ultravox.py

# Adapted from https://github.com/fixie-ai/ultravox/blob/ecd58c4041030bae2ad15aa6bcf04ab43199ea02/ultravox/model/ultravox_model.py
"""PyTorch Ultravox model."""

import math
from functools import cached_property, lru_cache
from typing import (Any, Dict, Iterable, List, Literal, Mapping, Optional, Set,
                    Tuple, TypedDict, Union)

import numpy as np
import torch
import torch.utils.checkpoint
from torch import nn
from torch.nn import functional as F
from transformers import BatchFeature
from transformers.models.whisper import WhisperFeatureExtractor
from transformers.models.whisper.modeling_whisper import WhisperEncoder

from vllm.attention import AttentionMetadata
from vllm.config import VllmConfig
from vllm.inputs import InputContext
from vllm.model_executor.layers.activation import SiluAndMul, get_act_fn
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
from vllm.model_executor.model_loader.loader import DefaultModelLoader
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.multimodal import MULTIMODAL_REGISTRY, NestedTensors
from vllm.multimodal.processing import (BaseMultiModalProcessor,
                                        MultiModalDataDict,
                                        MultiModalDataItems, ProcessorInputs,
                                        PromptReplacement)
from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.configs.ultravox import UltravoxConfig

from .interfaces import SupportsMultiModal, SupportsPP
from .utils import (AutoWeightsLoader, WeightsMapper, flatten_bn,
                    init_vllm_registered_model, maybe_prefix,
                    merge_multimodal_embeddings_from_map)

_AUDIO_TOKENS_PER_SECOND = 6.25


class UltravoxAudioFeatureInputs(TypedDict):
    type: Literal["audio_features"]
    data: NestedTensors
    """Shape: `(batch_size, num_audios, 80, M)`"""


class UltravoxAudioEmbeddingInputs(TypedDict):
    type: Literal["audio_embeds"]
    data: NestedTensors
    """Shape: `(batch_size, num_audios, audio_feature_size, hidden_size)`"""


UltravoxAudioInputs = Union[UltravoxAudioFeatureInputs,
                            UltravoxAudioEmbeddingInputs]


@lru_cache
def cached_feature_extractor(model_id: str) -> WhisperFeatureExtractor:
    return WhisperFeatureExtractor.from_pretrained(model_id)


def whisper_feature_extractor(ctx: InputContext) -> WhisperFeatureExtractor:
    return cached_feature_extractor(
        ctx.get_hf_config(UltravoxConfig).audio_model_id)


def get_ultravox_max_audio_tokens(ctx: InputContext):
    feature_extractor = whisper_feature_extractor(ctx)
    return math.ceil(feature_extractor.chunk_length * _AUDIO_TOKENS_PER_SECOND)


class UltravoxMultiModalProcessor(BaseMultiModalProcessor):

    def _get_feature_extractor(self) -> WhisperFeatureExtractor:
        return self._get_hf_processor().audio_processor.feature_extractor

    def _resample_audio(
        self,
        audio: np.ndarray,
        sr: int,
    ) -> Dict[str, Union[np.ndarray, int]]:
        # resample audio to the model's sampling rate
        feature_extractor = self._get_feature_extractor()
        if sr != feature_extractor.sampling_rate:
            try:
                import librosa
            except ImportError as exc:
                raise ImportError(
                    "Please install vllm[audio] for audio support.") from exc
            audio = librosa.resample(audio,
                                     orig_sr=sr,
                                     target_sr=feature_extractor.sampling_rate)
            sr = feature_extractor.sampling_rate
        return {"audio": audio, "sampling_rate": sr}

    def _apply_hf_processor(
        self,
        prompt: str,
        mm_data: MultiModalDataDict,
        mm_processor_kwargs: Mapping[str, object],
    ) -> BatchFeature:
        if not mm_data or not mm_data.get("audio", None):
            return super()._apply_hf_processor(prompt, mm_data,
                                               mm_processor_kwargs)

        audio_data = mm_data["audio"]
        if not isinstance(audio_data, list):
            audio_data = [audio_data]

