File size: 34,325 Bytes

---
license: mit
pipeline_tag: image-text-to-text
library_name: transformers
base_model:
  - OpenGVLab/InternViT-6B-448px-V2_5
  - Qwen/Qwen2.5-32B-Instruct
base_model_relation: merge
language:
  - multilingual
tags:
  - internvl
  - vision
  - ocr
  - multi-image
  - video
  - custom_code
---

# InternVL2_5-38B

[\[📂 GitHub\]](https://github.com/OpenGVLab/InternVL)  [\[🆕 Blog\]](https://internvl.github.io/blog/)  
[\[📜 InternVL 2.5 Report\]]()
[\[📜 InternVL 1.0 Paper\]](https://arxiv.org/abs/2312.14238)  [\[📜 InternVL 1.5 Report\]](https://arxiv.org/abs/2404.16821)
[\[🗨️ Chat Demo\]](https://internvl.opengvlab.com/)  [\[🤗 HF Demo\]](https://huggingface.co/spaces/OpenGVLab/InternVL)  [\[🚀 Quick Start\]](#quick-start)    [\[📖 Documents\]](https://internvl.readthedocs.io/en/latest/)

![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/64564b0e4a7ffb7d5a47f412/3i-8-6VSoTAo0-OKUUpec.jpeg)

## Introduction

We are excited to introduce InternVL 2.5, an advanced multimodal large language model (MLLM) series that builds upon InternVL 2.0, maintaining its core model architecture while introducing significant enhancements in training and testing strategies as well as data quality.

Through extensive evaluations on a wide range of benchmarks, including multi-discipline reasoning, document understanding, multi-image / video understanding, real-world comprehension, multimodal hallucination detection, visual grounding, multilingual capabilities, and pure language processing, InternVL 2.5 exhibits competitive performance, rivaling leading commercial models such as GPT-4o and Claude-3.5-Sonnet. Notably, our model is the first open-source MLLMs to achieve over **70%** on the **MMMU benchmark**. We hope this model contributes to the open-source community by setting new standards for developing and applying multimodal AI systems. This repository contains the instruction-tuned **InternVL2_5-38B** model.

We delve into the relationship between model scaling and performance, systematically exploring the performance trends in vision encoders, language models, dataset sizes, and test-time configurations. For more details, please refer to our [blog](), [tech report]() and [GitHub](https://github.com/OpenGVLab/InternVL).

|      Model Name      |                                     Vision Part                                     |                                        Language Part                                         |                             HF Link                              | 
| :------------------: | :---------------------------------------------------------------------------------: | :------------------------------------------------------------------------------------------: | :--------------------------------------------------------------: |
|     InternVL2_5-1B     |    [InternViT-300M-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-300M-448px-V2_5)    |            [Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct)            |     [🤗 link](https://huggingface.co/OpenGVLab/InternVL2_5-1B)     |  
|     InternVL2_5-2B     |    [InternViT-300M-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-300M-448px-V2_5)    |          [internlm2_5-1_8b-chat](https://huggingface.co/internlm/internlm2_5-1_8b-chat)          |     [🤗 link](https://huggingface.co/OpenGVLab/InternVL2_5-2B)     |     
|     InternVL2_5-4B     |    [InternViT-300M-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-300M-448px-V2_5)    |    [Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct)     |     [🤗 link](https://huggingface.co/OpenGVLab/InternVL2_5-4B)     |     
|     InternVL2_5-8B     |    [InternViT-300M-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-300M-448px-V2_5)    |          [internlm2_5-7b-chat](https://huggingface.co/internlm/internlm2_5-7b-chat)          |     [🤗 link](https://huggingface.co/OpenGVLab/InternVL2_5-8B)     |     
|    InternVL2_5-26B     | [InternViT-6B-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-6B-448px-V2_5) |           [internlm2_5-20b-chat](https://huggingface.co/internlm/internlm2_5-20b-chat)           |    [🤗 link](https://huggingface.co/OpenGVLab/InternVL2_5-26B)     |   
|    InternVL2_5-38B     | [InternViT-6B-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-6B-448px-V2_5) |       [Qwen2.5-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct)       |    [🤗 link](https://huggingface.co/OpenGVLab/InternVL2_5-38B)     | 
| InternVL2_5-78B | [InternViT-6B-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-6B-448px-V2_5) | [Qwen2.5-72B-Instruct](https://huggingface.co/Qwen/Qwen2.5-72B-Instruct) | [🤗 link](https://huggingface.co/OpenGVLab/InternVL2_5-78B) |

## Model Details

InternVL 2.5 is a multimodal large language model series, featuring models of various sizes. For each size, we release instruction-tuned models optimized for multimodal tasks. InternVL2_5-38B consists of [InternViT-6B-448px-V2_5](https://huggingface.co/OpenGVLab/InternViT-6B-448px-V2_5), an MLP projector, and [Qwen2.5-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct).

