nvidia/NVLM-D-72B

Model Overview

Description

This family of models performs vision-language and text-only tasks including optical character recognition, multimodal reasoning, localization, common sense reasoning, world knowledge utilization, and coding.

License/Terms of Use

Creative Commons Attribution: Non-Commercial 4.0 International

Model Details

Today (September 17th, 2024), we introduce NVLM 1.0, a family of frontier-class multimodal large language models (LLMs) that achieve state-of-the-art results on vision-language tasks, rivaling the leading proprietary models (e.g., GPT-4o) and open-access models (e.g., Llama 3-V 405B and InternVL 2). Remarkably, NVLM 1.0 shows improved text-only performance over its LLM backbone after multimodal training.

In this repo, we are open-sourcing NVLM-1.0-D-72B (decoder-only architecture), the decoder-only model weights and code for the community.

Reference(s)

Paper   Inference Code (HF)   Training Code (Coming soon)   Website

Benchmark Results

We train our model with legacy Megatron-LM and adapt the codebase to Huggingface for model hosting, reproducibility, and inference. We observe numerical differences between the Megatron and Huggingface codebases, which are within the expected range of variation. We provide the results from both the Huggingface codebase and the Megatron codebase for reproducibility and comparison with other models.

Results (as of September 17th, 2024) in the multimodal benchmarks are as follows:

Vision-language Benchmarks

Benchmark	MMMU (val / test)	MathVista	OCRBench	AI2D	ChartQA	DocVQA	TextVQA	RealWorldQA	VQAv2
NVLM-D 1.0 72B (Huggingface)	58.7 / 54.9	65.2	852	94.2	86.0	92.6	82.6	69.5	85.4
NVLM-D 1.0 72B (Megatron)	59.7 / 54.6	65.2	853	94.2	86.0	92.6	82.1	69.7	85.4
Llama 3.2 90B	60.3 / -	57.3	-	92.3	85.5	90.1	-	-	78.1
Llama 3-V 70B	60.6 / -	-	-	93.0	83.2	92.2	83.4	-	79.1
Llama 3-V 405B	64.5 / -	-	-	94.1	85.8	92.6	84.8	-	80.2
InternVL2-Llama3-76B	55.2 / -	65.5	839	94.8	88.4	94.1	84.4	72.2	-
GPT-4V	56.8 / 55.7	49.9	645	78.2	78.5	88.4	78.0	61.4	77.2
GPT-4o	69.1 / -	63.8	736	94.2	85.7	92.8	-	-	-
Claude 3.5 Sonnet	68.3 / -	67.7	788	94.7	90.8	95.2	-	-	-
Gemini 1.5 Pro (Aug 2024)	62.2 / -	63.9	754	94.4	87.2	93.1	78.7	70.4	80.2

Text-only Benchmarks

Tasks	Backbone LLM	MMLU	GSM8K	MATH	HumanEval	Avg. Accuracy
Proprietary
GPT-4.0	N/A	88.7	-	76.6	90.2	-
Gemini Pro 1.5 (Aug 2024)	N/A	85.9	90.8	67.7	84.1	82.1
Claude 3.5 Sonnet	N/A	88.7	96.4	71.1	92.0	87.0
Open LLM
(a) Nous-Hermes-2-Yi-34B	N/A	75.5	78.6	21.8	43.3	54.8
(b) Qwen-72B-Instruct	N/A	82.3	91.1	59.7	86.0	79.8
(c) Llama-3-70B-Instruct	N/A	82.0	93.0	51.0	81.7	76.6
(d) Llama-3.1-70B-Instruct	N/A	83.6	95.1	68.0	80.5	81.8
(e) Llama-3.1-405B-Instruct	N/A	87.3	96.8	73.8	89.0	86.7
Open Multimodal LLM
VILA-1.5 40B	(a)	73.3	67.5	16.8	34.1	🥶 47.9 (-6.9)
LLaVA-OneVision 72B	(b)	80.6	89.9	49.2	74.4	🥶 73.5 (-6.3)
InternVL-2-Llama3-76B	(c)	78.5	87.1	42.5	71.3	🥶 69.9 (-6.7)
*Llama 3-V 70B	(d)	83.6	95.1	68.0	80.5	🙂 81.8 (0)
*Llama 3-V 405B	(e)	87.3	96.8	73.8	89.0	🙂 86.7 (0)
NVLM-D 1.0 72B (Megatron)	(b)	82.0	92.9	73.1	88.4	🥳 84.1 (+4.3)
NVLM-D 1.0 72B (Huggingface)	(b)	81.7	93.2	73.1	89.0	🥳 84.3 (+4.5)