        # Ultravox processor doesn't support multiple inputs,
        # therefore we need to input text and audio one by one
        tokenizer = self._get_tokenizer()
        audio_features, audio_token_len = [], []
        processed_inputs = {}
        for audio, sr in audio_data:
            data = self._resample_audio(audio, sr)
            processed_inputs = super()._apply_hf_processor(
                prompt, data, mm_processor_kwargs)
            prompt = tokenizer.decode(processed_inputs["input_ids"][0],
                                      skip_special_tokens=False)
            audio_features.append(
                processed_inputs.pop("audio_values").squeeze(0))
            audio_token_len.append(
                processed_inputs.pop("audio_token_len").item())

        return dict(
            **processed_inputs,
            audio_features=audio_features,
            audio_token_len=audio_token_len,
        )

    def _get_processor_data(
        self,
        mm_data: MultiModalDataDict,
    ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
        # Ultravox uses "audio" instead of "audios" as calling keyword
        processor_data, passthrough_data = super()._get_processor_data(mm_data)
        if "audios" in processor_data:
            processor_data["audio"] = processor_data.pop("audios")
        return processor_data, passthrough_data

    def _get_prompt_replacements(
        self,
        mm_items: MultiModalDataItems,
        hf_inputs: BatchFeature,
        mm_processor_kwargs: Mapping[str, object],
    ) -> list[PromptReplacement]:
        hf_processor = self._get_hf_processor()
        placeholder = hf_processor.audio_token_replacement

        def get_replacement_ultravox(item_idx: int):
            audio_token_len = hf_inputs["audio_token_len"][item_idx]
            return placeholder * audio_token_len

        return [
            PromptReplacement(
                modality="audio",
                target="<|audio|>",
                replacement=get_replacement_ultravox,
            )
        ]

    def _get_dummy_mm_inputs(
        self,
        mm_counts: Mapping[str, int],
    ) -> ProcessorInputs:
        feature_extractor = self._get_feature_extractor()
        sampling_rate = feature_extractor.sampling_rate
        audio_len = feature_extractor.chunk_length * sampling_rate

        audio_count = mm_counts["audio"]
        audio = np.zeros(audio_len)
        data = {"audio": [(audio, sampling_rate)] * audio_count}

        return ProcessorInputs(
            prompt_text="<|audio|>" * audio_count,
            mm_data=data,
            mm_processor_kwargs={},
        )


class StackAudioFrames(nn.Module):
    """
    Stack the audio embedding frames to reduce the sequence length by a factor
    of `stack_factor`.
    """

    def __init__(self, stack_factor: int = 8):
        super().__init__()
        self.stack_factor = stack_factor

    def forward(self, audio_embeds: torch.Tensor) -> torch.Tensor:
        B, T, C = audio_embeds.shape
        T_pad = (T + self.stack_factor -
                 1) // self.stack_factor * self.stack_factor
        audio_embeds = F.pad(audio_embeds, (0, 0, 0, T_pad - T))
        B, T, C = audio_embeds.shape
        audio_embeds = audio_embeds.view(B, T // self.stack_factor,
                                         C * self.stack_factor)
        return audio_embeds


class FlippedSiluAndMul(SiluAndMul):
    """Ultravox is trained with SwiGLU with flipped halves."""

    def forward(self, x: torch.Tensor):
        a, b = x.chunk(2, dim=-1)
        flipped = torch.cat((b, a), dim=-1)
        return super().forward(flipped)


class UltravoxProjector(nn.Module):

    def __init__(self, config: UltravoxConfig):
        super().__init__()
        self.hidden_dim = config.hidden_size
        self._pad_and_stack = StackAudioFrames(config.stack_factor)
        dim = config.audio_config.hidden_size * config.stack_factor
        self.ln_pre = RMSNorm(dim)
        self.linear_1 = nn.Linear(dim, self.hidden_dim, bias=False)
        dim = self.hidden_dim

        if config.projector_act == "swiglu":
            self.act = FlippedSiluAndMul()
            dim = dim // 2
        else:
            self.act = get_act_fn(config.projector_act)

        self.linear_2 = nn.Linear(dim,
                                  config.text_config.hidden_size,
                                  bias=False)
        self.ln_post = RMSNorm(config.text_config.hidden_size)

    def forward(self, audio_features: torch.Tensor) -> torch.Tensor:
        audio_features = self._pad_and_stack(audio_features)
        audio_features = self.ln_pre(audio_features)
        hidden_states = self.linear_1(audio_features)
        hidden_states = self.act(hidden_states)
        hidden_states = self.linear_2(hidden_states)
        hidden_states = self.ln_post(hidden_states)
        return hidden_states


class ModifiedWhisperEncoder(WhisperEncoder):
    """
    Encoder portion of OpenAI's Whisper model.

