processing_llava.py

# coding=utf-8
# Copyright 2023 The HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Processor class for Llava.
"""


from typing import List, Optional, Union

from transformers.feature_extraction_utils import BatchFeature
from transformers.image_utils import ImageInput
from transformers.tokenization_utils_base import (
    PaddingStrategy,
    PreTokenizedInput,
    TextInput,
    TruncationStrategy,
)
from transformers.utils import TensorType
import torch
from open_clip.transform import PreprocessCfg, image_transform_v2
from modeling_llava import LlavaForConditionalGeneration


class OpenCLIPImageProcessor:
    def __init__(self, config):
        cfg = PreprocessCfg(**config)
        transform = image_transform_v2(cfg=cfg, is_train=False)
        self.transform = transform

    def __call__(self, image, return_tensors):
        if isinstance(image, list):
            outputs = []
            for item in image:
                outputs.append(self.transform(item))
            return {
                "pixel_values": torch.tensor(outputs),
            }
        output = self.transform(image)
        return {
            "pixel_values": output.unsqueeze(0),
        }

    @property
    def model_input_names(self):
        return ["pixel_values"]


class LlavaProcessor:
    def __init__(self, image_processor: OpenCLIPImageProcessor, tokenizer):
        self.image_processor = image_processor
        self.tokenizer = tokenizer

    def __call__(
        self,
        text: Union[
            TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]
        ] = None,
        images: ImageInput = None,
        model: LlavaForConditionalGeneration = None,
        padding: Union[bool, str, PaddingStrategy] = False,
        truncation: Union[bool, str, TruncationStrategy] = None,
        max_length=None,
        return_tensors: Optional[Union[str, TensorType]] = TensorType.PYTORCH,
    ) -> BatchFeature:
        if images is not None:
            pixel_values = self.image_processor(images, return_tensors=return_tensors)[
                "pixel_values"
            ]
            pixel_values = pixel_values.to(model.device)
            image_outputs = model.vision_model(pixel_values)
            image_features = model.multi_modal_projector(image_outputs)
        else:
            image_features = None
        text_inputs = self.tokenizer(
            text,
            return_tensors=return_tensors,
            padding=padding,
            truncation=truncation,
            max_length=max_length,
        )

        return BatchFeature(data={**text_inputs, "image_features": image_features})

    def batch_decode(self, *args, **kwargs):
        return self.tokenizer.batch_decode(*args, **kwargs)

    def decode(self, *args, **kwargs):
        return self.tokenizer.decode(*args, **kwargs)

    @property
    def model_input_names(self):
        tokenizer_input_names = self.tokenizer.model_input_names
        image_processor_input_names = self.image_processor.model_input_names
        return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))