Model Architectures

Network Architecture: Decoder-Only Transformer

Input

Input Type(s): Text, Image
Input Format(s): String, Pillow Library-Supported Formats
Input Dimensions: One-Dimensional (1D), Two Dimensional (2D)
Other Properties Related to Input: Maximum Token Length = 128K Tokens

Output

Output Type(s): Text
Output Format: String
Model Output: 1D
Other Properties Related to Output: None

How to use

When converting Megatron checkpoint to Huggingface, we adapt InternVL codebase to support model loading and multi-GPU inference in HF. We also use the tokenizer from Qwen2.5-72B-Instruct when adapting the tokenizer to Huggingface, as it contains extra special tokens for vision tasks, e.g., <|vision_pad|>. We train NVLM-1.0-D-72B based on the Qwen2-72B-Instruct text-only model and InternViT-6B-448px-V1-5 ViT model with our large-scale high-quality multimodal dataset. For training code, please refer to Megatron-LM (Coming soon).

Prepare the environment

We provide a docker build file in the Dockerfile for reproduction.

The docker image is based on nvcr.io/nvidia/pytorch:23.09-py3.

Note: We observe that different transformer versions / CUDA versions / docker versions can lead to slight benchmark number differences. We recommend using the Dockerfile above for precise reproduction.

Model loading

import torch
from transformers import AutoModel

path = "nvidia/NVLM-D-72B"
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=False,
    trust_remote_code=True).eval()

Multiple GPUs

The model can be loaded on multiple GPUs as follows:

import torch
import math
from transformers import AutoModel

def split_model():
    device_map = {}
    world_size = torch.cuda.device_count()
    num_layers = 80
    # Since the first GPU will be used for ViT, treat it as half a GPU.
    num_layers_per_gpu = math.ceil(num_layers / (world_size - 0.5))
    num_layers_per_gpu = [num_layers_per_gpu] * world_size
    num_layers_per_gpu[0] = math.ceil(num_layers_per_gpu[0] * 0.5)
    layer_cnt = 0
    for i, num_layer in enumerate(num_layers_per_gpu):
        for j in range(num_layer):
            device_map[f'language_model.model.layers.{layer_cnt}'] = i
            layer_cnt += 1
    device_map['vision_model'] = 0
    device_map['mlp1'] = 0
    device_map['language_model.model.tok_embeddings'] = 0
    device_map['language_model.model.embed_tokens'] = 0
    device_map['language_model.output'] = 0
    device_map['language_model.model.norm'] = 0
    device_map['language_model.lm_head'] = 0
    device_map['language_model.model.rotary_emb'] = 0
    device_map[f'language_model.model.layers.{num_layers - 1}'] = 0

    return device_map

path = "nvidia/NVLM-D-72B"
device_map = split_model()
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=False,
    trust_remote_code=True,
    device_map=device_map).eval()

Inference

import torch
from transformers import AutoTokenizer, AutoModel
import math
from PIL import Image
import torchvision.transforms as T
from torchvision.transforms.functional import InterpolationMode


def split_model():
    device_map = {}
    world_size = torch.cuda.device_count()
    num_layers = 80
    # Since the first GPU will be used for ViT, treat it as half a GPU.
    num_layers_per_gpu = math.ceil(num_layers / (world_size - 0.5))
    num_layers_per_gpu = [num_layers_per_gpu] * world_size
    num_layers_per_gpu[0] = math.ceil(num_layers_per_gpu[0] * 0.5)
    layer_cnt = 0
    for i, num_layer in enumerate(num_layers_per_gpu):
        for j in range(num_layer):
            device_map[f'language_model.model.layers.{layer_cnt}'] = i
            layer_cnt += 1
    device_map['vision_model'] = 0
    device_map['mlp1'] = 0
    device_map['language_model.model.tok_embeddings'] = 0
    device_map['language_model.model.embed_tokens'] = 0
    device_map['language_model.output'] = 0
    device_map['language_model.model.norm'] = 0
    device_map['language_model.lm_head'] = 0
    device_map['language_model.model.rotary_emb'] = 0
    device_map[f'language_model.model.layers.{num_layers - 1}'] = 0