    This implementation is a slightly modified version of HF Transformers'
    Whisper Encoder, with only a few fixes:
    1. base_model_prefix updated to allow for doing `.from_pretrained`
       directly on the encoder
    2. allow less than 30 second of audio padding to be passed in:
        - relaxed ValueError check for `input_features` length to be less
           than or equal to `expected_seq_length` instead of strictly equal
        - embed_pos is now sliced to match the length of `inputs_embeds`

    Original: https://github.com/huggingface/transformers/blob/main/src/transformers/models/whisper/modeling_whisper.py
    See commentary: https://github.com/huggingface/transformers/issues/25744
    """

    base_model_prefix = "model.encoder"

    def forward(
        self,
        input_features,
    ):
        expected_seq_length = (self.config.max_source_positions *
                               self.conv1.stride[0] * self.conv2.stride[0])
        if input_features.shape[-1] > expected_seq_length:
            raise ValueError(
                f"Whisper expects the mel input features to be of length "
                f"{expected_seq_length} or less, but found "
                f"{input_features.shape[-1]}. Make sure to pad the input mel "
                f"features to {expected_seq_length}.")

        inputs_embeds = nn.functional.gelu(self.conv1(input_features))
        inputs_embeds = nn.functional.gelu(self.conv2(inputs_embeds))

        inputs_embeds = inputs_embeds.permute(0, 2, 1)
        embed_pos = self.embed_positions.weight[:inputs_embeds.size(-2)]

        hidden_states = inputs_embeds + embed_pos
        hidden_states = nn.functional.dropout(hidden_states,
                                              p=self.dropout,
                                              training=self.training)

        for encoder_layer in self.layers:
            layer_outputs = encoder_layer(
                hidden_states,
                None,
                layer_head_mask=None,
            )

            hidden_states = layer_outputs[0]

        hidden_states = self.layer_norm(hidden_states)
        return hidden_states


@MULTIMODAL_REGISTRY.register_max_multimodal_tokens(
    "audio", get_ultravox_max_audio_tokens)
@MULTIMODAL_REGISTRY.register_processor(UltravoxMultiModalProcessor)
class UltravoxModel(nn.Module, SupportsMultiModal, SupportsPP):

    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
        super().__init__()
        config = vllm_config.model_config.hf_config
        multimodal_config = vllm_config.model_config.multimodal_config
        self.config = config
        self.multi_modal_config = multimodal_config
        assert self.multi_modal_config

        self.secondary_weights = []
        self.audio_tower = ModifiedWhisperEncoder(config.audio_config)
        if config.audio_model_id is not None:
            # this prefix is not for initialization, but for loading weights
            # note the trailing dot
            self.secondary_weights.append(
                DefaultModelLoader.Source(
                    model_or_path=config.audio_model_id,
                    revision=None,
                    prefix="audio_tower.",
                ))
        self.multi_modal_projector = UltravoxProjector(config)
        self.language_model = init_vllm_registered_model(
            vllm_config=vllm_config,
            hf_config=config.text_config,
            prefix=maybe_prefix(prefix, "language_model"),
        )
        if config.text_model_id is not None:
            # this prefix is not for initialization, but for loading weights
            # note the trailing dot
            self.secondary_weights.append(
                DefaultModelLoader.Source(model_or_path=config.text_model_id,
                                          revision=None,
                                          prefix="language_model."))

        self.make_empty_intermediate_tensors = (
            self.language_model.make_empty_intermediate_tensors)

    @cached_property
    def sampler(self):
        if hasattr(self.language_model, "sampler"):
            return self.language_model.sampler

        return get_sampler()

    def _audio_features_to_embeddings(
            self, input_features: torch.Tensor) -> torch.Tensor:
        audio_input = input_features.to(self.audio_tower.dtype)
        audio_features = self.audio_tower(audio_input)
        audio_features = audio_features.to(self.audio_tower.dtype)
        audio_embeddings = self.multi_modal_projector(audio_features)
        return audio_embeddings

    def _parse_and_validate_audio_input(
            self, **kwargs: object) -> Optional[UltravoxAudioInputs]:
        audio_features = kwargs.pop("audio_features", None)
        audio_embeds = kwargs.pop("audio_embeds", None)

        if audio_features is None and audio_embeds is None:
            return None

        if audio_features is not None:
            if not isinstance(audio_features, (torch.Tensor, list)):
                raise ValueError("Incorrect type of audio features. "
                                 f"Got type: {type(audio_features)}")

            return UltravoxAudioFeatureInputs(type="audio_features",
                                              data=audio_features)

        if audio_embeds is not None:
            if not isinstance(audio_embeds, (torch.Tensor, list)):
                raise ValueError("Incorrect type of audio embeds. "
                                 f"Got type: {type(audio_embeds)}")

            return UltravoxAudioEmbeddingInputs(type="audio_embeds",
                                                data=audio_embeds)

        raise AssertionError("This line should be unreachable.")

