Hugging Face's logo

Transformers documentation

Csm

You are viewing main version, which requires installation from source. If you'd like regular pip install, checkout the latest stable version (v4.51.3).
Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

Csm

Overview

The Conversational Speech Model (CSM) is the first open-source contextual text-to-speech model released by Sesame. It is designed to generate natural-sounding speech with or without conversational context. This context typically consists of multi-turn dialogue between speakers, represented as sequences of text and corresponding spoken audio.

Model Architecture: CSM is composed of two LLaMA-style auto-regressive transformer decoders: a backbone decoder that predicts the first codebook token and a depth decoder that generates the remaining tokens. It uses the pretrained codec model Mimi, introduced by Kyutai, to encode speech into discrete codebook tokens and decode them back into audio.

The original csm-1b checkpoint is available under the Sesame organization on Hugging Face.

Usage Tips

Without Conversational Context

CSM can be used to simply generate speech from a text prompt:

import torch
from transformers import CsmForConditionalGeneration, AutoProcessor

model_id = "eustlb/csm-1b"
device = "cuda" if torch.cuda.is_available() else "cpu"

# load the model and the processor
processor = AutoProcessor.from_pretrained(model_id)
model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=device)

# prepare the inputs
text = "[0]The past is just a story we tell ourselves." # `[0]` for speaker id 0
inputs = processor(text, add_special_tokens=True).to(device)

# another equivalent way to prepare the inputs
conversation = [
    {"role": "0", "content": [{"type": "text", "text": "The past is just a story we tell ourselves."}]},
]
inputs = processor.apply_chat_template(
    conversation,
    tokenize=True,
    return_dict=True,
).to(device)

# infer the model
audio = model.generate(**inputs, output_audio=True)
processor.save_audio(audio, "example_without_context.wav")

With Conversational Context

CSM can be used to generate speech given a conversation, allowing consistency in the voices and content-aware generation:

import torch
from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset, Audio

model_id = "eustlb/csm-1b"
device = "cuda" if torch.cuda.is_available() else "cpu"

# load the model and the processor
processor = AutoProcessor.from_pretrained(model_id)
model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=device)

# prepare the inputs
ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")
# ensure the audio is 24kHz
ds = ds.cast_column("audio", Audio(sampling_rate=24000))
conversation = []

# 1. context
for text, audio, speaker_id in zip(ds[:4]["text"], ds[:4]["audio"], ds[:4]["speaker_id"]):
    conversation.append(
        {
            "role": f"{speaker_id}",
            "content": [{"type": "text", "text": text}, {"type": "audio", "path": audio["array"]}],
        }
    )

# 2. text prompt
conversation.append({"role": f"{ds[4]['speaker_id']}", "content": [{"type": "text", "text": ds[4]["text"]}]})

inputs = processor.apply_chat_template(
    conversation,
    tokenize=True,
    return_dict=True,
).to(device)

# infer the model
audio = model.generate(**inputs, output_audio=True)
processor.save_audio(audio, "example_with_context.wav")

Batched Inference

CSM supports batched inference!

import torch
from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset, Audio

model_id = "eustlb/csm-1b"
device = "cuda" if torch.cuda.is_available() else "cpu"

# load the model and the processor
processor = AutoProcessor.from_pretrained(model_id)
model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=device)

# prepare the inputs 
ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")
# ensure the audio is 24kHz
ds = ds.cast_column("audio", Audio(sampling_rate=24000))
# here a batch with two prompts
conversation = [
    [
        {
            "role": f"{ds[0]['speaker_id']}",
            "content": [
                {"type": "text", "text": ds[0]["text"]},
                {"type": "audio", "path": ds[0]["audio"]["array"]},
            ],
        },
        {
            "role": f"{ds[1]['speaker_id']}",
            "content": [
                {"type": "text", "text": ds[1]["text"]},
            ],
        },
    ],
    [
        {
            "role": f"{ds[0]['speaker_id']}",
            "content": [
                {"type": "text", "text": ds[0]["text"]},
            ],
        }
    ],
]
inputs = processor.apply_chat_template(
    conversation,
    tokenize=True,
    return_dict=True,
).to(device)

audio = model.generate(**inputs, output_audio=True)
processor.save_audio(audio, [f"speech_batch_idx_{i}.wav" for i in range(len(audio))])

Making The Model Go Brrr

CSM supports full-graph compilation with CUDA graphs!

import torch
import copy
from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset

model_id = "eustlb/csm-1b"
device = "cuda"

# set logs to ensure no recompilation and graph breaks
torch._logging.set_logs(graph_breaks=True, recompiles=True, cudagraphs=True)

# load the model and the processor
processor = AutoProcessor.from_pretrained(model_id)
model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=device)

# use static cache, enabling automatically torch compile with fullgraph and reduce-overhead
model.generation_config.max_length = 250 # big enough to avoid recompilation
model.generation_config.max_new_tokens = None # would take precedence over max_length
model.generation_config.cache_implementation = "static"
model.depth_decoder.generation_config.cache_implementation = "static"

# generation kwargs
gen_kwargs = {
    "do_sample": False,
    "depth_decoder_do_sample": False,
    "temperature": 1.0,
    "depth_decoder_temperature": 1.0,
}

# Define a timing decorator
class TimerContext:
    def __init__(self, name="Execution"):
        self.name = name
        self.start_event = None
        self.end_event = None
        
    def __enter__(self):
        # Use CUDA events for more accurate GPU timing
        self.start_event = torch.cuda.Event(enable_timing=True)
        self.end_event = torch.cuda.Event(enable_timing=True)
        self.start_event.record()
        return self

    def __exit__(self, *args):
        self.end_event.record()
        torch.cuda.synchronize()
        elapsed_time = self.start_event.elapsed_time(self.end_event) / 1000.0
        print(f"{self.name} time: {elapsed_time:.4f} seconds")

