Pangea-7B Model Card

Pangea: A Fully Open Multilingual Multimodal LLM for 39 Languages

🇪🇹 🇸🇦 🇧🇬 🇧🇩 🇨🇿 🇩🇪 🇬🇷 🇬🇧 🇺🇸 🇪🇸 🇮🇷 🇫🇷 🇮🇪 🇮🇳 🇮🇩 🇳🇬 🇮🇹 🇮🇱 🇯🇵 🇮🇩 🇰🇷 🇳🇱 🇲🇳 🇲🇾 🇳🇴 🇵🇱 🇵🇹 🇧🇷 🇷🇴 🇷🇺 🇱🇰 🇮🇩 🇰🇪 🇹🇿 🇱🇰 🇹🇭 🇹🇷 🇺🇦 🇵🇰 🇻🇳 🇨🇳 🇹🇼

👁 description

Model details

Model: Pangea is a fully open-source Multilingual Multimodal Multicultural LLM.
Date: Pangea-7B was trained in 2024.
Training Dataset: 6M PangeaIns.
Architecture: Pangea-7B follows the architecture of LLaVA-NeXT, with a Qwen2-7B-Instruct backbone.

Uses

Pangea-7B follows the architecture of LLaVA-NeXT.

You could either (1) follow the same model loading procedures as of LLaVA-NeXT, an example of loading Pangea-7B directly is shown in the Python code below, or (2) use our hf version of Pangea-7B: [Pangea-7B-hf]https://huggingface.co/neulab/Pangea-7B-hf

Direct Use

First you would need to clone and install LLaVA-NeXT.

git clone https://github.com/LLaVA-VL/LLaVA-NeXT
cd LLaVA-NeXT
pip install -e ".[train]"

Then, you could load Pangea-7B using the following code:

from llava.model.builder import load_pretrained_model
model_path = 'neulab/Pangea-7B'
model_name = 'Pangea-7B-qwen'
args = {"multimodal": True}
tokenizer, model, image_processor, context_len = load_pretrained_model(model_path, None, model_name, **args)

Defining some helper functions for using the model:

import torch
from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN
from llava.utils import disable_torch_init
from llava.constants import IGNORE_INDEX, DEFAULT_IMAGE_TOKEN, IMAGE_TOKEN_INDEX
from typing import Dict
import transformers
import re
from PIL import Image

def preprocess_qwen(sources, tokenizer: transformers.PreTrainedTokenizer, has_image: bool = False, max_len=2048, system_message: str = "You are a helpful assistant.") -> Dict:
 roles = {"human": "<|im_start|>user", "gpt": "<|im_start|>assistant"}
 im_start, im_end = tokenizer.additional_special_tokens_ids
 nl_tokens = tokenizer("\n").input_ids
 _system = tokenizer("system").input_ids + nl_tokens
 _user = tokenizer("user").input_ids + nl_tokens
 _assistant = tokenizer("assistant").input_ids + nl_tokens
 input_ids = []
 source = sources
 if roles[source[0]["from"]] != roles["human"]: source = source[1:]
 input_id, target = [], []
 system = [im_start] + _system + tokenizer(system_message).input_ids + [im_end] + nl_tokens
 input_id += system
 target += [im_start] + [IGNORE_INDEX] * (len(system) - 3) + [im_end] + nl_tokens
 assert len(input_id) == len(target)
 for j, sentence in enumerate(source):
 role = roles[sentence["from"]]
 if has_image and sentence["value"] is not None and "<image>" in sentence["value"]:
 num_image = len(re.findall(DEFAULT_IMAGE_TOKEN, sentence["value"]))
 texts = sentence["value"].split('<image>')
 _input_id = tokenizer(role).input_ids + nl_tokens 
 for i,text in enumerate(texts):
 _input_id += tokenizer(text).input_ids 
 if i<len(texts)-1: _input_id += [IMAGE_TOKEN_INDEX] + nl_tokens
 _input_id += [im_end] + nl_tokens
 assert sum([i==IMAGE_TOKEN_INDEX for i in _input_id])==num_image
 else:
 if sentence["value"] is None: _input_id = tokenizer(role).input_ids + nl_tokens
 else: _input_id = tokenizer(role).input_ids + nl_tokens + tokenizer(sentence["value"]).input_ids + [im_end] + nl_tokens
 input_id += _input_id
 input_ids.append(input_id)
 return torch.tensor(input_ids, dtype=torch.long)

def generate_output(prompt, image=None, do_sample=False, temperature=0, top_p=0.5, num_beams=1, max_new_tokens=1024):
 image_tensors = []
 prompt = "<image>\n" + prompt
 image = Image.open(image)
 image_tensor = image_processor.preprocess(image, return_tensors='pt')['pixel_values']
 image_tensors.append(image_tensor.half().cuda())
 input_ids = preprocess_qwen([{'from': 'human', 'value': prompt},{'from': 'gpt','value': None}], tokenizer, has_image=True).cuda()
 with torch.inference_mode():
 output_ids = model.generate(
 input_ids,
 images=image_tensors,
 do_sample=do_sample,
 temperature=temperature,
 top_p=top_p,
 num_beams=num_beams,
 max_new_tokens=max_new_tokens,
 use_cache=True
 )
 outputs = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0]
 outputs = outputs.strip()
 return outputs

Now, an example of using the model:

prompt = "What did you see in the image?"
image = "image.png"
print(generate_output(prompt, image=image))

Note that the example above demonstrates multimodal usage. To use the model with text-only inputs, you would need to reload the model with :

args = {"multimodal": True}
tokenizer, model, _, context_len = load_pretrained_model(model_path, None, model_name, **args)

def generate_output_text_only(prompt, do_sample=False, temperature=0, top_p=0.5, num_beams=1, max_new_tokens=1024):
 input_ids = preprocess_qwen([{'from': 'human', 'value': prompt},{'from': 'gpt','value': None}], tokenizer, has_image=False).cuda()
 with torch.inference_mode():
 generated_ids = model.generate(
 input_ids,
 do_sample=do_sample,
 temperature=temperature,
 top_p=top_p,
 num_beams=num_beams,
 max_new_tokens=max_new_tokens,
 use_cache=True
 )
 generated_ids = [output_ids[len(input_ids) :] for input_ids, output_ids in zip(input_ids, generated_ids)]
 outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
 outputs = outputs.strip()
 return outputs

prompt = "Write me a python function that could sort a input integer list by descending order"
print(generate_output_text_only(prompt))

Citing the Model

BibTeX Citation:

@article{yue2024pangeafullyopenmultilingual,
 title={Pangea: A Fully Open Multilingual Multimodal LLM for 39 Languages},
 author={Xiang Yue and Yueqi Song and Akari Asai and Seungone Kim and Jean de Dieu Nyandwi and Simran Khanuja and Anjali Kantharuban and Lintang Sutawika and Sathyanarayanan Ramamoorthy and Graham Neubig},
 year={2024},
 journal={arXiv preprint arXiv:2410.16153},
 url={https://arxiv.org/abs/2410.16153}
}