## Performance

### Image Benchmarks

| Benchmark                  | InternVL2.5-26B | Cambrian-34B | VILA-1.5-40B | InternVL2.5-38B |
|----------------------------|-----------------|--------------|--------------|-----------------|
| MMMU (val)                 | 60.0            | 49.7         | 55.1         | 63.9            |
| MMMU (test)                | 51.8            | -            | 46.9         | 57.6            |
| MMMU-PRO (overall)         | 37.1            | -            | 25.0         | 46.0            |
| MathVista (mini)           | 67.7            | 53.2         | 49.5         | 71.9            |
| MathVision (mini)          | 28.0            | -            | -            | 32.2            |
| MathVision (full)          | 23.1            | -            | -            | 31.8            |
| MathVerse (mini)           | 40.1            | -            | -            | 49.4            |
| Olympiad Bench             | 8.8             | -            | -            | 12.1            |
| AI2D (w / wo M)            | 86.4 / 94.4     | 79.5 / -     | 69.9 / -     | 87.6 / 95.1     |
| ChartQA (test avg.)        | 87.2            | 75.6         | 67.2         | 88.2            |
| TextVQA (val)              | 82.4            | 76.7         | 73.6         | 82.7            |
| DocVQA (test)              | 94.0            | 75.5         | -            | 95.3            |
| InfoVQA (test)             | 79.8            | 46.0         | -            | 83.6            |
| OCR-Bench                  | 852             | 600          | 460          | 842             |
| SEED-2 Plus                | 70.8            | -            | -            | 71.2            |
| CharXiv (RQ / DQ)          | 35.9 / 73.5     | 27.3 / 59.7  | 24.0 / 38.7  | 42.4 / 79.6     |
| VCR-EN-Easy (EM / Jaccard) | 94.4 / 98.0     | 79.7 / 89.3  | -            | 94.7 / 98.2     |
| BLINK (val)                | 61.8            | -            | -            | 63.2            |
| Mantis Eval                | 75.6            | -            | -            | 78.3            |
| MMIU                       | 49.4            | -            | -            | 55.3            |
| Muir Bench                 | 61.1            | -            | -            | 62.7            |
| MMT (val)                  | 66.9            | -            | -            | 70.0            |
| MIRB (avg.)                | 55.7            | -            | -            | 61.2            |
| RealWorld QA               | 74.5            | 67.8         | -            | 73.5            |
| MME-RW (EN)                | 61.8            | 44.1         | -            | 64.0            |
| WildVision (win rate)      | 65.2            | -            | -            | 66.4            |
| R-Bench                    | 72.9            | -            | -            | 72.1            |
| MME (sum)                  | 2373.3          | -            | -            | 2455.8          |
| MMB (EN / CN)              | 85.4 / 85.5     | 80.4 / 79.2  | -            | 86.5 / 86.3     |
| MMBv1.1 (EN)               | 84.2            | 78.3         | -            | 85.5            |
| MMVet (turbo)              | 65.0            | 53.2         | -            | 68.8            |
| MMVetv2 (0613)             | 60.8            | -            | -            | 62.1            |
| MMStar                     | 66.5            | 54.2         | -            | 67.9            |
| HallBench (avg.)           | 55.0            | 41.6         | -            | 56.8            |
| MMHal (score)              | 3.70            | -            | -            | 3.71            |
| CRPE (relation)            | 79.1            | -            | -            | 78.3            |
| POPE (avg.)                | 90.6            | -            | -            | 90.7            |