# prepare the inputs 
ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")

conversation = [
    {
        "role": f"{ds[0]['speaker_id']}",
        "content": [
            {"type": "text", "text": ds[0]["text"]},
            {"type": "audio", "path": ds[0]["audio"]["array"]},
        ],
    },
    {
        "role": f"{ds[1]['speaker_id']}",
        "content": [
            {"type": "text", "text": ds[1]["text"]},
            {"type": "audio", "path": ds[1]["audio"]["array"]},
        ],
    },
    {
        "role": f"{ds[2]['speaker_id']}",
        "content": [
            {"type": "text", "text": ds[2]["text"]},
        ],
    },
]

padded_inputs_1 = processor.apply_chat_template(
    conversation,
    tokenize=True,
    return_dict=True,
).to(device)

print("\n" + "="*50)
print("First generation - compiling and recording CUDA graphs...")
with TimerContext("First generation"):
    _ = model.generate(**padded_inputs_1, **gen_kwargs)
print("="*50)

print("\n" + "="*50)
print("Second generation - fast !!!")
with TimerContext("Second generation"):
    _ = model.generate(**padded_inputs_1, **gen_kwargs)
print("="*50)

# now with different inputs
conversation = [
    {
        "role": f"{ds[0]['speaker_id']}",
        "content": [
            {"type": "text", "text": ds[2]["text"]},
            {"type": "audio", "path": ds[2]["audio"]["array"]},
        ],
    },
    {
        "role": f"{ds[1]['speaker_id']}",
        "content": [
            {"type": "text", "text": ds[3]["text"]},
            {"type": "audio", "path": ds[3]["audio"]["array"]},
        ],
    },
    {
        "role": f"{ds[2]['speaker_id']}",
        "content": [
            {"type": "text", "text": ds[4]["text"]},
        ],
    },
]
padded_inputs_2 = processor.apply_chat_template(
    conversation,
    tokenize=True,
    return_dict=True,
).to(device)

print("\n" + "="*50)
print("Generation with other inputs!")
with TimerContext("Generation with different inputs"):
    _ = model.generate(**padded_inputs_2, **gen_kwargs)
print("="*50)

Training

CSM Transformers integration supports training!

from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset, Audio

model_id = "eustlb/csm-1b"
device = "cuda"

# load the model and the processor
processor = AutoProcessor.from_pretrained(model_id)
model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=device)
model.train()

ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")
# ensure the audio is 24kHz
ds = ds.cast_column("audio", Audio(sampling_rate=24000))
conversation = []

# context
for text, audio, speaker_id in zip(ds[:4]["text"], ds[:4]["audio"], ds[:4]["speaker_id"]):
    conversation.append(
        {
            "role": f"{speaker_id}",
            "content": [{"type": "text", "text": text}, {"type": "audio", "path": audio["array"]}],
        }
    )

inputs = processor.apply_chat_template(
    conversation,
    tokenize=True,
    return_dict=True,
    output_labels=True,
).to(device)

out = model(**inputs)
out.loss.backward()

This model was contributed by Eustache Le Bihan. The original code can be found here.

CsmConfig

class transformers.CsmConfig

< >

( num_codebooks = 32 vocab_size = 2051 text_vocab_size = 128256 hidden_size = 2048 intermediate_size = 8192 num_hidden_layers = 16 num_attention_heads = 32 num_key_value_heads = 8 hidden_act = 'silu' max_position_embeddings = 2048 initializer_range = 0.02 rms_norm_eps = 1e-05 use_cache = True pad_token_id = 128002 codebook_pad_token_id = 2050 codebook_eos_token_id = 0 bos_token_id = 128000 eos_token_id = None audio_token_id = 128002 audio_eos_token_id = 128003 rope_theta = 500000 rope_scaling = None attention_bias = False attention_dropout = 0.0 mlp_bias = False head_dim = None tie_codebooks_embeddings = True depth_decoder_config = None codec_config = None **kwargs )

