---
license: cc-by-nc-4.0
language:
- ja
library_name: transformers
tags:
- vision
- image-captioning
---

# Chatvector-llava-v1.5-plus-Houou-v3-7b Model Card
# Model Details
※好奇心から生まれたモデルです。精度は保証できませんが、v1.6を用いたものよりは良い気がしています。<br>
chatvector-llava-v1.5-plus-houou-v3-7bは日本語で画像を説明することが可能なVLMです。<br>
[Chat Vector](https://arxiv.org/abs/2310.04799)の手法に影響を受けています。
このモデルはChat Vectorを参考に[llava-v1.5-7b](https://huggingface.co/liuhaotian/llava-v1.5-7b)と[houou-instruction-7b-v3](https://huggingface.co/moneyforward/houou-instruction-7b-v3)、[Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf)
の重みを以下のように加減算することで作成してみました。<br>
```
houou-instruction-7b-v3 + (llava-v1.5-7b - Llama-2-7b-hf)
```

次のプログラムは引用させていただいたサイトにあったものをベースにしています。以下文献もぜひご覧ください。

## Uses
```sh
git clone https://github.com/haotian-liu/LLaVA.git
cd LLaVA
pip install -e .
```
```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

model_path = "shinyice/chatvector-llava-v1.5-plus-houou-v3-7b"
device = "cuda" if torch.cuda.is_available() else "cpu"

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_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)
image_sizes = [img.size for img in image]

if isinstance(image_tensor, list):
    image_tensor = [img.to(model.device, dtype=torch.float16) for img in image_tensor]
else:
    image_tensor = image_tensor.to(device, dtype=torch.float16)

image_sizes_tensor = torch.tensor(image_sizes, dtype=torch.int32, device=device)

conv_mode = "v1" 
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)

temperature = 0.0
top_p = 1.0
max_new_tokens = 256

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

print(tokenizer.decode(output[0]))
```
## Bibliography 

- [Chat VectorでLLaVAを日本語対応させる](https://zenn.dev/toshi_456/articles/0166a6eaa81c7b)
- [Chat Vectorを使って日本語LLMをチャットモデルに改造する](https://qiita.com/jovyan/items/ee6affa5ee5bdaada6b4)