### Video Benchmarks

| Model Name                                  | Video-MME (wo / w sub)        | MVBench | MMBench-Video (val) | MLVU (M-Avg) | LongVideoBench (val total) | CG-Bench v1.1 (long / clue acc.)     |
|---------------------------------------------|-------------|------|-------|-------|------|-------------|
| **InternVL2.5-1B**                              | 50.3 / 52.3 | 64.3 | 1.36  | 57.3  | 47.9 | -           |
| Qwen2-VL-2B          | 55.6 / 60.4 | 63.2 | -     | -     | -    | -           |
| **InternVL2.5-2B**                              | 51.9 / 54.1 | 68.8 | 1.44  | 61.4  | 52.0 | -           |
| **InternVL2.5-4B**                              | 62.3 / 63.6 | 71.6 | 1.73  | 68.3  | 55.2 | -           |
| VideoChat2-HD        | 45.3 / 55.7 | 62.3 | 1.22  | 47.9  | -    | -           |
| MiniCPM-V-2.6         | 60.9 / 63.6 | -    | 1.70  | -     | 54.9 | -           |
| LLaVA-OneVision-7B     | 58.2 /  -  | 56.7 | -     | -     | -    | -           |
| Qwen2-VL-7B          | 63.3 / 69.0 | 67.0 | 1.44  | -     | 55.6 | -           |
| **InternVL2.5-8B**                              | 64.2 / 66.9 | 72.0 | 1.68  | 68.9  | 60.0 | -           |
| **InternVL2.5-26B**                             | 66.9 / 69.2 | 75.2 | 1.86  | 72.3  | 59.9 | -           |
| Oryx-1.5-32B                                | 67.3 / 74.9 | 70.1 | 1.52  | 72.3  | -    | -           |
| VILA-1.5-40B             | 60.1 / 61.1 | -    | 1.61  | 56.7  | -    | -           |
| **InternVL2.5-38B**                             | 70.7 / 73.1 | 74.4 | 1.82  | 75.3  | 63.3 | -           |
| GPT-4V/4T             | 59.9 / 63.3 | 43.7 | 1.53  | 49.2  | 59.1 | -           |
| GPT-4o-20240513                | 71.9 / 77.2 | -    | 1.63  | 64.6  | 66.7 | -           |
| GPT-4o-20240806                | -           | -    | 1.87  | -     | -    | -           |
| Gemini-1.5-Pro     | 75.0 / 81.3 | -    | 1.30  | -     | 64.0 | -           |
| VideoLLaMA2-72B | 61.4 / 63.1 | 62.0 | -     | -     | -    | -           |
| LLaVA-OneVision-72B    | 66.2 / 69.5 | 59.4 | -     | 66.4  | 61.3 | -           |
| Qwen2-VL-72B         | 71.2 / 77.8 | 73.6 | 1.70  | -     | -    | 41.3 / 56.2 |
| InternVL2-Llama3-76B     | 64.7 / 67.8 | 69.6 | 1.71  | 69.9  | 61.1 | -           |
| **InternVL2.5-78B**                             | 72.1 / 74.0 | 76.4 | 1.97  | 75.7  | 63.6 | 42.2 / 58.5 |

