentation linked above.
Resources
A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Whisper. If you’re interested in submitting a resource to be included here, please feel free to open a Pull Request and we’ll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
- Fine-tune Whisper on your own dataset for better downstream performance.
- Distil-Whisper: Upto 6x faster, 2x smaller distilled Whisper models for English. We release the model checkpoints, and distillation code.
- A fork with a script to convert a Whisper model in Hugging Face format to OpenAI format. 🌎 Usage example:
pip install -U openai-whisper python convert_hf_to_openai.py \ --checkpoint openai/whisper-tiny \ --whisper_dump_path whisper-tiny-openai.pt
WhisperConfig
class transformers.WhisperConfig
< source >( vocab_size = 51865 num_mel_bins = 80 encoder_layers = 4 encoder_attention_heads = 6 decoder_layers = 4 decoder_attention_heads = 6 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 = 384 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 = 50256 eos_token_id = 50256 suppress_tokens = None begin_suppress_tokens = [220, 50256] use_weighted_layer_sum = False classifier_proj_size = 256 apply_spec_augment = False mask_time_prob = 0.05 mask_time_length = 10 mask_time_min_masks = 2 mask_feature_prob = 0.0 mask_feature_length = 10 mask_feature_min_masks = 0 median_filter_width = 7 **kwargs )
Parameters
- vocab_size (
int
, optional, defaults to 51865) — Vocabulary size of the Whisper model. Defines the number of different tokens that can be represented by thedecoder_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 theWhisperProcessor
class. - encoder_layers (
int
, optional, defaults to 4) — Number of encoder layers. - decoder_layers (
int
, optional, defaults to 4) — Number of decoder layers. - encoder_attention_heads (
int
, optional, defaults to 6) — Number of attention heads for each attention layer in the Transformer encoder. - decoder_attention_heads (
int
, optional, defaults to 6) — 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 nodecoder_input_ids
are provided to thegenerate
function. It is used to guide the model`s generation process depending on the task. - use_cache (
bool
, optional, defaults toTrue
) — Whether or not the model should return the last key/values attentions (not used by all models). - is_encoder_decoder (
bool
, optional, defaults toTrue
) — 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 384) — 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 50256) — End of stream token id. - suppress_tokens (
List[int]
, optional) — A list containing the non-speech tokens that will be used by the logit processor in thegenerate
function. NON_SPEECH_TOKENS and NON_SPEECH_TOKENS_MULTI each correspond to theenglish-only
and themultilingual
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 theeos_token_id
- use_weighted_layer_sum (
bool
, optional, defaults toFalse
) — Whether to use a weighted average of layer outputs with learned weights. Only relevant when using an instance of WhisperForAudioClassification. - classifier_proj_size (
int
, optional, defaults to 256) — Dimensionality of the projection before token mean-pooling for classification. Only relevant when using an instance of WhisperForAudioClassification. - apply_spec_augment (
bool
, optional, defaults toFalse
) — Whether to apply SpecAugment data augmentation to the outputs of the feature encoder. For reference see SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition. - mask_time_prob (
float
, optional, defaults to 0.05) — Percentage (between 0 and 1) of all feature vectors along the time axis which will be masked. The masking procecure generatesmask_time_prob*len(time_axis)/mask_time_length
independent masks over the axis. If reasoning from the propability of each feature vector to be chosen as the start of the vector span to be masked, mask_time_prob should beprob_vector_start*mask_time_length
. Note that overlap may decrease the actual percentage of masked vectors. This is only relevant ifapply_spec_augment == True
. - mask_time_length (
int
, optional, defaults to 10) — Length of vector span along the time axis. - mask_time_min_masks (
int
, optional, defaults to 2), — The minimum number of masks of lengthmask_feature_length
generated along the time axis, each time step, irrespectively ofmask_feature_prob
. Only relevant if ”mask_time_prob*len(time_axis)/mask_time_length < mask_time_min_masks” - mask_feature_prob (
float
, optional, defaults to 0.0) — Percentage (between 0 and 1) of all feature vectors along the feature axis which will be masked. The masking procecure generatesmask_feature_prob*len(feature_axis)/mask_time_length
independent masks over the axis. If reasoning from the propability of each feature vector to be chosen as the start of the vector span to be masked, mask_feature_prob should beprob_vector_start*mask_feature_length
. Note that overlap may decrease the actual percentage of masked vectors. This is only relevant ifapply_spec_augment is True
. - mask_feature_length (
int
, optional, defaults to 10) — Length of vector span along the feature axis. - mask_feature_min_masks (
int
, optional, defaults to 0), — The minimum number of masks of lengthmask_feature_length
generated along the feature axis, each time step, irrespectively ofmask_feature_prob
. Only relevant ifmask_feature_prob*len(feature_axis)/mask_feature_length < mask_feature_min_masks
. - median_filter_width (
int
, optional, defaults to 7) — Width of the median filter used to smoothen to cross-attention outputs when computing token timestamps. Should be an odd number.
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 architecture.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.
Example:
>>> from transformers import WhisperConfig, WhisperModel
>>> # Initializing a Whisper tiny style configuration
>>> configuration = WhisperConfig()
>>> # Initializing a model (with random weights) from the tiny style configuration
>>> model = WhisperModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config
WhisperTokenizer
class transformers.WhisperTokenizer
< source >( vocab_file merges_file normalizer_file = None errors = 'replace' unk_token = '<|endoftext|>' bos_token = '<|endoftext|>' eos_token = '<|endoftext|>' pad_token = None add_prefix_space = False language = None task = None predict_timestamps = False **kwargs )
Parameters
- vocab_file (
str
) — Path to the vocabulary file. - merges_file (
str
) — Path to the merges file. - normalizer_file (
str
, optional) — Path to the normalizer_file file. - errors (
str
, optional, defaults to"replace"
) — Paradigm to follow when decoding bytes to UTF-8. See bytes.decode for more information. - unk_token (
str
, optional, defaults to"<|endoftext|>"
) — The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead. - bos_token (
str
, optional, defaults to"<|endoftext|>"
) — The beginning of sequence token. Thedecoder_start_token_id
is used to set the first token as"<|startoftranscript|>"
when generating. - eos_token (
str
, optional, defaults to"<|endoftext|>"
) — The end of sequence token. - pad_token (
str
, optional) — The token used for padding, for example when batching sequences of different lengths. - add_prefix_space (
bool
, optional, defaults toFalse
) — Whether or not to add an initial space to the input. This allows to treat the leading word just as any other word. - language (
str
, optional) — The language of the transcription text. The corresponding language id token is appended to the start of the sequence for multilingual speech recognition and speech translation tasks, e.g. for Spanish the token"<|es|>"
is appended to the start of sequence. This should be used for multilingual fine-tuning only. - task (
str
, optional) — Task identifier to append at the start of sequence (if any). This should be used for mulitlingual fine-tuning, with"transcribe"
for speech recognition and"translate"
for speech translation. - predict_timestamps (
bool
, optional, defaults toFalse
) — Whether to omit the<|notimestamps|>
token at the start of the sequence.
Construct a Whisper tokenizer.
This tokenizer inherits from PreTrainedTokenizer which contains some of the main methods. Users should refer to the superclass for more information regarding such methods.
set_prefix_tokens
< source >( language: str = None task: str = None predict_timestamps: bool = None )
Parameters
- language (
str
, optional, defaults toNone
) — The language of the transcription text. - task (
str
, optional, defaults toNone
) — Task identifier to append at the start of sequence (if any). - predict_timestamps (
bool
, optional, defaults toNone
) — Whether to omit the<|notimestamps|>
token at the start of the sequence.
Override the prefix tokens appended to the start of the label sequence. This method can be used standalone to
update the prefix tokens as required when fine-tuning. Example:
>>> # instantiate the tokenizer and set the prefix token to Spanish
>>> tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-tiny", language="spanish")
>>> # now switch the prefix token from Spanish to French
>>> tokenizer.set_prefix_tokens(language="french")
Build model inputs from a sequence by appending eos_token_id.
get_special_tokens_mask
< source >( token_ids_0: List token_ids_1: Optional = None already_has_special_tokens: bool = False ) → List[int]
Parameters
- token_ids_0 (
List[int]
) — List of IDs. - token_ids_1 (
List[int]
, optional) — Optional second list of IDs for sequence pairs. - already_has_special_tokens (
bool
, optional, defaults toFalse
) — Whether or not the token list is already formatted with special tokens for the model.
Returns
List[int]
A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
special tokens using the tokenizer prepare_for_model
method.
create_token_type_ids_from_sequences
< source >( token_ids_0: List token_ids_1: Optional = None ) → List[int]
Create the token type IDs corresponding to the sequences passed. What are token type IDs?
Should be overridden in a subclass if the model has a special way of building those.
batch_decode
< source >( sequences: Union skip_special_tokens: bool = False clean_up_tokenization_spaces: bool = None **kwargs ) → List[str]
Parameters
- sequences (
Union[List[int], List[List[int]], np.ndarray, torch.Tensor, tf.Tensor]
) — List of tokenized input ids. Can be obtained using the__call__
method. - skip_special_tokens (
bool
, optional, defaults toFalse
) — Whether or not to remove special tokens in the decoding. - clean_up_tokenization_spaces (
bool
, optional) — Whether or not to clean up the tokenization spaces. IfNone
, will default toself.clean_up_tokenization_spaces
. - kwargs (additional keyword arguments, optional) — Will be passed to the underlying model specific decode method.
Returns
List[str]
The list of decoded sentences.