Parameters

  • num_codebooks (int, optional, defaults to 32) — Number of codebooks used in the underlying codec model responsible for tokenizing the audio.
  • vocab_size (int, optional, defaults to 2051) — Vocabulary size of the Csm model. Defines the number of different audio tokens that can be represented by each codebook.
  • text_vocab_size (int, optional, defaults to 128256) — Vocabulary size of the text input for the Csm model. Defines the number of different text tokens that can be represented.
  • hidden_size (int, optional, defaults to 2048) — Dimension of the hidden representations of the backbone model.
  • intermediate_size (int, optional, defaults to 8192) — Dimension of the MLP representations of the backbone model.
  • num_hidden_layers (int, optional, defaults to 16) — Number of hidden layers in the backbone model Transformer decoder.
  • num_attention_heads (int, optional, defaults to 32) — Number of attention heads for each attention layer in the backbone model Transformer decoder.
  • num_key_value_heads (int, optional, defaults to 8) — This is the number of key_value heads that should be used to implement Grouped Query Attention. If num_key_value_heads=num_attention_heads, the model will use Multi Head Attention (MHA), if num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details checkout this paper.
  • hidden_act (str or function, optional, defaults to "silu") — The non-linear activation function (function or string) in the backbone model Transformer decoder.
  • max_position_embeddings (int, optional, defaults to 2048) — The maximum sequence length that this model might ever be used with.
  • initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
  • rms_norm_eps (float, optional, defaults to 1e-05) — The epsilon used by the rms normalization layers.
  • use_cache (bool, optional, defaults to True) — Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if config.is_decoder=True.
  • pad_token_id (int, optional, defaults to 128002) — Padding token id.
  • codebook_pad_token_id (int, optional, defaults to 2050) — Padding token id for codebook tokens.
  • codebook_eos_token_id (int, optional, defaults to 0) — End of stream token id for codebook tokens.
  • bos_token_id (int, optional, defaults to 128000) — Beginning of stream token id.
  • eos_token_id (int, optional) — End of stream token id.
  • audio_token_id (int, optional, defaults to 128002) — Audio token id in the text input.
  • audio_eos_token_id (int, optional, defaults to 128003) — End of stream token id for audio in the text input.
  • rope_theta (float, optional, defaults to 500000) — The base period of the RoPE embeddings.
  • rope_scaling (Dict, optional, defaults to {'factor' -- 32.0, 'high_freq_factor': 0.5, 'low_freq_factor': 0.125, 'original_max_position_embeddings': 1024, 'rope_type': 'llama3'}): Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type and you expect the model to work on longer max_position_embeddings, we recommend you to update this value accordingly. Expected contents: rope_type (str): The sub-variant of RoPE to use. Can be one of [‘default’, ‘linear’, ‘dynamic’, ‘yarn’, ‘longrope’, ‘llama3’], with ‘default’ being the original RoPE implementation. factor (float, optional): Used with all rope types except ‘default’. The scaling factor to apply to the RoPE embeddings. In most scaling types, a factor of x will enable the model to handle sequences of length x original maximum pre-trained length. original_max_position_embeddings (int, optional): Used with ‘dynamic’, ‘longrope’ and ‘llama3’. The original max position embeddings used during pretraining. attention_factor (float, optional): Used with ‘yarn’ and ‘longrope’. The scaling factor to be applied on the attention computation. If unspecified, it defaults to value recommended by the implementation, using the factor field to infer the suggested value. beta_fast (float, optional): Only used with ‘yarn’. Parameter to set the boundary for extrapolation (only) in the linear ramp function. If unspecified, it defaults to 32. beta_slow (float, optional): Only used with ‘yarn’. Parameter to set the boundary for interpolation (only) in the linear ramp function. If unspecified, it defaults to 1. short_factor (List[float], optional): Only used with ‘longrope’. The scaling factor to be applied to short contexts (< original_max_position_embeddings). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 long_factor (List[float], optional): Only used with ‘longrope’. The scaling factor to be applied to long contexts (< original_max_position_embeddings). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 low_freq_factor (float, optional): Only used with ‘llama3’. Scaling factor applied to low frequency components of the RoPE high_freq_factor (float, optional*): Only used with ‘llama3’. Scaling factor applied to high frequency components of the RoPE
  • attention_bias (bool, optional, defaults to False) — Whether to use a bias in the query, key, value and output projection layers during self-attention.
  • attention_dropout (float, optional, defaults to 0.0) — The dropout ratio for the attention probabilities.
  • mlp_bias (bool, optional, defaults to False) — Whether to use a bias in up_proj, down_proj and gate_proj layers in the MLP layers.
  • head_dim (int, optional) — The attention head dimension. If None, it will default to hidden_size // num_attention_heads
  • tie_codebooks_embeddings (bool, optional, defaults to True) — Whether to tie the codebook tokens embeddings of the backbone model to the codebook tokens embeddings of the depth decoder.
  • depth_decoder_config (CsmDepthDecoderConfig, optional) — Configuration for the depth decoder.
  • codec_config (PretrainedConfig, optional) — Configuration for the codec.

This is the configuration class to store the configuration of a CsmForConditionalGeneration. It is used to instantiate an CSM 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 csm-1b.

e.g. eustlb/csm-1b

Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.

>>> from transformers import CsmForConditionalGeneration, CsmConfig

>>> # Initializing a CsmConfig
>>> configuration = CsmConfig()

>>> # Initializing a model
>>> model = CsmForConditionalGeneration(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config

CsmDepthDecoderConfig

class transformers.CsmDepthDecoderConfig

< >

( num_codebooks = 32 backbone_hidden_size = 2048 vocab_size = 2051 hidden_size = 1024 intermediate_size = 8192 num_hidden_layers = 4 num_attention_heads = 8 num_key_value_heads = 2 hidden_act = 'silu' max_position_embeddings = 33 initializer_range = 0.02 rms_norm_eps = 1e-05 use_cache = True pad_token_id = None bos_token_id = None eos_token_id = None rope_theta = 500000 rope_scaling = None attention_bias = False attention_dropout = 0.0 mlp_bias = False head_dim = None **kwargs )