### Multimodal Multilingual Understanding
<table style="width: 100%; font-size: 8px; border-collapse: collapse; text-align: center;">
  <thead>
    <tr>
      <th rowspan="2">Model Name</th>
      <th colspan="6">MMMB</th>
      <th colspan="6">Multilingual MMBench</th>
      <th>MTVQA</th>
    </tr>
    <tr>
      <th>en</th>
      <th>zh</th>
      <th>pt</th>
      <th>ar</th>
      <th>tr</th>
      <th>ru</th>
      <th>en</th>
      <th>zh</th>
      <th>pt</th>
      <th>ar</th>
      <th>tr</th>
      <th>ru</th>
      <th>(avg)</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>InternVL-Chat-V1.5</td>
      <td>82.6</td>
      <td>80.8</td>
      <td>76.3</td>
      <td>65.2</td>
      <td>68.6</td>
      <td>74.0</td>
      <td>81.1</td>
      <td>80.2</td>
      <td>76.9</td>
      <td>56.2</td>
      <td>66.7</td>
      <td>71.0</td>
      <td>20.5</td>
    </tr>
    <tr>
      <td>InternVL2-26B</td>
      <td>83.8</td>
      <td>81.7</td>
      <td>78.0</td>
      <td>68.8</td>
      <td>69.3</td>
      <td>76.3</td>
      <td>82.7</td>
      <td>81.8</td>
      <td>77.8</td>
      <td>61.9</td>
      <td>69.6</td>
      <td>74.4</td>
      <td>17.7</td>
    </tr>
    <tr>
      <td>InternVL2.5-26B</td>
      <td>86.2</td>
      <td>83.8</td>
      <td>81.6</td>
      <td>73.3</td>
      <td>73.7</td>
      <td>82.8</td>
      <td>86.1</td>
      <td>85.5</td>
      <td>80.7</td>
      <td>67.5</td>
      <td>75.0</td>
      <td>79.6</td>
      <td>28.5</td>
    </tr>
    <tr>
      <td>InternVL2-40B</td>
      <td>85.3</td>
      <td>84.1</td>
      <td>81.1</td>
      <td>70.3</td>
      <td>74.2</td>
      <td>81.4</td>
      <td>86.2</td>
      <td>85.8</td>
      <td>82.8</td>
      <td>64.0</td>
      <td>74.2</td>
      <td>81.8</td>
      <td>20.6</td>
    </tr>
    <tr>
      <td>InternVL2.5-38B</td>
      <td>86.4</td>
      <td>85.1</td>
      <td>84.1</td>
      <td>84.3</td>
      <td>82.8</td>
      <td>84.9</td>
      <td>87.5</td>
      <td>88.6</td>
      <td>85.3</td>
      <td>84.5</td>
      <td>84.0</td>
      <td>85.9</td>
      <td>31.7</td>
    </tr>
  </tbody>
</table>



### Invitation to Evaluate InternVL

We welcome MLLM benchmark developers to assess our InternVL series models. If you need to add your evaluation results here, please contact me at [wztxy89@163.com](mailto:wztxy89@163.com).

## Quick Start

We provide an example code to run InternVL2_5-38B using `transformers`.

We also welcome you to experience the InternVL series models in our [online demo](https://internvl.opengvlab.com/).

> Please use transformers ≳ 4.37.2 to ensure the model works normally.

### Model Loading

#### 16-bit (bf16 / fp16)

```python
import torch
from transformers import AutoTokenizer, AutoModel
path = "OpenGVLab/InternVL2_5-38B"
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=True,
    trust_remote_code=True).eval().cuda()
```

#### BNB 8-bit Quantization

```python
import torch
from transformers import AutoTokenizer, AutoModel
path = "OpenGVLab/InternVL2_5-38B"
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    load_in_8bit=True,
    low_cpu_mem_usage=True,
    use_flash_attn=True,
    trust_remote_code=True).eval()
```

#### BNB 4-bit Quantization

```python
import torch
from transformers import AutoTokenizer, AutoModel
path = "OpenGVLab/InternVL2_5-38B"
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    load_in_4bit=True,
    low_cpu_mem_usage=True,
    use_flash_attn=True,
    trust_remote_code=True).eval()
```

#### Multiple GPUs

The reason for writing the code this way is to avoid errors that occur during multi-GPU inference due to tensors not being on the same device. By ensuring that the first and last layers of the large language model (LLM) are on the same device, we prevent such errors.

```python
import math
import torch
from transformers import AutoTokenizer, AutoModel

def split_model(model_name):
    device_map = {}
    world_size = torch.cuda.device_count()
    num_layers = {
        'InternVL2_5-1B': 24, 'InternVL_5-2B': 24, 'InternVL2_5-4B': 36, 'InternVL2_5-8B': 32,
        'InternVL2_5-26B': 48, 'InternVL2_5-38B': 64, 'InternVL2_5-78B': 80}[model_name]
    # 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[f'language_model.model.layers.{num_layers - 1}'] = 0

    return device_map

path = "OpenGVLab/InternVL2_5-38B"
device_map = split_model('InternVL2_5-38B')
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=True,
    trust_remote_code=True,
    device_map=device_map).eval()
```