Convert a list of lists of token ids into a list of strings by calling decode.
decode
< source >( token_ids skip_special_tokens: bool = False clean_up_tokenization_spaces: bool = None output_offsets: bool = False time_precision: float = 0.02 decode_with_timestamps: bool = False normalize: bool = False basic_normalize: bool = False remove_diacritics: bool = False **kwargs ) → str
Parameters
- token_ids (
Union[int, List[int], np.ndarray, torch.Tensor, tf.Tensor]
) — List of tokenized input ids. Can be obtained using the__call__
method. - skip_special_tokens (
bool
, optional, defaults toFalse
) — Whether or not to remove special tokens in the decoding. - clean_up_tokenization_spaces (
bool
, optional) — Whether or not to clean up the tokenization spaces. IfNone
, will default toself.clean_up_tokenization_spaces
(available in thetokenizer_config
). - output_offsets (
bool
, optional, defaults toFalse
) — Whether or not to output the offsets of the tokens. This should only be set if the model predicted timestamps. - time_precision (
float
,optional
, defaults to 0.02) — The time ratio to convert from token to time. - decode_with_timestamps (
bool
, optional, defaults toFalse
) — Whether or not to decode with timestamps included in the raw text. - normalize (
bool
, optional, defaults toFalse
) — Whether or not to apply the English text normalizer to the decoded text. Only applicable when the target text is in English. Otherwise, the basic text normalizer should be applied. - basic_normalize (
bool
, optional, defaults toFalse
) — Whether or not to apply the Basic text normalizer to the decoded text. Applicable to multilingual target text. - remove_diacritics (
bool
, optional, defaults toFalse
) — Whether or not to remove diacritics when applying the Basic text normalizer. Removing diacritics may destroy information in the decoded text, hence it should be used with caution. - kwargs (additional keyword arguments, optional) — Will be passed to the underlying model specific decode method.
Returns
str
The decoded sentence.
Converts a sequence of ids in a string, using the tokenizer and vocabulary with options to remove special tokens and clean up tokenization spaces.
Similar to doing self.convert_tokens_to_string(self.convert_ids_to_tokens(token_ids))
.
Normalize a given string using the BasicTextNormalizer
class, which preforms commons transformation on
multilingual text.
Normalize a given string using the EnglishTextNormalizer
class, which preforms commons transformation on
english text.
WhisperTokenizerFast
class transformers.WhisperTokenizerFast
< source >( vocab_file = None merges_file = None normalizer_file = None tokenizer_file = None unk_token = '<|endoftext|>' bos_token = '<|endoftext|>' eos_token = '<|endoftext|>' add_prefix_space = False language = None task = None predict_timestamps = False **kwargs )
Parameters
- vocab_file (
str
, optional) — Path to the vocabulary file. - merges_file (
str
, optional) — Path to the merges file. - normalizer_file (
str
, optional) — Path to the normalizer_file file. - tokenizer_file (
str
, optional) — Path to tokenizers file (generally has a .json extension) that contains everything needed to load the tokenizer. - unk_token (
str
, optional, defaults to"<|endoftext|>"
) — The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this token instead. - bos_token (
str
, optional, defaults to"<|endoftext|>"
) — The beginning of sequence token. Thedecoder_start_token_id
is used to set the first token as"<|startoftranscript|>"
when generating. - eos_token (
str
, optional, defaults to"<|endoftext|>"
) — The end of sequence token. - add_prefix_space (
bool
, optional, defaults toFalse
) — Whether or not to add an initial space to the input. This allows to treat the leading word just as any other word. (Whisper tokenizer detect beginning of words by the preceding space). - language (
str
, optional) — The language of the transcription text. The corresponding language id token is appended to the start of the sequence for multilingual speech recognition and speech translation tasks, e.g. for Spanish the token"<|es|>"
is appended to the start of sequence. This should be used for multilingual fine-tuning only. - task (
str
, optional) — Task identifier to append at the start of sequence (if any). This should be used for mulitlingual fine-tuning, with"transcribe"
for speech recognition and"translate"
for speech translation. - predict_timestamps (
bool
, optional, defaults toFalse
) — Whether to omit the<|notimestamps|>
token at the start of the sequence.
Construct a “fast” Whisper tokenizer (backed by HuggingFace’s tokenizers library).
This tokenizer inherits from PreTrainedTokenizerFast which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.
set_prefix_tokens
< source >( language: str = None task: str = None predict_timestamps: bool = None )
Parameters
- language (
str
, optional, defaults toNone
) — The language of the transcription text. - task (
str
, optional, defaults toNone
) — Task identifier to append at the start of sequence (if any). - predict_timestamps (
bool
, optional, defaults toNone
) — Whether to omit the<|notimestamps|>
token at the start of the sequence.
Override the prefix tokens appended to the start of the label sequence. This method can be used standalone to
update the prefix tokens as required when fine-tuning. Example:
>>> # instantiate the tokenizer and set the prefix token to Spanish
>>> tokenizer = WhisperTokenizerFast.from_pretrained("openai/whisper-tiny", language="spanish")
>>> # now switch the prefix token from Spanish to French
>>> tokenizer.set_prefix_tokens(language="french")
Build model inputs from a sequence by appending eos_token_id.
get_special_tokens_mask
< source >( token_ids_0: List token_ids_1: Optional = None already_has_special_tokens: bool = False ) → List[int]
Parameters
- token_ids_0 (
List[int]
) — List of IDs. - token_ids_1 (
List[int]
, optional) — Optional second list of IDs for sequence pairs. - already_has_special_tokens (
bool
, optional, defaults toFalse
) — Whether or not the token list is already formatted with special tokens for the model.
Returns
List[int]
A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
special tokens using the tokenizer prepare_for_model
method.
create_token_type_ids_from_sequences
< source >( token_ids_0: List token_ids_1: Optional = None ) → List[int]
Create the token type IDs corresponding to the sequences passed. What are token type IDs?
Should be overridden in a subclass if the model has a special way of building those.
batch_decode
< source >( sequences: Union skip_special_tokens: bool = False clean_up_tokenization_spaces: bool = None **kwargs ) → List[str]
Parameters
- sequences (
Union[List[int], List[List[int]], np.ndarray, torch.Tensor, tf.Tensor]
) — List of tokenized input ids. Can be obtained using the__call__
method. - skip_special_tokens (
bool
, optional, defaults toFalse
) — Whether or not to remove special tokens in the decoding. - clean_up_tokenization_spaces (
bool
, optional) — Whether or not to clean up the tokenization spaces. IfNone
, will default toself.clean_up_tokenization_spaces
. - kwargs (additional keyword arguments, optional) — Will be passed to the underlying model specific decode method.
Returns
List[str]
The list of decoded sentences.
Convert a list of lists of token ids into a list of strings by calling decode.
decode
< source >( token_ids skip_special_tokens: bool = False clean_up_tokenization_spaces: bool = None output_offsets: bool = False time_precision: float = 0.02 decode_with_timestamps: bool = False normalize: bool = False basic_normalize: bool = False remove_diacritics: bool = False **kwargs ) → str
Parameters
- token_ids (
Union[int, List[int], np.ndarray, torch.Tensor, tf.Tensor]
) — List of tokenized input ids. Can be obtained using the__call__
method. - skip_special_tokens (
bool
, optional, defaults toFalse
) — Whether or not to remove special tokens in the decoding. - clean_up_tokenization_spaces (
bool
, optional) — Whether or not to clean up the tokenization spaces. IfNone
, will default toself.clean_up_tokenization_spaces
(available in thetokenizer_config
). - output_offsets (
bool
, optional, defaults toFalse
) — Whether or not to output the offsets of the tokens. This should only be set if the model predicted timestamps. - time_precision (
float
,optional
, defaults to 0.02) — The time ratio to convert from token to time. - decode_with_timestamps (
bool
, optional, defaults toFalse
) — Whether or not to decode with timestamps included in the raw text. - normalize (
bool
, optional, defaults toFalse
) — Whether or not to apply the English text normalizer to the decoded text. Only applicable when the target text is in English. Otherwise, the basic text normalizer should be applied. - basic_normalize (
bool
, optional, defaults toFalse
) — Whether or not to apply the Basic text normalizer to the decoded text. Applicable to multilingual target text. - remove_diacritics (
bool
, optional, defaults toFalse
) — Whether or not to remove diacritics when applying the Basic text normalizer. Removing diacritics may destroy information in the decoded text, hence it should be used with caution. - kwargs (additional keyword arguments, optional) — Will be passed to the underlying model specific decode method.
Returns
str
The decoded sentence.
Converts a sequence of ids in a string, using the tokenizer and vocabulary with options to remove special tokens and clean up tokenization spaces.
Similar to doing self.convert_tokens_to_string(self.convert_ids_to_tokens(token_ids))
.
Normalize a given string using the BasicTextNormalizer
class, which preforms commons transformation on
multilingual text.
Normalize a given string using the EnglishTextNormalizer
class, which preforms commons transformation on
english text.
WhisperFeatureExtractor
class transformers.WhisperFeatureExtractor
< source >( feature_size = 80 sampling_rate = 16000 hop_length = 160 chunk_length = 30 n_fft = 400 padding_value = 0.0 return_attention_mask = False **kwargs )
Parameters
- feature_size (
int
, defaults to 80) — The feature dimension of the extracted features. - sampling_rate (
int
, defaults to 16000) — The sampling rate at which the audio files should be digitalized expressed in hertz (Hz). - hop_length (
int
, defaults to 160) — Length of the overlaping windows for the STFT used to obtain the Mel Frequency coefficients. - chunk_length (
int
, defaults to 30) — The maximum number of chuncks ofsampling_rate
samples used to trim and pad longer or shorter audio sequences. - n_fft (
int
, defaults to 400) — Size of the Fourier transform. - padding_value (
float
, optional, defaults to 0.0) — Padding value used to pad the audio. Should correspond to silences.
Constructs a Whisper feature extractor.
This feature extractor inherits from SequenceFeatureExtractor which contains most of the main methods. Users should refer to this superclass for more information regarding those methods.
This class extracts mel-filter bank features from raw speech using a custom numpy implementation of the Short Time Fourier Transform
which should match pytorch’s torch.stft
equivalent.
__call__
< source >( raw_speech: Union truncation: bool = True pad_to_multiple_of: Optional = None return_tensors: Union = None return_attention_mask: Optional = None padding: Optional = 'max_length' max_length: Optional = None sampling_rate: Optional = None do_normalize: Optional = None device: Optional = 'cpu' return_token_timestamps: Optional = None **kwargs )
Parameters
- raw_speech (
np.ndarray
,List[float]
,List[np.ndarray]
,List[List[float]]
) — The sequence or batch of sequences to be padded. Each sequence can be a numpy array, a list of float values, a list of numpy arrays or a list of list of float values. Must be mono channel audio, not stereo, i.e. single float per timestep. - truncation (
bool
, optional, default toTrue
) — Activates truncation to cut input sequences longer than max_length to max_length. - pad_to_multiple_of (
int
, optional, defaults to None) — If set will pad the sequence to a multiple of the provided value.This is especially useful to enable the use of Tensor Cores on NVIDIA hardware with compute capability
>= 7.5
(Volta), or on TPUs which benefit from having sequence lengths be a multiple of 128. - return_attention_mask (
bool
, optional) — Whether to return the attention mask. If left to the default, will return the attention mask according to the specific feature_extractor’s default.For Whisper models,
attention_mask
should always be passed for batched inference, to avoid subtle bugs. - return_tensors (
str
or TensorType, optional) — If set, will return tensors instead of list of python integers. Acceptable values are:'tf'
: Return TensorFlowtf.constant
objects.'pt'
: Return PyTorchtorch.Tensor
objects.'np'
: Return Numpynp.ndarray
objects.