Parameters

  • num_codebooks (int, optional, defaults to 32) — Number of codebooks used in the underlying codec model responsible for tokenizing the audio.
  • backbone_hidden_size (int, optional, defaults to 2048) — Dimension of the hidden representations of the backbone model used with this depth decoder.
  • vocab_size (int, optional, defaults to 2051) — Vocabulary size of the CsmDepthDecoder model. Defines the number of different audio tokens that can be represented by each codebook.
  • hidden_size (int, optional, defaults to 1024) — Dimension of the hidden representations.
  • intermediate_size (int, optional, defaults to 8192) — Dimension of the MLP representations.
  • num_hidden_layers (int, optional, defaults to 4) — Number of hidden layers in the Transformer decoder.
  • num_attention_heads (int, optional, defaults to 8) — Number of attention heads for each attention layer in the Transformer decoder.
  • num_key_value_heads (int, optional, defaults to 2) — This is the number of key_value heads that should be used to implement Grouped Query Attention. If num_key_value_heads=num_attention_heads, the model will use Multi Head Attention (MHA), if num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details checkout this paper. If it is not specified, will default to num_attention_heads.
  • hidden_act (str or function, optional, defaults to "silu") — The non-linear activation function (function or string) in the decoder.
  • max_position_embeddings (int, optional, defaults to 33) — The maximum sequence length that this model might ever be used with.
  • initializer_range (float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
  • rms_norm_eps (float, optional, defaults to 1e-05) — The epsilon used by the rms normalization layers.
  • use_cache (bool, optional, defaults to True) — Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if config.is_decoder=True.
  • pad_token_id (int, optional, defaults to 2050) — Padding token id.
  • bos_token_id (int, optional) — Beginning of stream token id.
  • eos_token_id (int, optional) — End of stream token id.
  • rope_theta (float, optional, defaults to 500000) — The base period of the RoPE embeddings.
  • rope_scaling (Dict, optional) — Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type and you expect the model to work on longer max_position_embeddings, we recommend you to update this value accordingly. Expected contents: rope_type (str): The sub-variant of RoPE to use. Can be one of [‘default’, ‘linear’, ‘dynamic’, ‘yarn’, ‘longrope’, ‘llama3’], with ‘default’ being the original RoPE implementation. factor (float, optional): Used with all rope types except ‘default’. The scaling factor to apply to the RoPE embeddings. In most scaling types, a factor of x will enable the model to handle sequences of length x original maximum pre-trained length. original_max_position_embeddings (int, optional): Used with ‘dynamic’, ‘longrope’ and ‘llama3’. The original max position embeddings used during pretraining. attention_factor (float, optional): Used with ‘yarn’ and ‘longrope’. The scaling factor to be applied on the attention computation. If unspecified, it defaults to value recommended by the implementation, using the factor field to infer the suggested value. beta_fast (float, optional): Only used with ‘yarn’. Parameter to set the boundary for extrapolation (only) in the linear ramp function. If unspecified, it defaults to 32. beta_slow (float, optional): Only used with ‘yarn’. Parameter to set the boundary for interpolation (only) in the linear ramp function. If unspecified, it defaults to 1. short_factor (List[float], optional): Only used with ‘longrope’. The scaling factor to be applied to short contexts (< original_max_position_embeddings). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 long_factor (List[float], optional): Only used with ‘longrope’. The scaling factor to be applied to long contexts (< original_max_position_embeddings). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 low_freq_factor (float, optional): Only used with ‘llama3’. Scaling factor applied to low frequency components of the RoPE high_freq_factor (float, optional*): Only used with ‘llama3’. Scaling factor applied to high frequency components of the RoPE
  • attention_bias (bool, optional, defaults to False) — Whether to use a bias in the query, key, value and output projection layers during self-attention.
  • attention_dropout (float, optional, defaults to 0.0) — The dropout ratio for the attention probabilities.
  • mlp_bias (bool, optional, defaults to False) — Whether to use a bias in up_proj, down_proj and gate_proj layers in the MLP layers.
  • head_dim (int, optional) — The attention head dimension. If None, it will default to hidden_size // num_attention_heads

This is the configuration class to store the configuration of a CsmDepthDecoderModel. It is used to instantiate an CSM depth decoder 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 csm-1b.

e.g. eustlb/csm-1b

Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.

>>> from transformers import CsmDepthDecoder, CsmDepthDecoderConfig

>>> # Initializing a CsmDepthDecoder
>>> configuration = CsmDepthDecoderConfig()
>>> model = CsmDepthDecoderModel(configuration)

>>> # Accessing the model configuration
>>> configuration = model.config

CsmProcessor

class transformers.CsmProcessor

< >

( feature_extractor tokenizer chat_template = None )

Parameters

  • feature_extractor (EncodecFeatureExtractor) — The feature extractor is a required input.
  • tokenizer ([PreTrainedTokenizer, PreTrainedTokenizerFast]) — The tokenizer is a required input.
  • chat_template (str, optional) — A Jinja template which will be used to convert lists of messages in a chat into a tokenizable string.

Constructs a Csm processor which wraps EncodecFeatureExtractor and PretrainedTokenizerFast into a single processor that inherits both the audio feature extraction and tokenizer functionalities. See the call() for more information. The preferred way of passing kwargs is as a dictionary per modality, see usage example below.

from transformers import CsmProcessor
from datasets import load_dataset

ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")
audio = ds[0]["audio"]["array"]

processor = CsmProcessor.from_pretrained("eustlb/csm-1b")

processor(
    text=["<|begin_of_text|>[0]What are you working on?<|end_of_text|><|AUDIO|><|audio_eos|><|begin_of_text|>[1]I'm figuring out my budget.<|end_of_text|>"],
    audio=audio,
    text_kwargs = {"padding": False},
    audio_kwargs = {"sampling_rate": 16000},
    common_kwargs = {"return_tensors": "pt"},
)
# this should error out because EncodecFeatureExtractor expects a 24kHz audio :)

__call__

< >

( text: typing.Union[str, typing.List[str], typing.List[typing.List[str]], NoneType] audio: typing.Union[numpy.ndarray, ForwardRef('torch.Tensor'), typing.List[numpy.ndarray], typing.Tuple[numpy.ndarray], typing.List[ForwardRef('torch.Tensor')], typing.Tuple[ForwardRef('torch.Tensor')], NoneType] = None output_labels: typing.Optional[bool] = False depth_decoder_labels_ratio: typing.Optional[float] = 1.0 **kwargs: typing_extensions.Unpack[transformers.models.csm.processing_csm.CsmProcessorKwargs] ) → BatchFeature