### Inference with Transformers

```python
import numpy as np
import torch
import torchvision.transforms as T
from decord import VideoReader, cpu
from PIL import Image
from torchvision.transforms.functional import InterpolationMode
from transformers import AutoModel, AutoTokenizer

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

# If you want to load a model using multiple GPUs, please refer to the `Multiple GPUs` section.
path = 'OpenGVLab/InternVL2_5-38B'
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    use_flash_attn=True,
    trust_remote_code=True).eval().cuda()
tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True, use_fast=False)

# set the max number of tiles in `max_num`
pixel_values = load_image('./examples/image1.jpg', max_num=12).to(torch.bfloat16).cuda()
generation_config = dict(max_new_tokens=1024, do_sample=True)

# pure-text conversation (纯文本对话)
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}')

question = 'Can you tell me a story?'
response, history = model.chat(tokenizer, None, question, generation_config, history=history, return_history=True)
print(f'User: {question}\nAssistant: {response}')

# single-image single-round conversation (单图单轮对话)
question = '<image>\nPlease describe the image shortly.'
response = model.chat(tokenizer, pixel_values, question, generation_config)
print(f'User: {question}\nAssistant: {response}')

# single-image multi-round conversation (单图多轮对话)
question = '<image>\nPlease describe the image in detail.'
response, history = model.chat(tokenizer, pixel_values, question, generation_config, history=None, return_history=True)
print(f'User: {question}\nAssistant: {response}')

question = 'Please write a poem according to the image.'
response, history = model.chat(tokenizer, pixel_values, question, generation_config, history=history, return_history=True)
print(f'User: {question}\nAssistant: {response}')

# multi-image multi-round conversation, combined images (多图多轮对话，拼接图像)
pixel_values1 = load_image('./examples/image1.jpg', max_num=12).to(torch.bfloat16).cuda()
pixel_values2 = load_image('./examples/image2.jpg', max_num=12).to(torch.bfloat16).cuda()
pixel_values = torch.cat((pixel_values1, pixel_values2), dim=0)

question = '<image>\nDescribe the two images in detail.'
response, history = model.chat(tokenizer, pixel_values, question, generation_config,
                               history=None, return_history=True)
print(f'User: {question}\nAssistant: {response}')

question = 'What are the similarities and differences between these two images.'
response, history = model.chat(tokenizer, pixel_values, question, generation_config,
                               history=history, return_history=True)
print(f'User: {question}\nAssistant: {response}')

# multi-image multi-round conversation, separate images (多图多轮对话，独立图像)
pixel_values1 = load_image('./examples/image1.jpg', max_num=12).to(torch.bfloat16).cuda()
pixel_values2 = load_image('./examples/image2.jpg', max_num=12).to(torch.bfloat16).cuda()
pixel_values = torch.cat((pixel_values1, pixel_values2), dim=0)
num_patches_list = [pixel_values1.size(0), pixel_values2.size(0)]

question = 'Image-1: <image>\nImage-2: <image>\nDescribe the two images in detail.'
response, history = model.chat(tokenizer, pixel_values, question, generation_config,
                               num_patches_list=num_patches_list,
                               history=None, return_history=True)
print(f'User: {question}\nAssistant: {response}')

question = 'What are the similarities and differences between these two images.'
response, history = model.chat(tokenizer, pixel_values, question, generation_config,
                               num_patches_list=num_patches_list,
                               history=history, return_history=True)
print(f'User: {question}\nAssistant: {response}')

# batch inference, single image per sample (单图批处理)
pixel_values1 = load_image('./examples/image1.jpg', max_num=12).to(torch.bfloat16).cuda()
pixel_values2 = load_image('./examples/image2.jpg', max_num=12).to(torch.bfloat16).cuda()
num_patches_list = [pixel_values1.size(0), pixel_values2.size(0)]
pixel_values = torch.cat((pixel_values1, pixel_values2), dim=0)

questions = ['<image>\nDescribe the image in detail.'] * len(num_patches_list)
responses = model.batch_chat(tokenizer, pixel_values,
                             num_patches_list=num_patches_list,
                             questions=questions,
                             generation_config=generation_config)
for question, response in zip(questions, responses):
    print(f'User: {question}\nAssistant: {response}')

# video multi-round conversation (视频多轮对话)
def get_index(bound, fps, max_frame, first_idx=0, num_segments=32):
    if bound:
        start, end = bound[0], bound[1]
    else:
        start, end = -100000, 100000
    start_idx = max(first_idx, round(start * fps))
    end_idx = min(round(end * fps), max_frame)
    seg_size = float(end_idx - start_idx) / num_segments
    frame_indices = np.array([
        int(start_idx + (seg_size / 2) + np.round(seg_size * idx))
        for idx in range(num_segments)
    ])
    return frame_indices

def load_video(video_path, bound=None, input_size=448, max_num=1, num_segments=32):
    vr = VideoReader(video_path, ctx=cpu(0), num_threads=1)
    max_frame = len(vr) - 1
    fps = float(vr.get_avg_fps())

    pixel_values_list, num_patches_list = [], []
    transform = build_transform(input_size=input_size)
    frame_indices = get_index(bound, fps, max_frame, first_idx=0, num_segments=num_segments)
    for frame_index in frame_indices:
        img = Image.fromarray(vr[frame_index].asnumpy()).convert('RGB')
        img = dynamic_preprocess(img, image_size=input_size, use_thumbnail=True, max_num=max_num)
        pixel_values = [transform(tile) for tile in img]
        pixel_values = torch.stack(pixel_values)
        num_patches_list.append(pixel_values.shape[0])
        pixel_values_list.append(pixel_values)
    pixel_values = torch.cat(pixel_values_list)
    return pixel_values, num_patches_list

video_path = './examples/red-panda.mp4'
pixel_values, num_patches_list = load_video(video_path, num_segments=8, max_num=1)
pixel_values = pixel_values.to(torch.bfloat16).cuda()
video_prefix = ''.join([f'Frame{i+1}: <image>\n' for i in range(len(num_patches_list))])
question = video_prefix + 'What is the red panda doing?'
# Frame1: <image>\nFrame2: <image>\n...\nFrame8: <image>\n{question}
response, history = model.chat(tokenizer, pixel_values, question, generation_config,
                               num_patches_list=num_patches_list, history=None, return_history=True)
print(f'User: {question}\nAssistant: {response}')

question = 'Describe this video in detail. Don\'t repeat.'
response, history = model.chat(tokenizer, pixel_values, question, generation_config,
                               num_patches_list=num_patches_list, history=history, return_history=True)
print(f'User: {question}\nAssistant: {response}')
```