- sampling_rate (
int
, optional) — The sampling rate at which theraw_speech
input was sampled. It is strongly recommended to passsampling_rate
at the forward call to prevent silent errors and allow automatic speech recognition pipeline. - padding_value (
float
, defaults to 0.0) — The value that is used to fill the padding values / vectors. - do_normalize (
bool
, optional, defaults toFalse
) — Whether or not to zero-mean unit-variance normalize the input. Normalizing can help to significantly improve the performance of the model. - device (
str
, optional, defaults to'cpu'
) — Specifies the device for computation of the log-mel spectrogram of audio signals in the_torch_extract_fbank_features
method. (e.g., “cpu”, “cuda”) - return_token_timestamps (
bool
, optional, defaults toNone
) — Whether or not to return the number of frames of the input raw_speech. These num_frames can be used by the model to compute word level timestamps.
Main method to featurize and prepare for the model one or several sequence(s). Implementation uses PyTorch for the STFT computation if available, otherwise a slower NumPy based one.
WhisperProcessor
class transformers.WhisperProcessor
< source >( feature_extractor tokenizer )
Parameters
- feature_extractor (
WhisperFeatureExtractor
) — An instance of WhisperFeatureExtractor. The feature extractor is a required input. - tokenizer (
WhisperTokenizer
) — An instance of WhisperTokenizer. The tokenizer is a required input.
Constructs a Whisper processor which wraps a Whisper feature extractor and a Whisper tokenizer into a single processor.
WhisperProcessor offers all the functionalities of WhisperFeatureExtractor and WhisperTokenizer. See the call() and decode() for more information.
Forwards the audio
argument to WhisperFeatureExtractor’s call() and the text
argument to call(). Please refer to the doctsring of the above two methods for more
information.
from_pretrained
< source >( pretrained_model_name_or_path: Union cache_dir: Union = None force_download: bool = False local_files_only: bool = False token: Union = None revision: str = 'main' **kwargs )
Parameters
- pretrained_model_name_or_path (
str
oros.PathLike
) — This can be either:- a string, the model id of a pretrained feature_extractor hosted inside a model repo on huggingface.co.
- a path to a directory containing a feature extractor file saved using the
save_pretrained() method, e.g.,
./my_model_directory/
. - a path or url to a saved feature extractor JSON file, e.g.,
./my_model_directory/preprocessor_config.json
. **kwargs — Additional keyword arguments passed along to both from_pretrained() and~tokenization_utils_base.PreTrainedTokenizer.from_pretrained
.
Instantiate a processor associated with a pretrained model.
This class method is simply calling the feature extractor
from_pretrained(), image processor
ImageProcessingMixin and the tokenizer
~tokenization_utils_base.PreTrainedTokenizer.from_pretrained
methods. Please refer to the docstrings of the
methods above for more information.
save_pretrained
< source >( save_directory push_to_hub: bool = False **kwargs )
Parameters
- save_directory (
str
oros.PathLike
) — Directory where the feature extractor JSON file and the tokenizer files will be saved (directory will be created if it does not exist). - push_to_hub (
bool
, optional, defaults toFalse
) — Whether or not to push your model to the Hugging Face model hub after saving it. You can specify the repository you want to push to withrepo_id
(will default to the name ofsave_directory
in your namespace). - kwargs (
Dict[str, Any]
, optional) — Additional key word arguments passed along to the push_to_hub() method.
Saves the attributes of this processor (feature extractor, tokenizer…) in the specified directory so that it can be reloaded using the from_pretrained() method.
This class method is simply calling save_pretrained() and save_pretrained(). Please refer to the docstrings of the methods above for more information.
This method forwards all its arguments to WhisperTokenizer’s batch_decode(). Please refer to the docstring of this method for more information.
This method forwards all its arguments to WhisperTokenizer’s decode(). Please refer to the docstring of this method for more information.
WhisperModel
class transformers.WhisperModel
< source >( config: WhisperConfig )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The bare Whisper Model outputting raw hidden-states without any specific head on top. This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( input_features: Optional = None attention_mask: Optional = None decoder_input_ids: Optional = None decoder_attention_mask: Optional = None head_mask: Optional = None decoder_head_mask: Optional = None cross_attn_head_mask: Optional = None encoder_outputs: Optional = None past_key_values: Union = None decoder_inputs_embeds: Optional = None decoder_position_ids: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None cache_position: Optional = None ) → transformers.modeling_outputs.Seq2SeqModelOutput or tuple(torch.FloatTensor)
Parameters
- input_features (
torch.FloatTensor
of shape(batch_size, feature_size, sequence_length)
) — Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the AutoFeatureExtractor should be used for extracting the mel features, padding and conversion into a tensor of typetorch.FloatTensor
. See call() - attention_mask (
torch.LongTensor
of shape(batch_size, sequence_length)
, optional) — Mask to avoid performing SpecAugment data augmentation on padding token indices. Mask values selected in[0, 1]
:- 1 for tokens that are not masked,
- 0 for tokens that are masked.
- decoder_input_ids (
torch.LongTensor
of shape(batch_size, target_sequence_length)
, optional) — Indices of decoder input sequence tokens in the vocabulary.Indices can be obtained using WhisperTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
Whisper uses the
decoder_start_token_id
as the starting token fordecoder_input_ids
generation. Ifpast_key_values
is used, optionally only the lastdecoder_input_ids
have to be input (seepast_key_values
). - decoder_attention_mask (
torch.LongTensor
of shape(batch_size, target_sequence_length)
, optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids
. Causal mask will also be used by default.If you want to change padding behavior, you should read
modeling_whisper._prepare_decoder_attention_mask
and modify to your needs. See diagram 1 in the BART paper for more information on the default strategy. - head_mask (
torch.Tensor
of shape(encoder_layers, encoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- decoder_head_mask (
torch.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the decoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- cross_attn_head_mask (
torch.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the cross-attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- encoder_outputs (
tuple(tuple(torch.FloatTensor)
, optional) — Tuple consists of (last_hidden_state
, optional:hidden_states
, optional:attentions
)last_hidden_state
of shape(batch_size, sequence_length, hidden_size)
, optional) is a sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of the decoder. - past_key_values (
EncoderDecoderCache
ortuple(tuple(torch.FloatTensor))
, optional) — Pre-computed hidden-states that can be used to speed up auto-regressive (sequential) decoding. There are four sets of pre-computed hidden-states: key and values states in the self-attention blocks (2) and in the cross-attention blocks (2). Thepast_key_values
are returned whenuse_cache=True
is passed or whenconfig.use_cache=True
Two formats are allowed:
- An EncoderDecoderCache instance;
- Tuple of
tuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.
If
past_key_values
are used, the user can optionally input only the lastdecoder_input_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of alldecoder_input_ids
of shape(batch_size, sequence_length)
. - decoder_inputs_embeds (
torch.FloatTensor
of shape(batch_size, target_sequence_length, hidden_size)
, optional) — Optionally, instead of passingdecoder_input_ids
you can choose to directly pass an embedded representation. Ifpast_key_values
is used, optionally only the lastdecoder_inputs_embeds
have to be input (seepast_key_values
). This is useful if you want more control over how to convertdecoder_input_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - use_cache (
bool
, optional) — If set toTrue
,past_key_values
key value states are returned and can be used to speed up decoding (seepast_key_values
). - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - cache_position (
torch.LongTensor
of shape(sequence_length)
, optional) — Indices depicting the position of the input sequence tokens in the sequence. It is used to update the cache in the correct position and to infer the complete sequence length.
Returns
transformers.modeling_outputs.Seq2SeqModelOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.Seq2SeqModelOutput or a tuple of
torch.FloatTensor
(if return_dict=False
is passed or when config.return_dict=False
) comprising various
elements depending on the configuration (WhisperConfig) and inputs.
-
last_hidden_state (
torch.FloatTensor
of shape(batch_size, sequence_length, hidden_size)
) — Sequence of hidden-states at the output of the last layer of the decoder of the model.If
past_key_values
is used only the last hidden-state of the sequences of shape(batch_size, 1, hidden_size)
is output. -
past_key_values (
tuple(tuple(torch.FloatTensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding. -
decoder_hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftorch.FloatTensor
(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the decoder at the output of each layer plus the optional initial embedding outputs.
-
decoder_attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
encoder_last_hidden_state (
torch.FloatTensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model. -
encoder_hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftorch.FloatTensor
(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the encoder at the output of each layer plus the optional initial embedding outputs.
-
encoder_attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The WhisperModel forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> import torch
>>> from transformers import AutoFeatureExtractor, WhisperModel
>>> from datasets import load_dataset
>>> model = WhisperModel.from_pretrained("openai/whisper-base")
>>> feature_extractor = AutoFeatureExtractor.from_pretrained("openai/whisper-base")
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> inputs = feature_extractor(ds[0]["audio"]["array"], return_tensors="pt")
>>> input_features = inputs.input_features
>>> decoder_input_ids = torch.tensor([[1, 1]]) * model.config.decoder_start_token_id
>>> last_hidden_state = model(input_features, decoder_input_ids=decoder_input_ids).last_hidden_state
>>> list(last_hidden_state.shape)
[1, 2, 512]
Masks extracted features along time axis and/or along feature axis according to SpecAugment.
