---
license: cc-by-nc-4.0
language:
- ja
pipeline_tag: image-to-text
tags:
- vision
- image-captioning
- VQA
---

# Chat-Vector-LLaVA-v1.5-7b-JA Model Card
## Model detail
**Model type:**
Chat-Vector-LLaVA-v1.5-7b-JA is a vision-language model that can converse about input images in Japanese.<br>
This model was created by adding and subtracting the weights of the [llava-v1.5-7b](https://huggingface.co/liuhaotian/llava-v1.5-7b), [Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf), and [ELYZA-japanese-Llama-2-7b](https://huggingface.co/elyza/ELYZA-japanese-Llama-2-7b) models using the Chat Vector method as follows.
```
ELYZA-japanese-Llama-2-7b + (llava-v1.5-7b - Llama-2-7b-hf)
```

Chat-Vector-LLaVA-v1.5-7b-JAは、入力画像について日本語で会話できるvision-language modelです。<br>
このモデルはChat Vectorの手法で[llava-v1.5-7b](https://huggingface.co/liuhaotian/llava-v1.5-7b)と[Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf)と[ELYZA-japanese-Llama-2-7b](https://huggingface.co/elyza/ELYZA-japanese-Llama-2-7b)のモデルの重みを以下の通り加減算することで作成しました。
```
ELYZA-japanese-Llama-2-7b + (llava-v1.5-7b - Llama-2-7b-hf)
```

**Comparing VLMs**
|Model|JA-VG-VQA-500<br>(ROUGE-L)|JA-VLM-Bench-In-the-Wild<br>(ROUGE-L)|Heron-Bench(Detail)|Heron-Bench(Conv)|Heron-Bench(Complex)|Heron-Bench(Average)
|-|-|-|-|-|-|-|
|[Japanese Stable VLM](https://huggingface.co/stabilityai/japanese-stable-vlm)|-|40.50|25.15|51.23|37.84|38.07|
|[EvoVLM-JP-v1-7B](https://huggingface.co/SakanaAI/EvoVLM-JP-v1-7B)|**19.70**|**51.25**|50.31|44.42|40.47|45.07|
|[Heron BLIP Japanese StableLM Base 7B llava-620k](https://huggingface.co/turing-motors/heron-chat-blip-ja-stablelm-base-7b-v1-llava-620k)|14.51|33.26|49.09|41.51|45.72|45.44|
|[Heron GIT Japanese StableLM Base 7B](https://huggingface.co/turing-motors/heron-chat-git-ja-stablelm-base-7b-v1)|15.18|37.82|42.77|**54.20**|43.53|46.83|
|[llava-jp-1.3b-v1.0-620k](https://huggingface.co/toshi456/llava-jp-1.3b-v1.0-620k)|12.69|44.58|51.21|41.05|45.95|44.84|
|[llava-jp-1.3b-v1.1](https://huggingface.co/toshi456/llava-jp-1.3b-v1.1)|13.33|44.40|50.00|51.83|**48.98**|**50.39**|
|[chat-vector-llava-v1.5-7b-ja](https://huggingface.co/toshi456/chat-vector-llava-v1.5-7b-ja)|18.64|42.23|**53.61**|44.36|44.48|46.10|


![image/png](https://cdn-uploads.huggingface.co/production/uploads/630af71ffaaea618ebc973db/jSW9RYPccrxaqrxntwtUb.png)

## How to use the model
**1. Download dependencies**
```
git clone https://github.com/tosiyuki/vlm-chat-vector-ja.git
```

**2. Inference**
```python
import requests
import torch
import transformers
from PIL import Image

from transformers.generation.streamers import TextStreamer
from llava.constants import DEFAULT_IMAGE_TOKEN, IMAGE_TOKEN_INDEX
from llava.conversation import conv_templates, SeparatorStyle
from llava.model.language_model.llava_llama import LlavaLlamaForCausalLM
from llava.mm_utils import tokenizer_image_token, process_images


if __name__ == "__main__":
    model_path = 'toshi456/chat-vector-llava-v1.5-7b-ja'
    device = "cuda" if torch.cuda.is_available() else "cpu"
    torch_dtype = torch.bfloat16 if device=="cuda" else torch.float32

    model = LlavaLlamaForCausalLM.from_pretrained(
        model_path, 
        device_map=device,
        low_cpu_mem_usage=True,
        use_safetensors=True,
        torch_dtype=torch.float16,
    ).eval()
    tokenizer = transformers.AutoTokenizer.from_pretrained(
        model_path,
        model_max_length=1024,
        padding_side="right",
        use_fast=False,
    )
    model.get_model().vision_tower.load_model()
    model = model.to(device)

    eos_token_id_list = [
        tokenizer.eos_token_id,
        tokenizer.bos_token_id,
    ]

    # image pre-process
    image_url = "https://huggingface.co/rinna/bilingual-gpt-neox-4b-minigpt4/resolve/main/sample.jpg"
    image = Image.open(requests.get(image_url, stream=True).raw).convert('RGB')

    if not isinstance(image, list):
        image = [image]
    
    image_tensor = process_images(image, model.get_model().vision_tower.image_processor, model.config)
    if type(image_tensor) is list:
        image_tensor = [image.to(model.device, dtype=torch.float16) for image in image_tensor]
    else:
        image_tensor = image_tensor.to(model.device, dtype=torch.float16)

    # create prompt
    # ユーザー: <image>\n{prompt}
        conv_mode = "llava_llama_2"
    conv = conv_templates[conv_mode].copy()
    prompt = "猫の隣には何がありますか？"
    inp = DEFAULT_IMAGE_TOKEN + '\n' + prompt
    conv.append_message(conv.roles[0], inp)
    conv.append_message(conv.roles[1], None)
    prompt = conv.get_prompt()

    input_ids = tokenizer_image_token(
        prompt, 
        tokenizer, 
        IMAGE_TOKEN_INDEX, 
        return_tensors='pt'
    ).unsqueeze(0)
    if device == "cuda":
        input_ids = input_ids.to(device)

    stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
    keywords = [stop_str]
    streamer = TextStreamer(tokenizer, skip_prompt=True, timeout=20.0)

    # parameter
    temperature = 0.0
    top_p = 1.0
    max_new_tokens=256

    # predict
    with torch.inference_mode():
        model.generate(
            inputs=input_ids,
            images=image_tensor,
            do_sample=True if temperature > 0 else False,
            temperature=temperature,
            top_p=top_p,
            max_new_tokens=max_new_tokens,
            streamer=streamer,
            use_cache=True,
            eos_token_id=eos_token_id_list,
        )

    """猫の隣には、コンピューター（パソコン）があります。<s>"""

```

## Acknowledgement
- [LLaVA](https://llava-vl.github.io/)
- [Chat Vector](https://arxiv.org/abs/2310.04799)

## License
cc-by-nc-4.0