    return device_map


IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)


def build_transform(input_size):
    MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    transform = T.Compose([
        T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
        T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
        T.ToTensor(),
        T.Normalize(mean=MEAN, std=STD)
    ])
    return transform


def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float('inf')
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    return best_ratio


def dynamic_preprocess(image, min_num=1, max_num=12, image_size=448, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set(
        (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
        i * j <= max_num and i * j >= min_num)
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    # find the closest aspect ratio to the target
    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, target_ratios, orig_width, orig_height, image_size)

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size
        )
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)
    assert len(processed_images) == blocks
    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images


def load_image(image_file, input_size=448, max_num=12):
    image = Image.open(image_file).convert('RGB')
    transform = build_transform(input_size=input_size)
    images = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, max_num=max_num)
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)
    return pixel_values

path = "nvidia/NVLM-D-72B"
device_map = split_model()
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=False,
    trust_remote_code=True,
    device_map=device_map).eval()

print(model)

tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True, use_fast=False)
generation_config = dict(max_new_tokens=1024, do_sample=False)

# pure-text conversation
question = 'Hello, who are you?'
response, history = model.chat(tokenizer, None, question, generation_config, history=None, return_history=True)
print(f'User: {question}\nAssistant: {response}')

# single-image single-round conversation
pixel_values = load_image('path/to/your/example/image.jpg', max_num=6).to(
    torch.bfloat16)
question = '<image>\nPlease describe the image shortly.'
response = model.chat(tokenizer, pixel_values, question, generation_config)
print(f'User: {question}\nAssistant: {response}')

Software Integration

Runtime Engine(s)

• PyTorch
  

Supported Hardware Microarchitecture Compatibility:

• NVIDIA Hopper
  

[Preferred/Supported] Operating System(s):

• Linux
  

Inference

Engine: PyTorch
Test Hardware:

• H100
  

Model Version(s)

• v1.0-D (NVLM-D)

Training, Testing, and Evaluation Datasets

Pre-Training Dataset

Link

• See Table 4
  

Data Collection Method by dataset

• Hybrid: Automated, Human, Synthetic, Unknown
  

Labeling Method by dataset

• Hybrid: Automated, Human, Synthetic, Unknown
  

Properties

• Trained on image captions, image-text pairs, natural images, charts, documents, scene descriptions, and mathematical reasoning.
  

Supervised Fine-Tuning Dataset

Link

• See Table 6
  

Data Collection Method by dataset

• Hybrid: Automated, Human, Synthetic, Unknown
  

Labeling Method by dataset

• Hybrid: Automated, Human, Synthetic, Unknown
  

Properties

• Trained on image captions; general knowledge; image-text pairs; natural images; charts; diagrams; documents; scene descriptions; science diagrams, lessons, textbook data, and question-answer pairs; visual instruction tuning; and mathematical reasoning.
  

Evaluation Dataset

Link

• See Section 6.1, "Benchmark"
  

Data collection method by dataset

• Human
  

Labeling method by dataset

• Human
  

Properties

• Evaluated on general knowledge, visual answering, chart understanding, table, optical character recognition, and mathematical reasoning.
  

Correspondence to

Wenliang Dai* ([email protected]), Nayeon Lee* ([email protected]), Boxin Wang* ([email protected]), Zhuolin Yang* ([email protected]), Wei Ping* ([email protected])

*Equal contribution

Citation

@article{nvlm2024,
  title={NVLM: Open Frontier-Class Multimodal LLMs},
  author={Dai, Wenliang and Lee, Nayeon and Wang, Boxin and Yang, Zhuolin and Liu, Zihan and Barker, Jon and Rintamaki, Tuomas and Shoeybi, Mohammad and Catanzaro, Bryan and Ping, Wei},
  journal={arXiv preprint},
  year={2024}}

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Please report security vulnerabilities or NVIDIA AI Concerns here.