WhisperForConditionalGeneration
class transformers.WhisperForConditionalGeneration
< source >( config: WhisperConfig )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The Whisper Model with a language modeling head. Can be used for automatic speech recognition. This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( input_features: Optional = None attention_mask: Optional = None decoder_input_ids: Optional = None decoder_attention_mask: Optional = None head_mask: Optional = None decoder_head_mask: Optional = None cross_attn_head_mask: Optional = None encoder_outputs: Optional = None past_key_values: Union = None decoder_inputs_embeds: Optional = None decoder_position_ids: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None cache_position: Optional = None ) → transformers.modeling_outputs.Seq2SeqLMOutput or tuple(torch.FloatTensor)
Parameters
- input_features (
torch.FloatTensor
of shape(batch_size, feature_size, sequence_length)
) — Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the AutoFeatureExtractor should be used for extracting the mel features, padding and conversion into a tensor of typetorch.FloatTensor
. See call() - attention_mask (
torch.LongTensor
of shape(batch_size, sequence_length)
, optional) — Mask to avoid performing SpecAugment data augmentation on padding token indices. Mask values selected in[0, 1]
:- 1 for tokens that are not masked,
- 0 for tokens that are masked.
- decoder_input_ids (
torch.LongTensor
of shape(batch_size, target_sequence_length)
, optional) — Indices of decoder input sequence tokens in the vocabulary.Indices can be obtained using WhisperTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.
Whisper uses the
decoder_start_token_id
as the starting token fordecoder_input_ids
generation. Ifpast_key_values
is used, optionally only the lastdecoder_input_ids
have to be input (seepast_key_values
). - decoder_attention_mask (
torch.LongTensor
of shape(batch_size, target_sequence_length)
, optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids
. Causal mask will also be used by default.If you want to change padding behavior, you should read
modeling_whisper._prepare_decoder_attention_mask
and modify to your needs. See diagram 1 in the BART paper for more information on the default strategy. - head_mask (
torch.Tensor
of shape(encoder_layers, encoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- decoder_head_mask (
torch.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the decoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- cross_attn_head_mask (
torch.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the cross-attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- encoder_outputs (
tuple(tuple(torch.FloatTensor)
, optional) — Tuple consists of (last_hidden_state
, optional:hidden_states
, optional:attentions
)last_hidden_state
of shape(batch_size, sequence_length, hidden_size)
, optional) is a sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of the decoder. - past_key_values (
EncoderDecoderCache
ortuple(tuple(torch.FloatTensor))
, optional) — Pre-computed hidden-states that can be used to speed up auto-regressive (sequential) decoding. There are four sets of pre-computed hidden-states: key and values states in the self-attention blocks (2) and in the cross-attention blocks (2). Thepast_key_values
are returned whenuse_cache=True
is passed or whenconfig.use_cache=True
Two formats are allowed:
- An EncoderDecoderCache instance;
- Tuple of
tuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.
If
past_key_values
are used, the user can optionally input only the lastdecoder_input_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of alldecoder_input_ids
of shape(batch_size, sequence_length)
. - decoder_inputs_embeds (
torch.FloatTensor
of shape(batch_size, target_sequence_length, hidden_size)
, optional) — Optionally, instead of passingdecoder_input_ids
you can choose to directly pass an embedded representation. Ifpast_key_values
is used, optionally only the lastdecoder_inputs_embeds
have to be input (seepast_key_values
). This is useful if you want more control over how to convertdecoder_input_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - use_cache (
bool
, optional) — If set toTrue
,past_key_values
key value states are returned and can be used to speed up decoding (seepast_key_values
). - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - cache_position (
torch.LongTensor
of shape(sequence_length)
, optional) — Indices depicting the position of the input sequence tokens in the sequence. It is used to update the cache in the correct position and to infer the complete sequence length. - labels (
torch.LongTensor
of shape(batch_size, sequence_length)
, optional) — Labels for computing the language modeling loss. Indices should either be in[0, ..., config.vocab_size]
or -100 (seeinput_ids
docstring). Tokens with indices set to-100
are ignored (masked), the loss is only computed for the tokens with labels in[0, ..., config.vocab_size]
.
Returns
transformers.modeling_outputs.Seq2SeqLMOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.Seq2SeqLMOutput or a tuple of
torch.FloatTensor
(if return_dict=False
is passed or when config.return_dict=False
) comprising various
elements depending on the configuration (WhisperConfig) and inputs.
-
loss (
torch.FloatTensor
of shape(1,)
, optional, returned whenlabels
is provided) — Language modeling loss. -
logits (
torch.FloatTensor
of shape(batch_size, sequence_length, config.vocab_size)
) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). -
past_key_values (
tuple(tuple(torch.FloatTensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding. -
decoder_hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftorch.FloatTensor
(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
-
decoder_attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
encoder_last_hidden_state (
torch.FloatTensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model. -
encoder_hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftorch.FloatTensor
(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
-
encoder_attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The WhisperForConditionalGeneration forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> import torch
>>> from transformers import AutoProcessor, WhisperForConditionalGeneration
>>> from datasets import load_dataset
>>> processor = AutoProcessor.from_pretrained("openai/whisper-tiny.en")
>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> inputs = processor(ds[0]["audio"]["array"], return_tensors="pt")
>>> input_features = inputs.input_features
>>> generated_ids = model.generate(inputs=input_features)
>>> transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
>>> transcription
' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
generate
< source >( input_features: Optional = None generation_config: Optional = None logits_processor: Optional = None stopping_criteria: Optional = None prefix_allowed_tokens_fn: Optional = None synced_gpus: bool = False return_timestamps: Optional = None task: Optional = None language: Union = None is_multilingual: Optional = None prompt_ids: Optional = None prompt_condition_type: Optional = None condition_on_prev_tokens: Optional = None temperature: Union = None compression_ratio_threshold: Optional = None logprob_threshold: Optional = None no_speech_threshold: Optional = None num_segment_frames: Optional = None attention_mask: Optional = None time_precision: float = 0.02 return_token_timestamps: Optional = None return_segments: bool = False return_dict_in_generate: Optional = None **kwargs ) → ModelOutput or torch.LongTensor
or Dict[str, Any]
Parameters
- input_features (
torch.Tensor
of shape(batch_size, feature_size, sequence_length)
, optional) — Float values of log-mel features extracted from the raw speech waveform. The raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the AutoFeatureExtractor should be used for extracting the mel features, padding and conversion into a tensor of typetorch.FloatTensor
. See call() for details. - generation_config (
~generation.GenerationConfig
, optional) — The generation configuration to be used as base parametrization for the generation call.**kwargs
passed to generate matching the attributes ofgeneration_config
will override them. Ifgeneration_config
is not provided, the default will be used, which had the following loading priority: 1) from thegeneration_config.json
model file, if it exists; 2) from the model configuration. Please note that unspecified parameters will inherit GenerationConfig’s default values, whose documentation should be checked to parameterize generation. - logits_processor (
LogitsProcessorList
, optional) — Custom logits processors that complement the default logits processors built from arguments and generation config. If a logit processor is passed that is already created with the arguments or a generation config an error is thrown. This feature is intended for advanced users. - stopping_criteria (
StoppingCriteriaList
, optional) — Custom stopping criteria that complement the default stopping criteria built from arguments and a generation config. If a stopping criteria is passed that is already created with the arguments or a generation config an error is thrown. This feature is intended for advanced users. - prefix_allowed_tokens_fn (
Callable[[int, torch.Tensor], List[int]]
, optional) — If provided, this function constraints the beam search to allowed tokens only at each step. If not provided no constraint is applied. This function takes 2 arguments: the batch IDbatch_id
andinput_ids
. It has to return a list with the allowed tokens for the next generation step conditioned on the batch IDbatch_id
and the previously generated tokensinputs_ids
. This argument is useful for constrained generation conditioned on the prefix, as described in Autoregressive Entity Retrieval. - synced_gpus (
bool
, optional, defaults toFalse
) — Whether to continue running the while loop until max_length (needed for ZeRO stage 3) - return_timestamps (
bool
, optional) — Whether to return the timestamps with the text. This enables theWhisperTimestampsLogitsProcessor
. - task (
str
, optional) — Task to use for generation, either “translate” or “transcribe”. Themodel.config.forced_decoder_ids
will be updated accordingly. - language (
str
or list ofstr
, optional) — Language token to use for generation, can be either in the form of<|en|>
,en
orenglish
. For batched generation, a list of language tokens can be passed. You can find all the possible language tokens in themodel.generation_config.lang_to_id
dictionary. - is_multilingual (
bool
, optional) — Whether or not the model is multilingual. - prompt_ids (
torch.Tensor
, optional) — Rank-1 tensor of token IDs created by passing text toget_prompt_ids()
that is provided as a prompt to each chunk. This can be used to provide or “prompt-engineer” a context for transcription, e.g. custom vocabularies or proper nouns to make it more likely to predict those words correctly. It cannot be used in conjunction withdecoder_start_token_id
as it overwrites this value. - prompt_condition_type (
str
, optional) — Only relevant for long-form transcription. Condition type ofprompt_ids
. ‘first-segment’ means only the first segment is conditioned onprompt_ids
. ‘all-segments’ means each segment is conditioned onprompt_ids
. Make sure to enablecondition_on_prev_tokens
for ‘all-segments’. Defaults to ‘first-segment’. For short-term transcription only ‘first-segment’ is possible. - condition_on_prev_tokens (
bool
, optional) — Only relevant for long-form transcription. Whether to condition each segment on the previous segment. As shown in the the Whisper paper, this can help to improve performance. - temperature (
float
or list offloat
, optional) — The temperature to be used for generation. Passing a singlefloat
value anddo_sample=True
activates generation using sampling. For long-form transcription, temperature fallback can be activated by passing a list of float values such as (0.0, 0.2, 0.4, 0.6, 0.8, 1.0). As shown in the the Whisper paper, this can help to improve performance. - compression_ratio_threshold (
float
, optional) — Only relevant for long-form transcription. If defined, the zlib compression rate of each segment will be computed. If the compression rate of a segment is higher thancompression_ratio_threshold
, temperature fallback is activated: the generated segment is discarded and the generation is repeated using a higher temperature. The intuition behind this feature is that segments with very high compression rates suffer from a lot of repetition. The unwanted repetition can be reduced by injecting more randomness by increasing the temperature. Ifcompression_ratio_threshold
is defined make sure thattemperature
is a list of values. A common value forcompression_ratio_threshold
is 1.35. As shown in the the Whisper paper, this can help to improve performance. - logprob_threshold (
float
, optional) — Only relevant for long-form transcription. If defined, the average log-probability of each segment will be computed. If the log-probability of a given segment is lower thanlogprob_threshold
, temperature fallback is activated: the generated segment is discarded and the generation is repeated using a higher temperature. The intuition behind this feature is that segments of low log-probability can be improved by injecting more randomness by increasing the temperature. Iflogprob_threshold
is defined make sure thattemperature
is a list of values. A common value forlogprob_threshold
is -1.0. As shown in the the Whisper paper, this can help to improve performance. - no_speech_threshold (
float
, optional) — Only relevant for long-form transcription. If defined, the “no-speech” token combined with thelogprob_threshold
is used to determine whether a segment contains only silence. In this case, the transcription for this segment is skipped. As shown in the the Whisper paper, this can help to improve performance. - num_segment_frames (
int
, optional) — The number of frames a single segment is made of. If not defined,num_segment_frames
defaults to the model’s stride times the maximum input length. - attention_mask (
torch.Tensor
, optional) —attention_mask
needs to be passed when doing long-form transcription using a batch size > 1. - time_precision (
int
, optional, defaults to 0.02) — The duration of output token in seconds. E.g. 0.02 means that a generated token on average accounts for 20 ms. - return_token_timestamps (
bool
, optional) — Whether to return token-level timestamps with the text. This can be used with or without thereturn_timestamps
option. To get word-level timestamps, use the tokenizer to group the tokens into words. - return_segments (
bool
, optional, defaults toFalse
) — Whether to additionally return a list of all segments. Note that this option can only be enabled when doing long-form transcription. - return_dict_in_generate (
bool
, optional, defaults toFalse
) — Whether or not to return a ModelOutput instead of just returning the generated tokens. Note that when doing long-form transcription,return_dict_in_generate
can only be enabled whenreturn_segments
is set True. In this case the generation outputs of each segment is added to each segment. - kwargs (
Dict[str, Any]
, optional) — Ad hoc parametrization ofgenerate_config
and/or additional model-specific kwargs that will be forwarded to theforward
function of the model. If the model is an encoder-decoder model, encoder specific kwargs should not be prefixed and decoder specific kwargs should be prefixed with decoder_.