Parameters

  • audio (np.ndarray, torch.Tensor, List[np.ndarray], List[torch.Tensor]) — The audio or batch of audio to be prepared. Each audio can be a NumPy array or PyTorch tensor.
  • text (str, List[str], List[List[str]]) — The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set is_split_into_words=True (to lift the ambiguity with a batch of sequences).
  • output_labels (bool, optional, default=False) — Whether to return labels for training. Indices will be in [config.audio_token_id, -100, -101].
    • config.audio_token_id indicates an audio frame (considering sequence length elements as frames)
    • -100 will be ignored in the loss computation
    • -101 indicates the audio frame will be used only for the backbone model (using the first codebook token as labels)
  • depth_decoder_labels_ratio (float, optional, default=1.0) — The ratio of audio frames to keep for the depth decoder labels.
  • return_tensors (str or TensorType, optional) — If set, will return tensors of a particular framework. Acceptable values are:
    • 'tf': Return TensorFlow tf.constant objects.
    • 'pt': Return PyTorch torch.Tensor objects.
    • 'np': Return NumPy np.ndarray objects.
    • 'jax': Return JAX jnp.ndarray objects.

Returns

BatchFeature

A BatchFeature with the following fields:

  • input_ids — List of token ids to be fed to a model. Returned when text is not None.
  • input_values — List of audio values to be fed to a model. Returned when audio is not None.
  • attention_mask — List of indices specifying which tokens should be attended to by the model (when return_attention_mask=True or if “attention_mask” is in self.model_input_names and if text is not None).
  • labels — List of labels for the audio frames. Returned when output_labels=True.

Main method to prepare text(s) and audio to be fed as input to the model. This method forwards the text arguments to PreTrainedTokenizerFast’s call() to encode the text. To prepare the audio, this method forwards the audio arguments to EncodecFeatureExtractor’s call(). Please refer to the docstring of the above two methods for more information.

CsmForConditionalGeneration

class transformers.CsmForConditionalGeneration

< >

( config )

Parameters

  • config (CsmConfig) — 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 Csm model consists of two llama-like auto-regressive transformer models: a backbone model that predicts the first codebook token and a depth decoder that predicts the other codebook tokens.

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

< >

( input_ids: LongTensor = None input_values: typing.Optional[torch.Tensor] = None attention_mask: typing.Optional[torch.Tensor] = None input_values_cutoffs: typing.Optional[torch.Tensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Union[transformers.cache_utils.Cache, typing.List[torch.FloatTensor], NoneType] = None inputs_embeds: typing.Optional[torch.FloatTensor] = None labels: typing.Optional[torch.LongTensor] = None use_cache: typing.Optional[bool] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None cache_position: typing.Optional[torch.LongTensor] = None logits_to_keep: typing.Union[int, torch.Tensor] = 0 **kwargs: typing_extensions.Unpack[transformers.models.csm.modeling_csm.KwargsForCausalLM] ) → transformers.models.csm.modeling_csm.CsmOutputWithPast or tuple(torch.FloatTensor)

Parameters

  • input_ids (torch.LongTensor of shape (batch_size, sequence_length, num_codebooks) or (batch_size, sequence_length)) —

    1. (batch_size, sequence_length): corresponds to the input sequence prepared with the processor from the text prompt. Such input requires input_values to be provided so that audio can be encoded in codebook tokens and then merged with the text tokens.

    2. (batch_size, sequence_length, num_codebooks): codebook tokens generated during the autoregressive decoding. Such input is not meant to be used by end users.

    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?

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    If past_key_values is used, optionally only the last input_ids have to be input (see past_key_values).

    If you want to change padding behavior, you should read modeling_opt._prepare_decoder_attention_mask and modify to your needs. See diagram 1 in the paper for more information on the default strategy.

    • 1 indicates the head is not masked,
    • 0 indicates the head is masked.
  • position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.n_positions - 1].

    What are position IDs?

  • past_key_values (Cache or tuple(tuple(torch.FloatTensor)), optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the past_key_values returned by the model at a previous stage of decoding, when use_cache=True or config.use_cache=True.

    Two formats are allowed:

    • a Cache instance, see our kv cache guide;
    • Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head)). This is also known as the legacy cache format.

    The model will output the same cache format that is fed as input. If no past_key_values are passed, the legacy cache format will be returned.

    If past_key_values are used, the user can optionally input only the last input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all input_ids of shape (batch_size, sequence_length).

  • inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
  • use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
  • output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
  • output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_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. Contrarily to position_ids, this tensor is not affected by padding. It is used to update the cache in the correct position and to infer the complete sequence length.
  • input_values_cutoffs (torch.Tensor of shape (batch_size, max_num_audio), optional) — Specify the end positions of audio segments within each batch entry, relative to the concatenated audio input. If a batch entry has fewer segments than the maximum, it is padded with -1. For example, in a batch of 2 sequences where the first contains 2 audio segments of length l1, and the second contains 1 audio segment of length l2, the input_values_cutoffs would be: [[l1, 2 * l1], [l2, -1]].
  • labels (torch.LongTensor of shape (batch_size, sequence_length), optional) — Labels for computing the masked language modeling loss. Indices should be in [config.audio_token_id, -100, -101]. Requires targeted input_values to be provided as audio tokens will be infered from it using the codec_model.

    • config.audio_token_id indicates an audio frames (considering sequence length elements as frames)
    • -100 will be ignored in the loss computation
    • -101 indicates the audio frame will be used only for the backbone model (using the first codebook token as labels)

    Such labels can be prepared using output_labels=True when calling CsmProcessor.

  • logits_to_keep (int or torch.Tensor, optional) — Kept for compatibility. Does not support another value than:

    1. 0, which is equivalent to keeping all logits, used in the training regime
    2. 1, which is equivalent to keeping only the last logit, used in the generation regime

Returns

transformers.models.csm.modeling_csm.CsmOutputWithPast or tuple(torch.FloatTensor)

A transformers.models.csm.modeling_csm.CsmOutputWithPast 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 (CsmConfig) and inputs.

