InternVideo2-Stage2_6B by OpenGVLab: Open-Weight Model

Runs On

What it takes to serve InternVideo2-Stage2_6B (6.4B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

Precision	Weights	Memory needed	Cheapest setup	Per hour	Also fits
16-bit	12.7 GB	15.3 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
8-bit	6.4 GB	7.6 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
4-bit	3.2 GB	3.8 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.

Model Card

By OpenGVLab, published under mit, revision fd8ef758daf1.

This repository contains the 6B model of the paper InternVideo2 in stage 2. Code: https://github.com/OpenGVLab/InternVideo/tree/main/InternVideo2/multimodality Please refer to https://github.com/OpenGVLab/InternVideo/blob/main/InternVideo2/multimodality/INSTALL.md

Read OpenGVLab's full model card

Introduction

This repository contains the 6B model of the paper InternVideo2 in stage 2.

Code: https://github.com/OpenGVLab/InternVideo/tree/main/InternVideo2/multi_modality

Installation

Please refer to https://github.com/OpenGVLab/InternVideo/blob/main/InternVideo2/multi_modality/INSTALL.md

Usage

import cv2
from transformers import AutoModel
from modeling_internvideo2 import (retrieve_text, vid2tensor, _frame_from_video,)


model = AutoModel.from_pretrained("OpenGVLab/InternVideo2-Stage2_6B", trust_remote_code=True).eval()

video = cv2.VideoCapture('example1.mp4')
frames = [x for x in _frame_from_video(video)]
text_candidates = ["A playful dog and its owner wrestle in the snowy yard, chasing each other with joyous abandon.",
                "A man in a gray coat walks through the snowy landscape, pulling a sleigh loaded with toys.",
                "A person dressed in a blue jacket shovels the snow-covered pavement outside their house.",
                "A cat excitedly runs through the yard, chasing a rabbit.",
                "A person bundled up in a blanket walks through the snowy landscape, enjoying the serene winter scenery."]

texts, probs = retrieve_text(frames, text_candidates, model=model, topk=5)
for t, p in zip(texts, probs):
    print(f'text: {t} ~ prob: {p:.4f}')

vidtensor = vid2tensor('example1.mp4', fnum=4)
feat = model.get_vid_feat(vidtensor)

Configuration

Architecture: InternVideo2_Stage2
Stored precision: float32

Identity and Version

Repository: OpenGVLab/InternVideo2-Stage2_6B
Publisher: OpenGVLab
Task: Video classification
Modality: Video
Library: Not stated by the source
Parameters: 6.4B parameters
Languages: Not stated by the source
Revision: fd8ef758daf132d4738325d39546dd76ce26911d
First published: 2025-02-10
Last updated: 2025-02-27

Files and Weights

40 files, 25.5 GB in total. The weights are 13 files totalling 25.5 GB in safetensors.

Weights13 files · 25.5 GB

Configuration23 files · 272.2 KB

Documentation1 file · 1.5 KB

Other2 files · 704.0 KB

Repository1 file · 1.6 KB

Every file

File	Type	Size	SHA-256
model-00001-of-00013.safetensors	Weights	1.8 GB	9d177c77c672
model-00002-of-00013.safetensors	Weights	2.0 GB	98753c824d62
model-00003-of-00013.safetensors	Weights	2.0 GB	bc1ad5944716
model-00004-of-00013.safetensors	Weights	2.0 GB	bd4479e81e91
model-00005-of-00013.safetensors	Weights	2.0 GB	f16ff3ed45cc
model-00006-of-00013.safetensors	Weights	2.0 GB	5b4fadad100d
model-00007-of-00013.safetensors	Weights	2.0 GB	0d32dbcb9d7c
model-00008-of-00013.safetensors	Weights	2.0 GB	8f990c3a412c
model-00009-of-00013.safetensors	Weights	2.0 GB	e29811441d6a
model-00010-of-00013.safetensors	Weights	2.0 GB	57e01270ed6a
model-00011-of-00013.safetensors	Weights	2.0 GB	8d4ef0dede04
model-00012-of-00013.safetensors	Weights	2.0 GB	29a0162c3608
model-00013-of-00013.safetensors	Weights	2.0 GB	6328c13ba3af
config.json	Configuration	11.8 KB	—
configs/beit-base-patch16-224-pt22k-ft22k.json	Configuration	914 B	—
configs/config_bert_large.json	Configuration	601 B	—
configs/med_config.json	Configuration	529 B	—
configs/med_large_config.json	Configuration	530 B	—
configs/model.py	Configuration	3.2 KB	—
configs/pretrain.py	Configuration	2.9 KB	—
configs/qa.py	Configuration	346 B	—
configs/qa_anet.py	Configuration	658 B	—
configs/qa_msrvtt.py	Configuration	651 B	—
configs/ret_anet.py	Configuration	542 B	—
configs/ret_coco.py	Configuration	701 B	—
configs/ret_didemo.py	Configuration	757 B	—
configs/ret_flickr.py	Configuration	716 B	—
configs/ret_msrvtt.py	Configuration	629 B	—
configs/ret_msrvtt_9k.py	Configuration	148 B	—
configs/ret_msrvtt_mc.py	Configuration	607 B	—
configs/ret_ssv2_label.py	Configuration	475 B	—
configs/ret_ssv2_template.py	Configuration	481 B	—
configs/tvqa.py	Configuration	789 B	—
demo.py	Configuration	1.2 KB	—
model.safetensors.index.json	Configuration	102.2 KB	—
modeling_internvideo2.py	Configuration	140.8 KB	—
README.md	Documentation	1.5 KB	—
__pycache__/modeling_internvideo2.cpython-310.pyc	Other	95.7 KB	—
example1.mp4	Other	608.3 KB	8da3fb45df69
.gitattributes	Repository	1.6 KB	—

