DeepSeek-VL2/inference.py

# Copyright (c) 2023-2024 DeepSeek.
#
# Permission is hereby granted, free of charge, to any person obtaining a copy of
# this software and associated documentation files (the "Software"), to deal in
# the Software without restriction, including without limitation the rights to
# use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
# the Software, and to permit persons to whom the Software is furnished to do so,
# subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
# FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
# COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
# IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
# CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

from argparse import ArgumentParser
from typing import List, Dict
import torch
from transformers import AutoModelForCausalLM

import PIL.Image

from deepseek_vl.models import DeepseekVLV2ForCausalLM, DeepseekVLV2Processor
from deepseek_vl.serve.app_modules.utils import parse_ref_bbox


def load_pil_images(conversations: List[Dict[str, str]]) -> List[PIL.Image.Image]:
    """

    Args:
        conversations (List[Dict[str, str]]): the conversations with a list of messages. An example is :
            [
                {
                    "role": "User",
                    "content": "<image>\nExtract all information from this image and convert them into markdown format.",
                    "images": ["./examples/table_datasets.png"]
                },
                {"role": "Assistant", "content": ""},
            ]

    Returns:
        pil_images (List[PIL.Image.Image]): the list of PIL images.

    """

    pil_images = []

    for message in conversations:
        if "images" not in message:
            continue

        for image_path in message["images"]:
            pil_img = PIL.Image.open(image_path)
            pil_img = pil_img.convert("RGB")
            pil_images.append(pil_img)

    return pil_images


def main(args):

    dtype = torch.bfloat16

    # specify the path to the model
    model_path = args.model_path
    vl_chat_processor: DeepseekVLV2Processor = DeepseekVLV2Processor.from_pretrained(model_path)
    tokenizer = vl_chat_processor.tokenizer

    vl_gpt: DeepseekVLV2ForCausalLM = AutoModelForCausalLM.from_pretrained(
        model_path,
        trust_remote_code=True,
        torch_dtype=dtype
    )
    vl_gpt = vl_gpt.cuda().eval()

    # single image conversation example
    conversation = [
        {
            "role": "<|User|>",
            "content": "<image>\n<|ref|>The giraffe at the back.<|/ref|>.",
            "images": ["./images/visual_grounding.jpeg"],
        },
        {"role": "<|Assistant|>", "content": ""},
    ]

    # load images and prepare for inputs
    pil_images = load_pil_images(conversation)
    prepare_inputs = vl_chat_processor.__call__(
        conversations=conversation,
        images=pil_images,
        force_batchify=True,
        system_prompt=""
    ).to(vl_gpt.device, dtype=dtype)

    with torch.no_grad():
        # run image encoder to get the image embeddings
        inputs_embeds = vl_gpt.prepare_inputs_embeds(**prepare_inputs)

        # run the model to get the response
        outputs = vl_gpt.generate(
            inputs_embeds=inputs_embeds,
            attention_mask=prepare_inputs.attention_mask,
            pad_token_id=tokenizer.eos_token_id,
            bos_token_id=tokenizer.bos_token_id,
            eos_token_id=tokenizer.eos_token_id,
            max_new_tokens=1024,

            do_sample=False,
            # repetition_penalty=1.1,

            # do_sample=True,
            # temperature=1.0,
            # top_p=0.9,
            # repetition_penalty=1.1,

            use_cache=True,
        )

        answer = tokenizer.decode(outputs[0].cpu().tolist(), skip_special_tokens=False)
        print(f"{prepare_inputs['sft_format'][0]}", answer)

        vg_image = parse_ref_bbox(answer, image=pil_images[0])
        if vg_image is not None:
            vg_image.save("./vg.jpg", format="JPEG", quality=85)


if __name__ == "__main__":
    parser = ArgumentParser()
    parser.add_argument("--model_path", type=str, required=True,
                        default="deepseek-ai/deepseek-vl2-27b-moe",
                        help="model name or local path to the model")
    args = parser.parse_args()
    main(args)