Returns
ModelOutput or torch.LongTensor
or Dict[str, Any]
A ModelOutput (if return_dict_in_generate=True
or when config.return_dict_in_generate=True
) or a torch.FloatTensor
or a dict of segments when return_segments=True
.
If the passed input is > 30 seconds / > 3000 mel input features and return_segments=True
then a dictionary of generated sequence ids, called sequences
and a list of each generated segment is returned.
else if the passed input is <= 30 seconds / >= 3000 mel input features, the possible ModelOutput types are:
else only the generated output sequence ids are returned.
Transcribes or translates log-mel input features to a sequence of auto-regressively generated token ids.
Most generation-controlling parameters are set in generation_config
which, if not passed, will be set to the
model’s default generation configuration. You can override any generation_config
by passing the corresponding
parameters to generate(), e.g. .generate(inputs, num_beams=4, do_sample=True)
.
For an overview of generation strategies and code examples, check out the following guide.
Example:
- Longform transcription: To transcribe or translate audios longer than 30 seconds, process the audio files without truncation and pass all mel features at once to generate.
>>> import torch
>>> from transformers import AutoProcessor, WhisperForConditionalGeneration
>>> from datasets import load_dataset, Audio
>>> processor = AutoProcessor.from_pretrained("openai/whisper-tiny.en")
>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
>>> model.cuda()
>>> # load audios > 30 seconds
>>> ds = load_dataset("distil-whisper/meanwhile", "default")["test"]
>>> # resample to 16kHz
>>> ds = ds.cast_column("audio", Audio(sampling_rate=16000))
>>> # take first 8 audios and retrieve array
>>> audio = ds[:8]["audio"]
>>> audio = [x["array"] for x in audio]
>>> # make sure to NOT truncate the input audio, to return the `attention_mask` and to pad to the longest audio
>>> inputs = processor(audio, return_tensors="pt", truncation=False, padding="longest", return_attention_mask=True, sampling_rate=16_000)
>>> inputs = inputs.to("cuda", torch.float32)
>>> # transcribe audio to ids
>>> generated_ids = model.generate(**inputs)
>>> transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)
>>> transcription[0]
" Folks, if you watch the show, you know, I spent a lot of time right over there. Patiently and astutely scrutinizing the boxwood and mahogany chest set of the day's biggest stories developing the central headline pawns, definitely maneuvering an oso topical night to F6, fainting a classic Sicilian, nade door variation on the news, all the while seeing eight moves deep and patiently marshalling the latest press releases into a fisher's shows in Lip Nitsky attack that culminates in the elegant lethal slow-played, all-passant checkmate that is my nightly monologue. But sometimes, sometimes, folks, I. CHEERING AND APPLAUSE Sometimes I startle away, cubside down in the monkey bars of a condemned playground on a super fun site. Get all hept up on goofballs. Rummage that were discarded tag bag of defective toys. Yank out a fist bowl of disembodied doll limbs, toss them on a stained kid's place mat from a defunct dennies. set up a table inside a rusty cargo container down by the Wharf and challenged toothless drifters to the godless bughouse blitz of tournament that is my segment. Meanwhile."
- Shortform transcription: If passed mel input features are < 30 seconds, the whole audio will be transcribed with a single call to generate.
>>> import torch
>>> from transformers import AutoProcessor, WhisperForConditionalGeneration
>>> from datasets import load_dataset
>>> processor = AutoProcessor.from_pretrained("openai/whisper-tiny.en")
>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> inputs = processor(ds[0]["audio"]["array"], return_tensors="pt")
>>> input_features = inputs.input_features
>>> generated_ids = model.generate(inputs=input_features)
>>> transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
>>> transcription
' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
WhisperForCausalLM
class transformers.WhisperForCausalLM
< source >( config )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
Whisper decoder with a language modeling head on top (linear layer with weights tied to the input embeddings).
This model inherits from PreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >( input_ids: LongTensor = None attention_mask: Optional = None encoder_outputs: Optional = None head_mask: Optional = None cross_attn_head_mask: Optional = None past_key_values: Optional = None inputs_embeds: Optional = None labels: Optional = None use_cache: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None cache_position: Optional = None ) → transformers.modeling_outputs.CausalLMOutputWithCrossAttentions or tuple(torch.FloatTensor)
Parameters
- input_ids (
torch.LongTensor
of shape(batch_size, sequence_length)
) — Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide it. Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details. What are input IDs? - attention_mask (
torch.Tensor
of shape(batch_size, sequence_length)
, optional) — Mask to avoid performing attention on padding token indices. Mask values selected in[0, 1]
:- 1 for tokens that are not masked,
- 0 for tokens that are masked. What are attention masks?
- encoder_outputs (
torch.FloatTensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention if the model is configured as a decoder. - head_mask (
torch.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- cross_attn_head_mask (
torch.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the cross-attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- past_key_values (
tuple(tuple(torch.FloatTensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(torch.FloatTensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
. The two additional tensors are only required when the model is used as a decoder in a Sequence to Sequence model. Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (seepast_key_values
input) to speed up sequential decoding. Ifpast_key_values
are used, the user can optionally input only the lastdecoder_input_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of alldecoder_input_ids
of shape(batch_size, sequence_length)
. - inputs_embeds (
torch.FloatTensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Optionally, instead of passinginput_ids
you can choose to directly pass an embedded representation. This is useful if you want more control over how to convertinput_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - labels (
torch.LongTensor
of shape(batch_size, sequence_length)
, optional) — Labels for computing the masked language modeling loss. Indices should either be in[0, ..., config.vocab_size]
or -100 (seeinput_ids
docstring). Tokens with indices set to-100
are ignored (masked), the loss is only computed for the tokens with labels in[0, ..., config.vocab_size]
. - use_cache (
bool
, optional) — If set toTrue
,past_key_values
key value states are returned and can be used to speed up decoding (seepast_key_values
).- 1 for tokens that are not masked,
- 0 for tokens that are masked.
- output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - cache_position (
torch.LongTensor
of shape(sequence_length)
, optional) — Indices depicting the position of the input sequence tokens in the sequence. It is used to update the cache in the correct position and to infer the complete sequence length.
Returns
transformers.modeling_outputs.CausalLMOutputWithCrossAttentions or tuple(torch.FloatTensor)
A transformers.modeling_outputs.CausalLMOutputWithCrossAttentions or a tuple of
torch.FloatTensor
(if return_dict=False
is passed or when config.return_dict=False
) comprising various
elements depending on the configuration (WhisperConfig) and inputs.
-
loss (
torch.FloatTensor
of shape(1,)
, optional, returned whenlabels
is provided) — Language modeling loss (for next-token prediction). -
logits (
torch.FloatTensor
of shape(batch_size, sequence_length, config.vocab_size)
) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). -
hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftorch.FloatTensor
(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
-
attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Cross attentions weights after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
past_key_values (
tuple(tuple(torch.FloatTensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftorch.FloatTensor
tuples of lengthconfig.n_layers
, with each tuple containing the cached key, value states of the self-attention and the cross-attention layers if model is used in encoder-decoder setting. Only relevant ifconfig.is_decoder = True
.Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding.
Example:
>>> from transformers import WhisperForCausalLM, WhisperForConditionalGeneration, WhisperProcessor
>>> import torch
>>> from datasets import load_dataset
>>> processor = WhisperProcessor.from_pretrained("openai/whisper-large-v2")
>>> model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v2")
>>> assistant_model = WhisperForCausalLM.from_pretrained("distil-whisper/distil-large-v2")
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> sample = ds[0]["audio"]
>>> input_features = processor(
... sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt"
... ).input_features
>>> predicted_ids = model.generate(input_features, assistant_model=assistant_model)
>>> # decode token ids to text
>>> transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
>>> transcription
' Mr. Quilter is the apostle of the middle classes and we are glad to welcome his gospel.'