  • loss (torch.FloatTensor of shape (1,), optional, returned when labels 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).

  • past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head))

    Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.

  • hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.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.

  • depth_decoder_loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss (for next-token prediction) of the depth decoder model.

  • depth_decoder_logits (torch.FloatTensor of shape (batch_size, sequence_length, config.vocab_size)) — Prediction scores of the depth decoder (scores for each vocabulary token before SoftMax).

  • depth_decoder_past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head))

  • depth_decoder_hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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.

  • depth_decoder_attentions (tuple(torch.FloatTensor), optional, returned when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.FloatTensor (one for each layer) of shape (batch_size, num_heads, sequence_length, sequence_length).

  • backbone_loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Language modeling loss (for next-token prediction) of the backbone model.

The CsmForConditionalGeneration 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 CsmForConditionalGeneration, AutoProcessor
>>> from datasets import load_dataset, Audio

>>> model_id = "eustlb/csm-1b"
>>> torch_device = "cuda" if torch.cuda.is_available() else "cpu"

>>> processor = AutoProcessor.from_pretrained(model_id)

>>> ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")
>>> # ensure the audio is 24kHz
>>> ds = ds.cast_column("audio", Audio(sampling_rate=24000))

>>> conversation = []
>>> # prepare a conversation with text and corresponding audio
>>> for text, audio, speaker_id in zip(ds[:4]["text"], ds[:4]["audio"], ds[:4]["speaker_id"]):
...     conversation.append(
...         {
...             "role": f"{speaker_id}",
...             "content": [{"type": "text", "text": text}, {"type": "audio", "path": audio["array"]}],
...         }
...     )

>>> inputs = processor.apply_chat_template(
...     conversation,
...     tokenize=True,
...     return_dict=True,
...     output_labels=True,
... ).to(torch_device)

>>> model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=torch_device)
>>> output = model(**inputs)
>>> output.loss.backward()

generate

< >

( input_ids: typing.Optional[torch.Tensor] = None input_values: typing.Optional[torch.Tensor] = None input_values_cutoffs: typing.Optional[torch.Tensor] = None generation_config: typing.Optional[transformers.generation.configuration_utils.GenerationConfig] = None logits_processor: typing.Optional[transformers.generation.logits_process.LogitsProcessorList] = None stopping_criteria: typing.Optional[transformers.generation.stopping_criteria.StoppingCriteriaList] = None synced_gpus: typing.Optional[bool] = None streamer: typing.Optional[ForwardRef('BaseStreamer')] = None output_audio: typing.Optional[bool] = False **kwargs ) → CsmGenerateOutput or torch.LongTensor or List[torch.FloatTensor]

Parameters

  • inputs_ids (torch.Tensor of shape (batch_size, seq_length), optional) — The sequence used as a prompt for the backbone model.
  • input_values (torch.Tensor of shape (batch_size, channels, max_concatenated_audio_length), optional) — The batched audio input values, where each batch entry contains the concatenation of all audio segments for that entry. These values will be encoded into codebook tokens using the codec model and merged with the text input ids provided in input_ids.
  • input_values_cutoffs (torch.Tensor of shape (batch_size, max_num_audio), optional) — Specify the end positions of audio segments within each batch entry, relative to the concatenated audio input. If a batch entry has fewer segments than the maximum, it is padded with -1. For example, in a batch of 2 sequences where the first contains 2 audio segments of length l1, and the second contains 1 audio segment of length l2, the input_values_cutoffs would be: [[l1, 2 * l1], [l2, -1]].
  • generation_config (GenerationConfig, optional) — The generation configuration to be used as base parametrization for the generation call. **kwargs passed to generate matching the attributes of generation_config will override them. If generation_config is not provided, the default will be used, which has the following loading priority: 1) from the generation_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 complements 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. If your stopping criteria depends on the scores input, make sure you pass return_dict_in_generate=True, output_scores=True to generate. This feature is intended for advanced users.
  • synced_gpus (bool, optional) — Whether to continue running the while loop until max_length. Unless overridden, this flag will be set to True if using FullyShardedDataParallel or DeepSpeed ZeRO Stage 3 with multiple GPUs to avoid deadlocking if one GPU finishes generating before other GPUs. Otherwise, defaults to False.
  • streamer (BaseStreamer, optional) — Streamer object that will be used to stream the generated sequences. Generated tokens are passed through streamer.put(token_ids) and the streamer is responsible for any further processing.
  • output_audio (bool, optional) — Whether to return the generated audio.
  • kwargs (Dict[str, Any], optional) — Ad hoc parametrization of generation_config and/or additional model-specific kwargs that will be forwarded to the forward function of the model. Depth decoder specific kwargs should be prefixed with depthdecoder.

Returns

CsmGenerateOutput or torch.LongTensor or List[torch.FloatTensor]

A CsmGenerateOutput (if return_dict_in_generate=True or when config.return_dict_in_generate=True) or a torch.LongTensor when output_audio=False or a List[torch.FloatTensor] otherwise.

This method overrides generate() to match the specifics of the Csm model. Indeed, Csm model requires a custom generation sampling step:

  1. Infer the backbone model to sample the first codebook token
  2. Call generate on the depth decoder with the first codebook token as input_ids to sample the next codebook tokens
  3. Use these generated codebook tokens as input_ids to sample the next first codebook token using the backbone model
  4. Repeat until stopping criteria is met

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, do_sample=True).