License and Download

License: mit
Access: Open weights, no gate
Download size: 25.5 GB

Download from OpenGVLab

Released by OpenGVLab through its official repository on Hugging Face. Read the license.

Built From

Described by arXiv:2403.15377

Memory Requirements

Precision	Weights in memory
As published	25.5 GB
16-bit	12.7 GB
8-bit	6.4 GB
4-bit	3.2 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About InternVideo2-Stage2_6B

How much GPU memory does InternVideo2-Stage2_6B need?

About 15.3 GB at 16-bit and 3.8 GB at 4-bit: the weights (6.4B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run InternVideo2-Stage2_6B on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use InternVideo2-Stage2_6B commercially?

Yes. InternVideo2-Stage2_6B is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.

Similar Models

Model · Video classification

Cosmos-Embed1-448p-anomaly-detection

NVIDIA

Cosmos-Embed1 is a joint video-text embedder tailored for physical AI. It can be used for text-to-video retrieval, inverse video search, semantic deduplication, zero-shot and k-nearest-neighbors (kNN) classification, and as a base model for video curation tasks. It has state-of-the-art (SOTA) performance on autonomous vehicle (AV) and robotics datasets, while maintaining competitive performance in general domains. A fine-tuned variant is also provided for video anomaly detection and classification. This model is ready for commercial use. The Cosmos-Embed1 release includes the following embedders: Note: while each checkpoint was optimized at a specific fixed resolution (and default to…

Open weights other 1.2B parameters cosmos

View model

Model · Video classification

vjepa2-vitg-fpc64-384-ssv2

AI at Meta

A frontier video understanding model developed by FAIR, Meta, which extends the pretraining objectives of VJEPA, resulting in state-of-the-art video understanding capabilities, leveraging data and model sizes at scale. The code is released in this repository. This is V-JEPA 2 ViT-g 384 model with video classification head pretrained on Something-Something-V2 dataset. To run V-JEPA 2 model, ensure you have installed the latest transformers

Open weights mit 1.1B parameters transformers

View model

Model · Video classification

vjepa2-vitg-fpc32-384-diving48

AI at Meta

A frontier video understanding model developed by FAIR, Meta, which extends the pretraining objectives of VJEPA, resulting in state-of-the-art video understanding capabilities, leveraging data and model sizes at scale. The code is released in this repository. This is V-JEPA 2 ViT-g 384 model with video classification head pretrained on Diving 48 dataset. To run V-JEPA 2 model, ensure you have installed the latest transformers

Open weights mit 1.1B parameters transformers

View model

Model · Video classification

vjepa2.1-vit-giant-384

Antonio Apicella

A HuggingFace-format conversion of Meta AI's V-JEPA 2.1 ViT-g/16 video encoder and predictor, operating at 384x384 resolution. The weights are Meta's, copied without modification. This repository provides the transformers-compatible packaging plus a documented numerical validation against the original implementation. No prior HuggingFace port of this variant existed at the time of upload. The only structural change is that the fused QKV projection of each attention block is split into separate query / key / value matrices, following the convention used by transformers. This is a re-parameterization, not a change of weights. It is also convenient downstream: PEFT adapters apply to…

Open weights mit 1.1B parameters transformers

View model

Model · Video classification

vjepa2-vitg-fpc64-256

AI at Meta

A frontier video understanding model developed by FAIR, Meta, which extends the pretraining objectives of VJEPA, resulting in state-of-the-art video understanding capabilities, leveraging data and model sizes at scale. The code is released in this repository. To run V-JEPA 2 model, ensure you have installed the latest transformers: V-JEPA 2 is intended to represent any video (and image) to perform video classification, retrieval, or as a video encoder for VLMs. To load a video, sample the number of frames according to the model. For this model, we use 64. To load an image, simply copy the image to the desired number of frames. For more code examples, please refer to the V-JEPA 2…

Open weights apache-2.0 1B parameters transformers

View model

Model · Video classification

vjepa2-vitg-fpc64-384

AI at Meta

A frontier video understanding model developed by FAIR, Meta, which extends the pretraining objectives of VJEPA, resulting in state-of-the-art video understanding capabilities, leveraging data and model sizes at scale. The code is released in this repository. To run V-JEPA 2 model, ensure you have installed the latest transformers: V-JEPA 2 is intended to represent any video (and image) to perform video classification, retrieval, or as a video encoder for VLMs. To load a video, sample the number of frames according to the model. For this model, we use 64. To load an image, simply copy the image to the desired number of frames. For more code examples, please refer to the V-JEPA 2…

Open weights apache-2.0 1B parameters transformers

View model