WhisperForAudioClassification
class transformers.WhisperForAudioClassification
< source >( config )
Parameters
- input_features (
torch.FloatTensor
of shape(batch_size, feature_size, sequence_length)
) — Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the AutoFeatureExtractor should be used for extracting the mel features, padding and conversion into a tensor of typetorch.FloatTensor
. See call() - head_mask (
torch.Tensor
of shape(encoder_layers, encoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- encoder_outputs (
tuple(tuple(torch.FloatTensor)
, optional) — Tuple consists of (last_hidden_state
, optional:hidden_states
, optional:attentions
)last_hidden_state
of shape(batch_size, sequence_length, hidden_size)
, optional) is a sequence of hidden-states at the output of the last layer of the encoder. - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
Whisper Encoder Model with a sequence classification head on top (a linear layer over the pooled output) for tasks like SUPERB Keyword Spotting.
forward
< source >( input_features: Optional = None head_mask: Optional = None encoder_outputs: Optional = None labels: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None ) → transformers.modeling_outputs.SequenceClassifierOutput or tuple(torch.FloatTensor)
Parameters
- input_features (
torch.FloatTensor
of shape(batch_size, feature_size, sequence_length)
) — Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the AutoFeatureExtractor should be used for extracting the mel features, padding and conversion into a tensor of typetorch.FloatTensor
. See call() - head_mask (
torch.Tensor
of shape(encoder_layers, encoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- encoder_outputs (
tuple(tuple(torch.FloatTensor)
, optional) — Tuple consists of (last_hidden_state
, optional:hidden_states
, optional:attentions
)last_hidden_state
of shape(batch_size, sequence_length, hidden_size)
, optional) is a sequence of hidden-states at the output of the last layer of the encoder. - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - labels (
torch.LongTensor
of shape(batch_size,)
, optional) — Labels for computing the sequence classification/regression loss. Indices should be in[0, ..., config.num_labels - 1]
. Ifconfig.num_labels == 1
a regression loss is computed (Mean-Square loss), Ifconfig.num_labels > 1
a classification loss is computed (Cross-Entropy).
Returns
transformers.modeling_outputs.SequenceClassifierOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.SequenceClassifierOutput or a tuple of
torch.FloatTensor
(if return_dict=False
is passed or when config.return_dict=False
) comprising various
elements depending on the configuration (WhisperConfig) and inputs.
-
loss (
torch.FloatTensor
of shape(1,)
, optional, returned whenlabels
is provided) — Classification (or regression if config.num_labels==1) loss. -
logits (
torch.FloatTensor
of shape(batch_size, config.num_labels)
) — Classification (or regression if config.num_labels==1) scores (before SoftMax). -
hidden_states (
tuple(torch.FloatTensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftorch.FloatTensor
(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
-
attentions (
tuple(torch.FloatTensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftorch.FloatTensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
The WhisperForAudioClassification forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> import torch
>>> from transformers import AutoFeatureExtractor, WhisperForAudioClassification
>>> from datasets import load_dataset
>>> feature_extractor = AutoFeatureExtractor.from_pretrained("sanchit-gandhi/whisper-medium-fleurs-lang-id")
>>> model = WhisperForAudioClassification.from_pretrained("sanchit-gandhi/whisper-medium-fleurs-lang-id")
>>> ds = load_dataset("google/fleurs", "all", split="validation", streaming=True)
>>> sample = next(iter(ds))
>>> inputs = feature_extractor(
... sample["audio"]["array"], sampling_rate=sample["audio"]["sampling_rate"], return_tensors="pt"
... )
>>> input_features = inputs.input_features
>>> with torch.no_grad():
... logits = model(input_features).logits
>>> predicted_class_ids = torch.argmax(logits).item()
>>> predicted_label = model.config.id2label[predicted_class_ids]
>>> predicted_label
'Afrikaans'
TFWhisperModel
class transformers.TFWhisperModel
< source >( config: WhisperConfig **kwargs )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The bare Whisper Model outputting raw hidden-states without any specific head on top. This model inherits from TFPreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a keras.Model subclass. Use it as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and behavior.
call
< source >( input_features: TFModelInputType | None = None decoder_input_ids: np.ndarray | tf.Tensor | None = None decoder_attention_mask: np.ndarray | tf.Tensor | None = None decoder_position_ids: np.ndarray | tf.Tensor | None = None head_mask: np.ndarray | tf.Tensor | None = None decoder_head_mask: np.ndarray | tf.Tensor | None = None cross_attn_head_mask: np.ndarray | tf.Tensor | None = None encoder_outputs: Optional[Tuple[Tuple[Union[np.ndarray, tf.Tensor]]]] = None past_key_values: Optional[Tuple[Tuple[Union[np.ndarray, tf.Tensor]]]] = None decoder_inputs_embeds: Optional[Tuple[Union[np.ndarray, tf.Tensor]]] = None use_cache: Optional[bool] = None output_attentions: Optional[bool] = None output_hidden_states: Optional[bool] = None return_dict: Optional[bool] = None training: bool = False ) → transformers.modeling_tf_outputs.TFSeq2SeqModelOutput or tuple(tf.Tensor)
Parameters
- input_features (
tf.Tensor
of shape(batch_size, feature_size, sequence_length)
) — Float values of fbank features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the AutoFeatureExtractor should be used for extracting the fbank features, padding and conversion into a tensor of typetf.Tensor
. See call() - decoder_input_ids (
tf.Tensor
of shape(batch_size, target_sequence_length)
, optional) — Indices of decoder input sequence tokens in the vocabulary.Indices can be obtained using
SpeechToTextTokenizer
. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.SpeechToText uses the
eos_token_id
as the starting token fordecoder_input_ids
generation. Ifpast_key_values
is used, optionally only the lastdecoder_input_ids
have to be input (seepast_key_values
). - decoder_attention_mask (
tf.Tensor
of shape(batch_size, target_sequence_length)
, optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids
. Causal mask will also be used by default.If you want to change padding behavior, you should read
modeling_whisper._prepare_decoder_attention_mask
and modify to your needs. See diagram 1 in the paper for more information on the default strategy. - head_mask (
tf.Tensor
of shape(encoder_layers, encoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- decoder_head_mask (
tf.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the decoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- cross_attn_head_mask (
tf.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the cross-attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- encoder_outputs (
tuple(tuple(tf.Tensor)
, optional) — Tuple consists of (last_hidden_state
, optional:hidden_states
, optional:attentions
)last_hidden_state
of shape(batch_size, sequence_length, hidden_size)
, optional) is a sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of the decoder. - past_key_values (
tuple(tuple(tf.Tensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(tf.Tensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding.If
past_key_values
are used, the user can optionally input only the lastdecoder_input_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of alldecoder_input_ids
of shape(batch_size, sequence_length)
. - decoder_inputs_embeds (
tf.Tensor
of shape(batch_size, target_sequence_length, hidden_size)
, optional) — Optionally, instead of passingdecoder_input_ids
you can choose to directly pass an embedded representation. Ifpast_key_values
is used, optionally only the lastdecoder_inputs_embeds
have to be input (seepast_key_values
). This is useful if you want more control over how to convertdecoder_input_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - use_cache (
bool
, optional) — If set toTrue
,past_key_values
key value states are returned and can be used to speed up decoding (seepast_key_values
). - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
Returns
transformers.modeling_tf_outputs.TFSeq2SeqModelOutput or tuple(tf.Tensor)
A transformers.modeling_tf_outputs.TFSeq2SeqModelOutput or a tuple of tf.Tensor
(if
return_dict=False
is passed or when config.return_dict=False
) comprising various elements depending on the
configuration (WhisperConfig) and inputs.
-
last_hidden_state (
tf.Tensor
of shape(batch_size, sequence_length, hidden_size)
) — Sequence of hidden-states at the output of the last layer of the decoder of the model.If
past_key_values
is used only the last hidden-state of the sequences of shape(batch_size, 1, hidden_size)
is output. -
past_key_values (
List[tf.Tensor]
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — List oftf.Tensor
of lengthconfig.n_layers
, with each tensor of shape(2, batch_size, num_heads, sequence_length, embed_size_per_head)
).Contains pre-computed hidden-states (key and values in the attention blocks) of the decoder that can be used (see
past_key_values
input) to speed up sequential decoding. -
decoder_hidden_states (
tuple(tf.Tensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftf.Tensor
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
-
decoder_attentions (
tuple(tf.Tensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftf.Tensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(tf.Tensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftf.Tensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
encoder_last_hidden_state (
tf.Tensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model. -
encoder_hidden_states (
tuple(tf.Tensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftf.Tensor
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
-
encoder_attentions (
tuple(tf.Tensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftf.Tensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The TFWhisperModel forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> import tensorflow as tf
>>> from transformers import TFWhisperModel, AutoFeatureExtractor
>>> from datasets import load_dataset
>>> model = TFWhisperModel.from_pretrained("openai/whisper-base")
>>> feature_extractor = AutoFeatureExtractor.from_pretrained("openai/whisper-base")
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> inputs = feature_extractor(ds[0]["audio"]["array"], return_tensors="tf")
>>> input_features = inputs.input_features
>>> decoder_input_ids = tf.convert_to_tensor([[1, 1]]) * model.config.decoder_start_token_id
>>> last_hidden_state = model(input_features, decoder_input_ids=decoder_input_ids).last_hidden_state
>>> list(last_hidden_state.shape)
[1, 2, 512]
TFWhisperForConditionalGeneration
class transformers.TFWhisperForConditionalGeneration
< source >( config: WhisperConfig **kwargs )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The Whisper Model with a language modeling head. Can be used for automatic speech recognition. This model inherits from TFPreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads etc.)