    def _process_audio_input(
            self, audio_input: UltravoxAudioInputs) -> NestedTensors:
        if audio_input["type"] == "audio_embeds":
            return audio_input["data"]

        audio_features = audio_input["data"]
        if isinstance(audio_features, torch.Tensor):
            # Combine the B and N dimensions for the encoder/projector
            flattened = flatten_bn(audio_features)
            flattened_embeddings = self._audio_features_to_embeddings(
                flattened)

            # Restore the original dimensions
            embeddings = flattened_embeddings.unflatten(
                0, audio_features.shape[:2])
            return embeddings

        result = []
        # TODO: Batch heterogeneous tensors through the encoder/projector
        for audio_features_item in audio_features:
            if isinstance(audio_features_item, torch.Tensor):
                result.append(
                    self._audio_features_to_embeddings(audio_features_item))
            else:
                embeddings = [
                    # Add a batch dimension to embed it, then remove it.
                    self._audio_features_to_embeddings(tensor.unsqueeze(0)
                                                       ).squeeze(0)
                    for tensor in audio_features_item
                ]
                result.append(embeddings)

        return result

    def get_multimodal_embeddings(self, **kwargs) -> Optional[NestedTensors]:
        audio_input = self._parse_and_validate_audio_input(**kwargs)
        if audio_input is None:
            return None
        audio_embeddings = self._process_audio_input(audio_input)
        return audio_embeddings

    def get_input_embeddings(
        self,
        input_ids: torch.Tensor,
        multimodal_embeddings: Optional[NestedTensors] = None,
        attn_metadata: Optional[AttentionMetadata] = None,
    ) -> torch.Tensor:
        inputs_embeds = self.language_model.get_input_embeddings(input_ids)
        if multimodal_embeddings is not None:

            # TODO(ywang96): use merge_multimodal_embeddings after
            # v0 is deprecated
            merge_multimodal_embeddings_from_map(
                inputs_embeds, multimodal_embeddings,
                attn_metadata.multi_modal_placeholder_index_maps["audio"])
        return inputs_embeds

    def forward(self,
                input_ids: torch.Tensor,
                positions: torch.Tensor,
                kv_caches: List[torch.Tensor],
                attn_metadata: AttentionMetadata,
                intermediate_tensors: Optional[torch.Tensor] = None,
                inputs_embeds: Optional[torch.Tensor] = None,
                **kwargs) -> Union[torch.Tensor, IntermediateTensors]:
        """Run forward pass for Ultravox

        One key thing to understand is the `input_ids` already accounts for the
        positions of the to-be-inserted audio embeddings. The to-be-inserted
        audio has a size that is essentially 6.25 tokens per second of audio.

        This way, the `positions` and `attn_metadata` are consistent
        with the `input_ids`.

        Args:
            audio_features: A batch of audio inputs [B, N, 80, M].
        """

        if intermediate_tensors is not None:
            inputs_embeds = None

        # NOTE: In v1, inputs_embeds is always generated at model runner, this
        # condition is for v0 compatibility.
        elif inputs_embeds is None:
            multimodal_embeddings = self.get_multimodal_embeddings(**kwargs)

            # TODO(ywang96): remove attn_metadata from get_input_embeddings
            # after v0 is deprecated
            inputs_embeds = self.get_input_embeddings(input_ids,
                                                      multimodal_embeddings,
                                                      attn_metadata)
            input_ids = None

        hidden_states = self.language_model.model(input_ids,
                                                  positions,
                                                  kv_caches,
                                                  attn_metadata,
                                                  intermediate_tensors,
                                                  inputs_embeds=inputs_embeds)
        return hidden_states

    def compute_logits(self, hidden_states: torch.Tensor,
                       sampling_metadata: SamplingMetadata) -> torch.Tensor:
        return self.language_model.compute_logits(hidden_states,
                                                  sampling_metadata)

    def sample(
        self,
        logits: torch.Tensor,
        sampling_metadata: SamplingMetadata,
    ) -> Optional[SamplerOutput]:
        return self.language_model.sample(logits, sampling_metadata)

    def load_weights(self, weights: Iterable[Tuple[str,
                                                   torch.Tensor]]) -> Set[str]:
        hf_to_vllm_mapper = WeightsMapper(
            orig_to_new_prefix={"audio_tower.model.encoder.": "audio_tower."})

        loader = AutoWeightsLoader(self,
                                   ignore_unexpected_prefixes=["audio_tower."])
        return loader.load_weights(weights, mapper=hf_to_vllm_mapper)