#### Streaming output

Besides this method, you can also use the following code to get streamed output.

```python
from transformers import TextIteratorStreamer
from threading import Thread

# Initialize the streamer
streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True, timeout=10)
# Define the generation configuration
generation_config = dict(max_new_tokens=1024, do_sample=False, streamer=streamer)
# Start the model chat in a separate thread
thread = Thread(target=model.chat, kwargs=dict(
    tokenizer=tokenizer, pixel_values=pixel_values, question=question,
    history=None, return_history=False, generation_config=generation_config,
))
thread.start()

# Initialize an empty string to store the generated text
generated_text = ''
# Loop through the streamer to get the new text as it is generated
for new_text in streamer:
    if new_text == model.conv_template.sep:
        break
    generated_text += new_text
    print(new_text, end='', flush=True)  # Print each new chunk of generated text on the same line
```

## Finetune

Many repositories now support fine-tuning of the InternVL series models, including [InternVL](https://github.com/OpenGVLab/InternVL), [SWIFT](https://github.com/modelscope/ms-swift), [XTurner](https://github.com/InternLM/xtuner), and others. Please refer to their documentation for more details on fine-tuning.

## Deployment

### LMDeploy

LMDeploy is a toolkit for compressing, deploying, and serving LLM, developed by the MMRazor and MMDeploy teams.

```sh
pip install lmdeploy>=0.5.3
```

LMDeploy abstracts the complex inference process of multi-modal Vision-Language Models (VLM) into an easy-to-use pipeline, similar to the Large Language Model (LLM) inference pipeline.

#### A 'Hello, world' example

```python
from lmdeploy import pipeline, TurbomindEngineConfig
from lmdeploy.vl import load_image

model = 'OpenGVLab/InternVL2_5-38B'
image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/tests/data/tiger.jpeg')
pipe = pipeline(model, backend_config=TurbomindEngineConfig(session_len=8192))
response = pipe(('describe this image', image))
print(response.text)
```

If `ImportError` occurs while executing this case, please install the required dependency packages as prompted.

#### Multi-images inference

When dealing with multiple images, you can put them all in one list. Keep in mind that multiple images will lead to a higher number of input tokens, and as a result, the size of the context window typically needs to be increased.