This model is also a keras.Model subclass. Use it as a regular TF 2.0 Keras Model and refer to the TF 2.0 documentation for all matter related to general usage and behavior.
call
< source >( input_features: TFModelInputType | None = None decoder_input_ids: np.ndarray | tf.Tensor | None = None decoder_attention_mask: np.ndarray | tf.Tensor | None = None decoder_position_ids: np.ndarray | tf.Tensor | None = None head_mask: np.ndarray | tf.Tensor | None = None decoder_head_mask: np.ndarray | tf.Tensor | None = None cross_attn_head_mask: np.ndarray | tf.Tensor | None = None encoder_outputs: Optional[Tuple[Tuple[Union[np.ndarray, tf.Tensor]]]] = None past_key_values: Optional[Tuple[Tuple[Union[np.ndarray, tf.Tensor]]]] = None decoder_inputs_embeds: Optional[Tuple[Union[np.ndarray, tf.Tensor]]] = None labels: np.ndarray | tf.Tensor | None = None use_cache: Optional[bool] = None output_attentions: Optional[bool] = None output_hidden_states: Optional[bool] = None return_dict: Optional[bool] = None training: bool = False ) → transformers.modeling_tf_outputs.TFSeq2SeqLMOutput or tuple(tf.Tensor)
Parameters
- input_features (
tf.Tensor
of shape(batch_size, feature_size, sequence_length)
) — Float values of fbank features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the AutoFeatureExtractor should be used for extracting the fbank features, padding and conversion into a tensor of typetf.Tensor
. See call() - decoder_input_ids (
tf.Tensor
of shape(batch_size, target_sequence_length)
, optional) — Indices of decoder input sequence tokens in the vocabulary.Indices can be obtained using
SpeechToTextTokenizer
. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.SpeechToText uses the
eos_token_id
as the starting token fordecoder_input_ids
generation. Ifpast_key_values
is used, optionally only the lastdecoder_input_ids
have to be input (seepast_key_values
). - decoder_attention_mask (
tf.Tensor
of shape(batch_size, target_sequence_length)
, optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids
. Causal mask will also be used by default.If you want to change padding behavior, you should read
modeling_whisper._prepare_decoder_attention_mask
and modify to your needs. See diagram 1 in the paper for more information on the default strategy. - head_mask (
tf.Tensor
of shape(encoder_layers, encoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the encoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- decoder_head_mask (
tf.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the attention modules in the decoder. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- cross_attn_head_mask (
tf.Tensor
of shape(decoder_layers, decoder_attention_heads)
, optional) — Mask to nullify selected heads of the cross-attention modules. Mask values selected in[0, 1]
:- 1 indicates the head is not masked,
- 0 indicates the head is masked.
- encoder_outputs (
tuple(tuple(tf.Tensor)
, optional) — Tuple consists of (last_hidden_state
, optional:hidden_states
, optional:attentions
)last_hidden_state
of shape(batch_size, sequence_length, hidden_size)
, optional) is a sequence of hidden-states at the output of the last layer of the encoder. Used in the cross-attention of the decoder. - past_key_values (
tuple(tuple(tf.Tensor))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(tf.Tensor)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding.If
past_key_values
are used, the user can optionally input only the lastdecoder_input_ids
(those that don’t have their past key value states given to this model) of shape(batch_size, 1)
instead of alldecoder_input_ids
of shape(batch_size, sequence_length)
. - decoder_inputs_embeds (
tf.Tensor
of shape(batch_size, target_sequence_length, hidden_size)
, optional) — Optionally, instead of passingdecoder_input_ids
you can choose to directly pass an embedded representation. Ifpast_key_values
is used, optionally only the lastdecoder_inputs_embeds
have to be input (seepast_key_values
). This is useful if you want more control over how to convertdecoder_input_ids
indices into associated vectors than the model’s internal embedding lookup matrix. - use_cache (
bool
, optional) — If set toTrue
,past_key_values
key value states are returned and can be used to speed up decoding (seepast_key_values
). - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple. - labels (
tf.Tensor
of shape(batch_size, sequence_length)
, optional) — Labels for computing the language modeling loss. Indices should either be in[0, ..., config.vocab_size]
or -100 (seeinput_ids
docstring). Tokens with indices set to-100
are ignored (masked), the loss is only computed for the tokens with labels in[0, ..., config.vocab_size]
.
Returns
transformers.modeling_tf_outputs.TFSeq2SeqLMOutput or tuple(tf.Tensor)
A transformers.modeling_tf_outputs.TFSeq2SeqLMOutput or a tuple of tf.Tensor
(if
return_dict=False
is passed or when config.return_dict=False
) comprising various elements depending on the
configuration (WhisperConfig) and inputs.
-
loss (
tf.Tensor
of shape(n,)
, optional, where n is the number of non-masked labels, returned whenlabels
is provided) — Language modeling loss. -
logits (
tf.Tensor
of shape(batch_size, sequence_length, config.vocab_size)
) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). -
past_key_values (
List[tf.Tensor]
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — List oftf.Tensor
of lengthconfig.n_layers
, with each tensor of shape(2, batch_size, num_heads, sequence_length, embed_size_per_head)
).Contains pre-computed hidden-states (key and values in the attention blocks) of the decoder that can be used (see
past_key_values
input) to speed up sequential decoding. -
decoder_hidden_states (
tuple(tf.Tensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftf.Tensor
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
-
decoder_attentions (
tuple(tf.Tensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftf.Tensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(tf.Tensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftf.Tensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
encoder_last_hidden_state (
tf.Tensor
of shape(batch_size, sequence_length, hidden_size)
, optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model. -
encoder_hidden_states (
tuple(tf.Tensor)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple oftf.Tensor
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
-
encoder_attentions (
tuple(tf.Tensor)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple oftf.Tensor
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The TFWhisperForConditionalGeneration forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> import tensorflow as tf
>>> from transformers import AutoProcessor, TFWhisperForConditionalGeneration
>>> from datasets import load_dataset
>>> processor = AutoProcessor.from_pretrained("openai/whisper-tiny.en")
>>> model = TFWhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en")
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> inputs = processor(ds[0]["audio"]["array"], return_tensors="tf")
>>> input_features = inputs.input_features
>>> generated_ids = model.generate(input_features=input_features)
>>> transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
>>> transcription
' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
FlaxWhisperModel
class transformers.FlaxWhisperModel
< source >( config: WhisperConfig input_shape: Tuple = None seed: int = 0 dtype: dtype = <class 'jax.numpy.float32'> _do_init: bool = True gradient_checkpointing: bool = False **kwargs )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
- dtype (
jax.numpy.dtype
, optional, defaults tojax.numpy.float32
) — The data type of the computation. Can be one ofjax.numpy.float32
,jax.numpy.float16
(on GPUs) andjax.numpy.bfloat16
(on TPUs). This can be used to enable mixed-precision training or half-precision inference on GPUs or TPUs. If specified all the computation will be performed with the givendtype
. Note that this only specifies the dtype of the computation and does not influence the dtype of model parameters. If you wish to change the dtype of the model parameters, see to_fp16() and to_bf16().
The bare Whisper Model transformer outputting raw hidden-states without any specific head on top. This model inherits from FlaxPreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its models (such as downloading or saving, resizing the input embeddings, pruning heads etc.) This model is also a Flax Linen flax.nn.Module subclass. Use it as a regular Flax Module and refer to the Flax documentation for all matter related to general usage and behavior. Finally, this model supports inherent JAX features such as:
__call__
< source >( input_features: Array decoder_input_ids: Array attention_mask: Optional = None decoder_attention_mask: Optional = None position_ids: Optional = None decoder_position_ids: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None train: bool = False params: dict = None dropout_rng: PRNGKey = None ) → transformers.modeling_flax_outputs.FlaxSeq2SeqModelOutput or tuple(torch.FloatTensor)
Parameters
- input_features (
numpy.ndarray
of shape(batch_size, feature_size, sequence_length)
) — Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the WhisperFeatureExtractor should be used for extracting the features, padding and conversion into a tensor of typenumpy.ndarray
. See call() - attention_mask (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Whisper does not support masking of theinput_features
, this argument is preserved for compatibility, but is not used. By default the silence in the input log mel spectrogram are ignored. - decoder_input_ids (
numpy.ndarray
of shape(batch_size, target_sequence_length)
, optional) — Indices of decoder input sequence tokens in the vocabulary. Indices can be obtained using WhisperTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details. What are decoder input IDs? Whisper uses thedecoder_start_token_id
as the starting token fordecoder_input_ids
generation. - decoder_attention_mask (
numpy.ndarray
of shape(batch_size, target_sequence_length)
, optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids
. Causal mask will also be used by default. If you want to change padding behavior, you should modify to your needs. See diagram 1 in the paper for more information on the default strategy. - position_ids (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Whisper does not useposition_ids
in the encoder asinput_features
is always the same size and doesn’t use masking, but this argument is preserved for compatibility. By default the silence in the input log mel spectrogram are ignored. - decoder_position_ids (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Indices of positions of each decoder input sequence tokens in the position embeddings. Selected in the range[0, config.max_position_embeddings - 1]
. - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
Returns
transformers.modeling_flax_outputs.FlaxSeq2SeqModelOutput or tuple(torch.FloatTensor)
A transformers.modeling_flax_outputs.FlaxSeq2SeqModelOutput or a tuple of
torch.FloatTensor
(if return_dict=False
is passed or when config.return_dict=False
) comprising various
elements depending on the configuration (WhisperConfig) and inputs.
-
last_hidden_state (
jnp.ndarray
of shape(batch_size, sequence_length, hidden_size)
) — Sequence of hidden-states at the output of the last layer of the decoder of the model.If
past_key_values
is used only the last hidden-state of the sequences of shape(batch_size, 1, hidden_size)
is output. -
past_key_values (
tuple(tuple(jnp.ndarray))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(jnp.ndarray)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding. -
decoder_hidden_states (
tuple(jnp.ndarray)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple ofjnp.ndarray
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
-
decoder_attentions (
tuple(jnp.ndarray)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple ofjnp.ndarray
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(jnp.ndarray)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple ofjnp.ndarray
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
encoder_last_hidden_state (
jnp.ndarray
of shape(batch_size, sequence_length, hidden_size)
, optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model. -
encoder_hidden_states (
tuple(jnp.ndarray)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple ofjnp.ndarray
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
-
encoder_attentions (
tuple(jnp.ndarray)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple ofjnp.ndarray
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The FlaxWhisperPreTrainedModel
forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Example:
>>> from transformers import AutoTokenizer, FlaxWhisperModel
>>> tokenizer = AutoTokenizer.from_pretrained("openai/whisper-tiny")
>>> model = FlaxWhisperModel.from_pretrained("openai/whisper-tiny")
>>> inputs = tokenizer("Hello, my dog is cute", return_tensors="jax")
>>> outputs = model(**inputs)
>>> last_hidden_states = outputs.last_hidden_state
FlaxWhisperForConditionalGeneration
class transformers.FlaxWhisperForConditionalGeneration
< source >( config: WhisperConfig input_shape: Tuple = None seed: int = 0 dtype: dtype = <class 'jax.numpy.float32'> _do_init: bool = True gradient_checkpointing: bool = False **kwargs )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
- dtype (
jax.numpy.dtype
, optional, defaults tojax.numpy.float32
) — The data type of the computation. Can be one ofjax.numpy.float32
,jax.numpy.float16
(on GPUs) andjax.numpy.bfloat16
(on TPUs). This can be used to enable mixed-precision training or half-precision inference on GPUs or TPUs. If specified all the computation will be performed with the givendtype
. Note that this only specifies the dtype of the computation and does not influence the dtype of model parameters. If you wish to change the dtype of the model parameters, see to_fp16() and to_bf16().