Example:

>>> from transformers import CsmProcessor, CsmForConditionalGeneration
>>> from datasets import load_dataset, Audio

>>> model_id = "eustlb/csm-1b"
>>> torch_device = "cuda" if torch.cuda.is_available() else "cpu"

>>> processor = AutoProcessor.from_pretrained(model_id)

>>> ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")
>>> # ensure the audio is 24kHz
>>> ds = ds.cast_column("audio", Audio(sampling_rate=24000))

>>> conversation = []
>>> # prepare a conversation with text and corresponding audio
>>> for text, audio, speaker_id in zip(ds[:4]["text"], ds[:4]["audio"], ds[:4]["speaker_id"]):
...     conversation.append(
...         {
...             "role": f"{speaker_id}",
...             "content": [{"type": "text", "text": text}, {"type": "audio", "path": audio["array"]}],
...         }
...     )

>>> # text prompt
>>> conversation.append({"role": f"{ds[4]['speaker_id']}", "content": [{"type": "text", "text": ds[4]["text"]}]})

>>> inputs = processor.apply_chat_template(
...     conversation,
...     tokenize=True,
...     return_dict=True,
... ).to(torch_device)

>>> model = CsmForConditionalGeneration.from_pretrained(model_id, device_map=torch_device)
>>> audio = model.generate(**inputs, output_audio=True)
>>> processor.save_audio(audio, "output.wav")

CsmDepthDecoderForCausalLM

class transformers.CsmDepthDecoderForCausalLM

< >

( config )

Parameters

  • config (CsmDepthDecoderConfig) — 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 CsmDepthDecoder Model transformer, with a CsmCodebooksHead on top, which can be seen a position-specific language modeling head, allowing to use a different linear layer for each codebook (e.g. position 0 is the first codebook and uses the first codebook head, etc.)

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

< >

( input_ids: LongTensor = None backbone_last_hidden_state: typing.Optional[torch.FloatTensor] = None attention_mask: typing.Optional[torch.Tensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Union[transformers.cache_utils.Cache, typing.List[torch.FloatTensor], NoneType] = None inputs_embeds: typing.Optional[torch.FloatTensor] = None labels: typing.Optional[torch.LongTensor] = None use_cache: typing.Optional[bool] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None cache_position: typing.Optional[torch.LongTensor] = None logits_to_keep: typing.Union[int, torch.Tensor] = 0 **kwargs: typing_extensions.Unpack[transformers.models.csm.modeling_csm.KwargsForCausalLM] ) → transformers.modeling_outputs.CausalLMOutputWithPast 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?

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    If past_key_values is used, optionally only the last input_ids have to be input (see past_key_values).

    If you want to change padding behavior, you should read modeling_opt._prepare_decoder_attention_mask and modify to your needs. See diagram 1 in the paper for more information on the default strategy.

    • 1 indicates the head is not masked,
    • 0 indicates the head is masked.
  • position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.n_positions - 1].

    What are position IDs?

  • past_key_values (Cache or tuple(tuple(torch.FloatTensor)), optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the past_key_values returned by the model at a previous stage of decoding, when use_cache=True or config.use_cache=True.

    Two formats are allowed:

    • a Cache instance, see our kv cache guide;
    • Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head)). This is also known as the legacy cache format.

    The model will output the same cache format that is fed as input. If no past_key_values are passed, the legacy cache format will be returned.

    If past_key_values are used, the user can optionally input only the last input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all input_ids of shape (batch_size, sequence_length).

  • inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
  • use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
  • output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
  • output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_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. Contrarily to position_ids, this tensor is not affected by padding. 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 masked language modeling loss. Indices should either be in [0, ..., config.vocab_size] or -100 (see input_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].
  • logits_to_keep (int or torch.Tensor, optional) — If an int, compute logits for the last logits_to_keep tokens. If 0, calculate logits for all input_ids (special case). Only last token logits are needed for generation, and calculating them only for that token can save memory, which becomes pretty significant for long sequences or large vocabulary size. If a torch.Tensor, must be 1D corresponding to the indices to keep in the sequence length dimension. This is useful when using packed tensor format (single dimension for batch and sequence length).

Returns

transformers.modeling_outputs.CausalLMOutputWithPast or tuple(torch.FloatTensor)

A transformers.modeling_outputs.CausalLMOutputWithPast 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 (CsmConfig) and inputs.

  • loss (torch.FloatTensor of shape (1,), optional, returned when labels 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).

  • past_key_values (tuple(tuple(torch.FloatTensor)), optional, returned when use_cache=True is passed or when config.use_cache=True) — Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head))

    Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see past_key_values input) to speed up sequential decoding.

  • hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.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 CsmDepthDecoderForCausalLM 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, CsmDepthDecoderForCausalLM

>>> model = CsmDepthDecoderForCausalLM.from_pretrained("meta-csm_depth_decoder/CsmDepthDecoder-2-7b-hf")
>>> tokenizer = AutoTokenizer.from_pretrained("meta-csm_depth_decoder/CsmDepthDecoder-2-7b-hf")

>>> prompt = "Hey, are you conscious? Can you talk to me?"
>>> inputs = tokenizer(prompt, return_tensors="pt")

>>> # Generate
>>> generate_ids = model.generate(inputs.input_ids, max_length=30)
>>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
"Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
backbone_last_hidden_state (`torch.FloatTensor` of shape `(batch_size, backbone_hidden_size)`, *optional*): The last hidden state of the backbone model. Such input is required when the first codebook token (the one generated by the backbone model) is provided in the `input_ids` argument. 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 (see `input_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]`.

CsmDepthDecoderModel

class transformers.CsmDepthDecoderModel

< >

( config )

Parameters

  • config (CsmDepthDecoderConfig) — 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.
  • config — CsmDepthDecoderConfig

The bare CsmDepthDecoderModel 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.