> Warning: Due to the scarcity of multi-image conversation data, the performance on multi-image tasks may be unstable, and it may require multiple attempts to achieve satisfactory results.

```python
from lmdeploy import pipeline, TurbomindEngineConfig
from lmdeploy.vl import load_image
from lmdeploy.vl.constants import IMAGE_TOKEN

model = 'OpenGVLab/InternVL2_5-38B'
pipe = pipeline(model, backend_config=TurbomindEngineConfig(session_len=8192))

image_urls=[
    'https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/demo/resources/human-pose.jpg',
    'https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/demo/resources/det.jpg'
]

images = [load_image(img_url) for img_url in image_urls]
# Numbering images improves multi-image conversations
response = pipe((f'Image-1: {IMAGE_TOKEN}\nImage-2: {IMAGE_TOKEN}\ndescribe these two images', images))
print(response.text)
```

#### Batch prompts inference

Conducting inference with batch prompts is quite straightforward; just place them within a list structure:

```python
from lmdeploy import pipeline, TurbomindEngineConfig
from lmdeploy.vl import load_image

model = 'OpenGVLab/InternVL2_5-38B'
pipe = pipeline(model, backend_config=TurbomindEngineConfig(session_len=8192))

image_urls=[
    "https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/demo/resources/human-pose.jpg",
    "https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/demo/resources/det.jpg"
]
prompts = [('describe this image', load_image(img_url)) for img_url in image_urls]
response = pipe(prompts)
print(response)
```

#### Multi-turn conversation

There are two ways to do the multi-turn conversations with the pipeline. One is to construct messages according to the format of OpenAI and use above introduced method, the other is to use the `pipeline.chat` interface.

```python
from lmdeploy import pipeline, TurbomindEngineConfig, GenerationConfig
from lmdeploy.vl import load_image

model = 'OpenGVLab/InternVL2_5-38B'
pipe = pipeline(model, backend_config=TurbomindEngineConfig(session_len=8192))

image = load_image('https://raw.githubusercontent.com/open-mmlab/mmdeploy/main/demo/resources/human-pose.jpg')
gen_config = GenerationConfig(top_k=40, top_p=0.8, temperature=0.8)
sess = pipe.chat(('describe this image', image), gen_config=gen_config)
print(sess.response.text)
sess = pipe.chat('What is the woman doing?', session=sess, gen_config=gen_config)
print(sess.response.text)
```

#### Service

LMDeploy's `api_server` enables models to be easily packed into services with a single command. The provided RESTful APIs are compatible with OpenAI's interfaces. Below are an example of service startup:

```shell
lmdeploy serve api_server OpenGVLab/InternVL2_5-38B --backend turbomind --server-port 23333
```

To use the OpenAI-style interface, you need to install OpenAI:

```shell
pip install openai
```

Then, use the code below to make the API call:

```python
from openai import OpenAI

client = OpenAI(api_key='YOUR_API_KEY', base_url='http://0.0.0.0:23333/v1')
model_name = client.models.list().data[0].id
response = client.chat.completions.create(
    model=model_name,
    messages=[{
        'role':
        'user',
        'content': [{
            'type': 'text',
            'text': 'describe this image',
        }, {
            'type': 'image_url',
            'image_url': {
                'url':
                'https://modelscope.oss-cn-beijing.aliyuncs.com/resource/tiger.jpeg',
            },
        }],
    }],
    temperature=0.8,
    top_p=0.8)
print(response)
```

## License

This project is released under the MIT license, while Qwen2 is licensed under the Tongyi Qianwen LICENSE.

## Citation

If you find this project useful in your research, please consider citing:

```BibTeX
@article{chen2023internvl,
  title={InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks},
  author={Chen, Zhe and Wu, Jiannan and Wang, Wenhai and Su, Weijie and Chen, Guo and Xing, Sen and Zhong, Muyan and Zhang, Qinglong and Zhu, Xizhou and Lu, Lewei and Li, Bin and Luo, Ping and Lu, Tong and Qiao, Yu and Dai, Jifeng},
  journal={arXiv preprint arXiv:2312.14238},
  year={2023}
}
@article{chen2024far,
  title={How Far Are We to GPT-4V? Closing the Gap to Commercial Multimodal Models with Open-Source Suites},
  author={Chen, Zhe and Wang, Weiyun and Tian, Hao and Ye, Shenglong and Gao, Zhangwei and Cui, Erfei and Tong, Wenwen and Hu, Kongzhi and Luo, Jiapeng and Ma, Zheng and others},
  journal={arXiv preprint arXiv:2404.16821},
  year={2024}
}
```