The Whisper Model with a language modeling head. This model inherits from FlaxPreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its models (such as downloading or saving, resizing the input embeddings, pruning heads etc.) This model is also a Flax Linen flax.nn.Module subclass. Use it as a regular Flax Module and refer to the Flax documentation for all matter related to general usage and behavior. Finally, this model supports inherent JAX features such as:
__call__
< source >( input_features: Array decoder_input_ids: Array attention_mask: Optional = None decoder_attention_mask: Optional = None position_ids: Optional = None decoder_position_ids: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None train: bool = False params: dict = None dropout_rng: PRNGKey = None ) → transformers.modeling_flax_outputs.FlaxSeq2SeqLMOutput or tuple(torch.FloatTensor)
Parameters
- input_features (
numpy.ndarray
of shape(batch_size, feature_size, sequence_length)
) — Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the WhisperFeatureExtractor should be used for extracting the features, padding and conversion into a tensor of typenumpy.ndarray
. See call() - attention_mask (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Whisper does not support masking of theinput_features
, this argument is preserved for compatibility, but is not used. By default the silence in the input log mel spectrogram are ignored. - decoder_input_ids (
numpy.ndarray
of shape(batch_size, target_sequence_length)
, optional) — Indices of decoder input sequence tokens in the vocabulary. Indices can be obtained using WhisperTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details. What are decoder input IDs? Whisper uses thedecoder_start_token_id
as the starting token fordecoder_input_ids
generation. - decoder_attention_mask (
numpy.ndarray
of shape(batch_size, target_sequence_length)
, optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids
. Causal mask will also be used by default. If you want to change padding behavior, you should modify to your needs. See diagram 1 in the paper for more information on the default strategy. - position_ids (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Whisper does not useposition_ids
in the encoder asinput_features
is always the same size and doesn’t use masking, but this argument is preserved for compatibility. By default the silence in the input log mel spectrogram are ignored. - decoder_position_ids (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Indices of positions of each decoder input sequence tokens in the position embeddings. Selected in the range[0, config.max_position_embeddings - 1]
. - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
Returns
transformers.modeling_flax_outputs.FlaxSeq2SeqLMOutput or tuple(torch.FloatTensor)
A transformers.modeling_flax_outputs.FlaxSeq2SeqLMOutput or a tuple of
torch.FloatTensor
(if return_dict=False
is passed or when config.return_dict=False
) comprising various
elements depending on the configuration (WhisperConfig) and inputs.
-
logits (
jnp.ndarray
of shape(batch_size, sequence_length, config.vocab_size)
) — Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax). -
past_key_values (
tuple(tuple(jnp.ndarray))
, optional, returned whenuse_cache=True
is passed or whenconfig.use_cache=True
) — Tuple oftuple(jnp.ndarray)
of lengthconfig.n_layers
, with each tuple having 2 tensors of shape(batch_size, num_heads, sequence_length, embed_size_per_head)
) and 2 additional tensors of shape(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)
.Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used (see
past_key_values
input) to speed up sequential decoding. -
decoder_hidden_states (
tuple(jnp.ndarray)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple ofjnp.ndarray
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the decoder at the output of each layer plus the initial embedding outputs.
-
decoder_attentions (
tuple(jnp.ndarray)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple ofjnp.ndarray
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
-
cross_attentions (
tuple(jnp.ndarray)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple ofjnp.ndarray
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the decoder’s cross-attention layer, after the attention softmax, used to compute the weighted average in the cross-attention heads.
-
encoder_last_hidden_state (
jnp.ndarray
of shape(batch_size, sequence_length, hidden_size)
, optional) — Sequence of hidden-states at the output of the last layer of the encoder of the model. -
encoder_hidden_states (
tuple(jnp.ndarray)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple ofjnp.ndarray
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the encoder at the output of each layer plus the initial embedding outputs.
-
encoder_attentions (
tuple(jnp.ndarray)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple ofjnp.ndarray
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights of the encoder, after the attention softmax, used to compute the weighted average in the self-attention heads.
The FlaxWhisperPreTrainedModel
forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Transcription example:
>>> from transformers import WhisperProcessor, FlaxWhisperForConditionalGeneration
>>> from datasets import load_dataset
>>> processor = WhisperProcessor.from_pretrained("openai/whisper-tiny.en")
>>> model = FlaxWhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny.en", from_pt=True)
>>> ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
>>> inputs = processor(ds[0]["audio"]["array"], return_tensors="np")
>>> input_features = inputs.input_features
>>> generated_ids = model.generate(input_ids=input_features)
>>> transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
>>> transcription
' Mr. Quilter is the apostle of the middle classes, and we are glad to welcome his gospel.'
FlaxWhisperForAudioClassification
class transformers.FlaxWhisperForAudioClassification
< source >( config: WhisperConfig input_shape: Tuple = None seed: int = 0 dtype: dtype = <class 'jax.numpy.float32'> _do_init: bool = True gradient_checkpointing: bool = False **kwargs )
Parameters
- config (WhisperConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
- dtype (
jax.numpy.dtype
, optional, defaults tojax.numpy.float32
) — The data type of the computation. Can be one ofjax.numpy.float32
,jax.numpy.float16
(on GPUs) andjax.numpy.bfloat16
(on TPUs). This can be used to enable mixed-precision training or half-precision inference on GPUs or TPUs. If specified all the computation will be performed with the givendtype
. Note that this only specifies the dtype of the computation and does not influence the dtype of model parameters. If you wish to change the dtype of the model parameters, see to_fp16() and to_bf16().
The Whisper Model with an audio classification head on top. This model inherits from FlaxPreTrainedModel. Check the superclass documentation for the generic methods the library implements for all its models (such as downloading or saving, resizing the input embeddings, pruning heads etc.) This model is also a Flax Linen flax.nn.Module subclass. Use it as a regular Flax Module and refer to the Flax documentation for all matter related to general usage and behavior. Finally, this model supports inherent JAX features such as:
__call__
< source >( input_features: Array attention_mask: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None train: bool = False params: dict = None dropout_rng: PRNGKey = None **kwargs ) → transformers.modeling_flax_outputs.FlaxSequenceClassifierOutput or tuple(torch.FloatTensor)
Parameters
- input_features (
numpy.ndarray
of shape(batch_size, feature_size, sequence_length)
) — Float values mel features extracted from the raw speech waveform. Raw speech waveform can be obtained by loading a.flac
or.wav
audio file into an array of typeList[float]
or anumpy.ndarray
, e.g. via the soundfile library (pip install soundfile
). To prepare the array intoinput_features
, the WhisperFeatureExtractor should be used for extracting the features, padding and conversion into a tensor of typenumpy.ndarray
. See call() - attention_mask (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Whisper does not support masking of theinput_features
, this argument is preserved for compatibility, but is not used. By default the silence in the input log mel spectrogram are ignored. - decoder_input_ids (
numpy.ndarray
of shape(batch_size, target_sequence_length)
, optional) — Indices of decoder input sequence tokens in the vocabulary. Indices can be obtained using WhisperTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details. What are decoder input IDs? Whisper uses thedecoder_start_token_id
as the starting token fordecoder_input_ids
generation. - decoder_attention_mask (
numpy.ndarray
of shape(batch_size, target_sequence_length)
, optional) — Default behavior: generate a tensor that ignores pad tokens indecoder_input_ids
. Causal mask will also be used by default. If you want to change padding behavior, you should modify to your needs. See diagram 1 in the paper for more information on the default strategy. - position_ids (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Whisper does not useposition_ids
in the encoder asinput_features
is always the same size and doesn’t use masking, but this argument is preserved for compatibility. By default the silence in the input log mel spectrogram are ignored. - decoder_position_ids (
numpy.ndarray
of shape(batch_size, sequence_length)
, optional) — Indices of positions of each decoder input sequence tokens in the position embeddings. Selected in the range[0, config.max_position_embeddings - 1]
. - output_attentions (
bool
, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentions
under returned tensors for more detail. - output_hidden_states (
bool
, optional) — Whether or not to return the hidden states of all layers. Seehidden_states
under returned tensors for more detail. - return_dict (
bool
, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
Returns
transformers.modeling_flax_outputs.FlaxSequenceClassifierOutput or tuple(torch.FloatTensor)
A transformers.modeling_flax_outputs.FlaxSequenceClassifierOutput or a tuple of
torch.FloatTensor
(if return_dict=False
is passed or when config.return_dict=False
) comprising various
elements depending on the configuration (WhisperConfig) and inputs.
-
logits (
jnp.ndarray
of shape(batch_size, config.num_labels)
) — Classification (or regression if config.num_labels==1) scores (before SoftMax). -
hidden_states (
tuple(jnp.ndarray)
, optional, returned whenoutput_hidden_states=True
is passed or whenconfig.output_hidden_states=True
) — Tuple ofjnp.ndarray
(one for the output of the embeddings + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size)
.Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-
attentions (
tuple(jnp.ndarray)
, optional, returned whenoutput_attentions=True
is passed or whenconfig.output_attentions=True
) — Tuple ofjnp.ndarray
(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length)
.Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
The FlaxWhisperForAudioClassification forward method, overrides the __call__
special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Transcription example:
>>> import jax.numpy as jnp
>>> from transformers import AutoFeatureExtractor, FlaxWhisperForAudioClassification
>>> from datasets import load_dataset
>>> feature_extractor = AutoFeatureExtractor.from_pretrained("sanchit-gandhi/whisper-medium-fleurs-lang-id")
>>> model = FlaxWhisperForAudioClassification.from_pretrained(
... "sanchit-gandhi/whisper-medium-fleurs-lang-id", from_pt=True
... )
>>> ds = load_dataset("google/fleurs", "all", split="validation", streaming=True, trust_remote_code=True)
>>> sample = next(iter(ds))
>>> inputs = feature_extractor(
... sample["audio"]["array"], sampling_rate=sample["audio"]["sampling_rate"], return_tensors="np"
... )
>>> input_features = inputs.input_features
>>> logits = model(input_features).logits
>>> predicted_class_ids = jnp.argmax(logits).item()
>>> predicted_label = model.config.id2label[predicted_class_ids]
>>> predicted_label
'af_za'