OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Internvl

Internvl is an open-source family of multimodal vision-language models and developer resources for processing visual and text inputs. Its repository includes Python examples, documentation links, a chat demo and API documentation.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Internvl — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use the chat demo to explore multimodal conversation, or the repository and linked documentation when building a Python project with Internvl models.

Who it is for

  • Developers prototyping visual question-answering and image-chat applications.
  • Researchers and engineers exploring image-text retrieval or video conversation workflows.

Documented uses

  • Ask questions about an image and generate a conversational response.
  • Extract visual features from images.
  • Retrieve or compare images and text across modalities.
  • Prototype multi-round conversations using video frames.
Before you choose

Requirements & limits

  • The repository is MIT-licensed and identifies Python as its primary language.
  • Quick-start examples use PyTorch, PIL, Transformers and CUDA, so developers should check the linked model documentation for setup requirements.
  • The README links to distinct resources for documentation, a chat demo and API documentation.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install the InternVL repository dependencies

Set up the repository’s Python environment and install its default dependencies.

Before you start

  • Conda is available.
  • Python 3.9 is supported by the documented environment setup.

Steps

  1. Clone the repository with the command below.

    Documented command
    git clone https://github.com/OpenGVLab/InternVL.git

  2. Create and activate the Conda environment: conda create -n internvl python=3.9 followed by conda activate internvl.

  3. Install the default dependencies with the command below.

    Documented command
    pip install -r requirements.txt

  4. For zero-shot classification or retrieval evaluation, install the optional CLIP benchmark dependencies with the command below.

    Documented command
    pip install -r requirements/clip_benchmark.txt

Expected outcome

The repository is cloned and its default dependencies are installed in the `internvl` Conda environment; the optional benchmark dependencies can be installed separately.

Guide limitations

  • The default requirements do not include clip_benchmark.txt; install it only if that functionality is needed.
  • FlashAttention is optional and listed for chat-model training; mmcv-full and Apex are optional for segmentation.
  • The documentation notes that an Apex CUDA extension failure can result in fused_layer_norm_cuda being unavailable; uninstalling Apex lets the code use the PyTorch RMSNorm implementation.
Documentation 1 ↗
Source-based practical guide

Run InternViT-6B visual feature extraction with Hugging Face

Load the documented InternViT-6B model and pass it an image to obtain model outputs.

Before you start

  • The InternVL Python dependencies are installed.
  • A CUDA device is available, as the example moves the model and image tensors to CUDA.
  • The example image ./examples/image1.jpg is available.

Steps

  1. Load OpenGVLab/InternViT-6B-448px-V2_5 using AutoModel.from_pretrained with torch_dtype=torch.bfloat16, low_cpu_mem_usage=True, and trust_remote_code=True, then move it to CUDA and set evaluation mode.

  2. Open ./examples/image1.jpg as an RGB image.

  3. Load CLIPImageProcessor from OpenGVLab/InternViT-6B-448px-V1-5, preprocess the image into tensors, and move pixel_values to CUDA as bfloat16.

  4. Pass pixel_values to the model and collect the returned outputs.

Expected outcome

The example produces model outputs for the preprocessed image using InternViT-6B.

Guide limitations

  • The example uses CUDA and bfloat16; no hardware sizing or alternative device procedure is specified.
  • The documented snippet does not describe post-processing or interpretation of the returned outputs.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗Guide documentation ↗
Source record details
Source availability
Active
Sources checked
13 October 2026
Original website title
OpenGVLab/InternVL · Official homepage title · retained in its source language
How we review sourcesSuggest a correction