Transformer decoder consisting of config.num_hidden_layers layers. Each layer is a CsmDecoderLayer

forward

< >

( input_ids: LongTensor = None backbone_last_hidden_state: typing.Optional[torch.FloatTensor] = None attention_mask: typing.Optional[torch.Tensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Optional[transformers.cache_utils.Cache] = None inputs_embeds: typing.Optional[torch.FloatTensor] = None use_cache: typing.Optional[bool] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None cache_position: typing.Optional[torch.LongTensor] = None **flash_attn_kwargs: typing_extensions.Unpack[transformers.modeling_flash_attention_utils.FlashAttentionKwargs] )

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?

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    If past_key_values is used, optionally only the last input_ids have to be input (see past_key_values).

    If you want to change padding behavior, you should read modeling_opt._prepare_decoder_attention_mask and modify to your needs. See diagram 1 in the paper for more information on the default strategy.

    • 1 indicates the head is not masked,
    • 0 indicates the head is masked.
  • position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.n_positions - 1].

    What are position IDs?

  • past_key_values (Cache or tuple(tuple(torch.FloatTensor)), optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the past_key_values returned by the model at a previous stage of decoding, when use_cache=True or config.use_cache=True.

    Two formats are allowed:

    • a Cache instance, see our kv cache guide;
    • Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head)). This is also known as the legacy cache format.

    The model will output the same cache format that is fed as input. If no past_key_values are passed, the legacy cache format will be returned.

    If past_key_values are used, the user can optionally input only the last input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all input_ids of shape (batch_size, sequence_length).

  • inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
  • use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
  • output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
  • output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_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. Contrarily to position_ids, this tensor is not affected by padding. It is used to update the cache in the correct position and to infer the complete sequence length.
  • backbone_last_hidden_state (torch.FloatTensor of shape (batch_size, backbone_hidden_size), optional) — The last hidden state of the backbone model. Such input is required when the first codebook token (the one generated by the backbone model) is provided in the input_ids argument.

The CsmDepthDecoderModel 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.

CsmBackboneModel

class transformers.CsmBackboneModel

< >

( config )

Parameters

  • config (CsmConfig) — 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.
  • config — CsmBackboneModelConfig

The bare CsmBackboneModel 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.

Transformer decoder consisting of config.num_hidden_layers layers. Each layer is a CsmDecoderLayer

forward

< >

( input_ids: typing.Optional[torch.LongTensor] = None attention_mask: typing.Optional[torch.Tensor] = None position_ids: typing.Optional[torch.LongTensor] = None past_key_values: typing.Optional[transformers.cache_utils.Cache] = None inputs_embeds: typing.Optional[torch.FloatTensor] = None use_cache: typing.Optional[bool] = None output_attentions: typing.Optional[bool] = None output_hidden_states: typing.Optional[bool] = None cache_position: typing.Optional[torch.LongTensor] = None **flash_attn_kwargs: typing_extensions.Unpack[transformers.modeling_flash_attention_utils.FlashAttentionKwargs] )

Parameters

  • input_ids (torch.LongTensor of shape (batch_size, sequence_length, num_codebooks) or (batch_size, sequence_length)) —

    1. (batch_size, sequence_length): corresponds to the input sequence prepared with the processor from the text prompt. Such input requires input_values to be provided so that audio can be encoded in codebook tokens and then merged with the text tokens.

    2. (batch_size, sequence_length, num_codebooks): codebook tokens generated during the autoregressive decoding. Such input is not meant to be used by end users.

    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?

    Indices can be obtained using AutoTokenizer. See PreTrainedTokenizer.encode() and PreTrainedTokenizer.call() for details.

    If past_key_values is used, optionally only the last input_ids have to be input (see past_key_values).

    If you want to change padding behavior, you should read modeling_opt._prepare_decoder_attention_mask and modify to your needs. See diagram 1 in the paper for more information on the default strategy.

    • 1 indicates the head is not masked,
    • 0 indicates the head is masked.
  • position_ids (torch.LongTensor of shape (batch_size, sequence_length), optional) — Indices of positions of each input sequence tokens in the position embeddings. Selected in the range [0, config.n_positions - 1].

    What are position IDs?

  • past_key_values (Cache or tuple(tuple(torch.FloatTensor)), optional) — Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention blocks) that can be used to speed up sequential decoding. This typically consists in the past_key_values returned by the model at a previous stage of decoding, when use_cache=True or config.use_cache=True.

    Two formats are allowed:

    • a Cache instance, see our kv cache guide;
    • Tuple of tuple(torch.FloatTensor) of length config.n_layers, with each tuple having 2 tensors of shape (batch_size, num_heads, sequence_length, embed_size_per_head)). This is also known as the legacy cache format.

    The model will output the same cache format that is fed as input. If no past_key_values are passed, the legacy cache format will be returned.

    If past_key_values are used, the user can optionally input only the last input_ids (those that don’t have their past key value states given to this model) of shape (batch_size, 1) instead of all input_ids of shape (batch_size, sequence_length).

  • inputs_embeds (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size), optional) — Optionally, instead of passing input_ids you can choose to directly pass an embedded representation. This is useful if you want more control over how to convert input_ids indices into associated vectors than the model’s internal embedding lookup matrix.
  • use_cache (bool, optional) — If set to True, past_key_values key value states are returned and can be used to speed up decoding (see past_key_values).
  • output_attentions (bool, optional) — Whether or not to return the attentions tensors of all attention layers. See attentions under returned tensors for more detail.
  • output_hidden_states (bool, optional) — Whether or not to return the hidden states of all layers. See hidden_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. Contrarily to position_ids, this tensor is not affected by padding. It is used to update the cache in the correct position and to infer the complete sequence length.

The CsmBackboneModel 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.

< > Update on GitHub

←CLAP dac→