360VL is developed based on the LLama3 language model and is also the industry's first open source large multi-modal model based on LLama3-70B[🤗Meta-Llama-3-70B-Instruct]. In addition to applying the Llama3 language model, the 360VL model also designs a globally aware multi-branch projector architecture, which enables the model to have more sufficient image understanding capabilities.

Github:https://github.com/360CVGroup/360VL

Model Zoo

360VL has released the following versions.

Model	Download
360VL-8B	🤗 Hugging Face
360VL-70B	🤗 Hugging Face

Features

360VL offers the following features:

Multi-round text-image conversations: 360VL can take both text and images as inputs and produce text outputs. Currently, it supports multi-round visual question answering with one image.
Bilingual text support: 360VL supports conversations in both English and Chinese, including text recognition in images.
Strong image comprehension: 360VL is adept at analyzing visuals, making it an efficient tool for tasks like extracting, organizing, and summarizing information from images.
Fine-grained image resolution: 360VL supports image understanding at a higher resolution of 672×672.

Performance

Model	Checkpoints	MMB_T	MMB_D	MMB-CN_T	MMB-CN_D	MMMU_V	MMMU_T	MME
QWen-VL-Chat	🤗LINK	61.8	60.6	56.3	56.7	37	32.9	1860
mPLUG-Owl2	🤖LINK	66.0	66.5	60.3	59.5	34.7	32.1	1786.4
CogVLM	🤗LINK	65.8	63.7	55.9	53.8	37.3	30.1	1736.6
Monkey-Chat	🤗LINK	72.4	71	67.5	65.8	40.7	-	1887.4
MM1-7B-Chat	LINK	-	72.3	-	-	37.0	35.6	1858.2
IDEFICS2-8B	🤗LINK	75.7	75.3	68.6	67.3	43.0	37.7	1847.6
SVIT-v1.5-13B	🤗LINK	69.1	-	63.1	-	38.0	33.3	1889
LLaVA-v1.5-13B	🤗LINK	69.2	69.2	65	63.6	36.4	33.6	1826.7
LLaVA-v1.6-13B	🤗LINK	70	70.7	68.5	64.3	36.2	-	1901
Honeybee	LINK	73.6	74.3	-	-	36.2	-	1976.5
YI-VL-34B	🤗LINK	72.4	71.1	70.7	71.4	45.1	41.6	2050.2
360VL-8B	🤗LINK	75.3	73.7	71.1	68.6	39.7	37.1	1944.6
360VL-70B	🤗LINK	78.1	80.4	76.9	77.7	50.8	44.3	2012.3

Quick Start 🤗

Shell

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
from PIL import Image

checkpoint = "qihoo360/360VL-70B"

model = AutoModelForCausalLM.from_pretrained(checkpoint, torch_dtype=torch.float16, device_map='auto', trust_remote_code=True).eval()
tokenizer = AutoTokenizer.from_pretrained(checkpoint, trust_remote_code=True)
vision_tower = model.get_vision_tower()
vision_tower.load_model()
vision_tower.to(device="cuda", dtype=torch.float16)
image_processor = vision_tower.image_processor
tokenizer.pad_token = tokenizer.eos_token


image = Image.open("docs/008.jpg").convert('RGB')
query = "Who is this cartoon character?"
terminators = [
    tokenizer.convert_tokens_to_ids("<|eot_id|>",)
]

inputs = model.build_conversation_input_ids(tokenizer, query=query, image=image, image_processor=image_processor)

input_ids = inputs["input_ids"].to(device='cuda', non_blocking=True)
images = inputs["image"].to(dtype=torch.float16, device='cuda', non_blocking=True)

output_ids = model.generate(
    input_ids,
    images=images,
    do_sample=False,
    eos_token_id=terminators,
    num_beams=1,
    max_new_tokens=512,
    use_cache=True)

input_token_len = input_ids.shape[1]
outputs = tokenizer.batch_decode(output_ids[:, input_token_len:], skip_special_tokens=True)[0]
outputs = outputs.strip()
print(outputs)

Model type: 360VL-70B is an open-source chatbot trained by fine-tuning LLM on multimodal instruction-following data. It is an auto-regressive language model, based on the transformer architecture. Base LLM: meta-llama/Meta-Llama-3-70B-Instruct

Model date: 360VL-70B was trained in May 2024.

License

This project utilizes certain datasets and checkpoints that are subject to their respective original licenses. Users must comply with all terms and conditions of these original licenses. The content of this project itself is licensed under the Apache license 2.0

Where to send questions or comments about the model: https://github.com/360CVGroup/360VL

Related Projects

This work wouldn't be possible without the incredible open-source code of these projects. Huge thanks!