LLaVA-OneVision-2.5-4B-Instruct

GitHub Project page: coming soon Technical report: coming soon License

English | ไธญๆ–‡

LLaVA-OneVision-2.5-4B-Instruct is a 4B vision-language model for single-image, multi-image and video input. Beyond the usual offline tasks, it is trained for streaming video: frames arrive one at a time, and at every step the model decides whether to stay silent or speak. The same weights and the same chat template serve both modes.

Highlights

  • Offline and streaming in one model. apply_chat_template(..., streaming=True) switches mode; eight protocol tokens carry the per-frame decision to stay silent, respond or caption. Nothing else about the input changes.
  • Codec-aligned video. Besides uniform frame sampling, codec mode uses the encoder's bit-cost to pack the informative regions of a video into a handful of canvases.
  • 9ร— spatial compression. A patch-16 vision tower with 3ร—3 merging gives one visual token per 48ร—48 pixels, so long videos fit comfortably in the 262k context.

Performance

Video

Benchmark OV2.5-4B (codec) OV2.5-4B (frames) Qwen3.5-4B
General video understanding
VideoMME (no subtitles)71.969.771.7
VideoMME (subtitles)74.775.275.9
MotionBench64.2 TC128ร—313,60062.4 128fร—153,66460.1 128fร—153,664
TOMATO36.0 TC16ร—153,66439.2 16fร—153,66432.4 16fร—153,664
MVBench75.5 TC64ร—313,60072.0 64fร—153,66464.0 64fร—153,664
NextQA MC test86.4 TC64ร—313,60086.1 64fร—153,66480.7 64fร—153,664
TempCompass MC79.7 TC64ร—313,60075.8 64fร—153,66473.4 64fร—153,664
Video-MME-v217.2 TC64ร—313,60016.4 64fร—330,00015.5 64fร—330,000
EgoSchema subset79.8 TC64ร—313,60080.0 64fร—153,66468.2 64fร—153,664
Video-Index23.1 TC512ร—55,29620.0 1fps, โ‰ค512fร—55,29619.6 1fps, โ‰ค512fร—55,296
Long video
LVBench61.6 TC1024ร—156,80062.0 950fร—200,00056.5 2048fร—120,000
MLVU dev77.1 TC512ร—313,60075.3 512fร—72,90073.8 512fร—72,900
LongVideoBench val71.1 TC768ร—235,20069.0 1280fร—80,00062.7 700fร—300,000
VideoEval-Pro64.6 TC768ร—235,20064.4 950fร—200,00062.8 2048fร—100,000
Temporal grounding (mIoU)
TimeLens ActivityNet59.1 TC128ร—313,60058.8 2048fร—409,600, fps=453.8 2048fร—409,600, fps=4
TimeLens Charades61.3 TC128ร—313,60060.6 2048fร—409,600, fps=448.1 2048fร—409,600, fps=4
TimeLens QVHighlights70.3 TC128ร—313,60068.4 2048fร—409,600, fps=467.5 2048fร—409,600, fps=4

Streaming video

Benchmark OV2.5-4B
(codec, 24 fps)
MOSS-VL-Realtime 11B
(1 fps)
JoyAI-VL-Interaction 8B
(1 fps)
OVO-Bench
Real-Time Visual Perception (RTVP)81.075.976.4
Backward Tracing (BT)68.872.653.3
Forward Active Responding (FAR)60.762.154.6
Overall70.270.261.4
StreamingBench
Real-Time Visual Understanding (RTVU)85.382.983.2
Misleading Context Understanding (MCU)36.8--37.2
Anomaly Context Understanding (ACU)67.6--51.6
Sequential Question Answering (SQA)54.850.453.6
Proactive Output (PO)63.660.039.6
Contextual Understanding (CU)55.756.445.5
Proactive and interactive
OmniMMI: Proactive Alerting (PA)68.566.013.5
ProactiveVideoQA: PAUC (ฯ‰=0.5)33.147.237.0
InteractionBench: Total38.6----
StreamArena: Proactive Interaction (Pro)25.511.89.4

Image

General perception and reading OV2.5-4B Qwen3.5-4B Grounding and spatial OV2.5-4B Qwen3.5-4B
MMStar63.966.9RefCOCO val (Acc@0.5)92.689.1
MMBench-EN dev88.288.8RefCOCOg val (Acc@0.5)88.987.7
RealWorldQA73.974.4RefSpatial-Bench62.129.1
AI2D82.282.4CrossPoint80.533.7
BLINK67.163.6CV-Bench 2D83.380.7
V*-Bench89.581.7CV-Bench 3D94.592.0
OCRBench79.885.2EmbSpatial83.074.7
DocVQA val92.995.2MMSI-Bench33.331.4

Compared with Qwen3.5-4B, thinking disabled. Bold is the best score in each row; -- means not available. The superscript on each video score is the setting that produced it: frames ร— pixels per frame, or for codec, canvases (TC) ร— pixels per canvas. VideoMME is the mean over the short, medium and long splits. Each side reports its best score over the settings tried, from single runs. The image table shows selected benchmarks; full results, settings and protocol will be given in the technical report.

Model Overview

Typeinstruction-tuned vision-language model, LLaVA-OneVision-2.5 architecture
Parameters5.25B total (4B language model plus vision tower and adapter)
Language side32 layers, hidden size 2560, FFN 9216, 16 attention heads / 4 KV heads, linear attention with one full-attention layer every 4
Vision side24 layers, hidden size 1024, patch_size=16, spatial_merge_size=3, attention within a frame
Vocabulary248320, including the 8 streaming protocol tokens
Context length262144
Weightsbfloat16, 3 safetensors shards, about 9.9 GB

Quickstart

The modeling code ships with the weights, so trust_remote_code=True is required.

pip install "transformers>=5.7.0" accelerate torch torchvision
pip install flash-attn --no-build-isolation   # optional, recommended
pip install decord                            # frame-sampling video path only

Image

import torch
from PIL import Image
from transformers import AutoProcessor, AutoModelForImageTextToText

MODEL = "lmms-lab-encoder/LLaVA-OneVision-2.5-4B-Instruct"

processor = AutoProcessor.from_pretrained(MODEL, trust_remote_code=True)
model = AutoModelForImageTextToText.from_pretrained(
    MODEL, trust_remote_code=True, dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",   # "sdpa" without flash-attn
    device_map="auto",
).eval()

messages = [{"role": "user", "content": [
    {"type": "image"},
    {"type": "text", "text": "Describe this image in detail."},
]}]
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=text, images=[Image.open("example.jpg").convert("RGB")],
                   return_tensors="pt").to(model.device)

with torch.inference_mode():
    out = model.generate(**inputs, max_new_tokens=512, do_sample=False)
print(processor.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Video

messages = [{"role": "user", "content": [
    {"type": "video"},
    {"type": "text", "text": "What is happening in this video?"},
]}]
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

inputs = processor(
    text=text, videos=["example.mp4"],
    max_frames=64,
    min_pixels=153664, max_pixels=153664,   # set equal to fix the per-frame budget
    return_tensors="pt",
).to(model.device)

For codec mode, install codec-video-prep and pass video_backend="codec":

pip install -i https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple \
            "codec-video-prep==0.3.0"
codec-video-prep-doctor
inputs = processor(text=text, videos=["example.mp4"],
                   video_backend="codec", return_tensors="pt").to(model.device)

Streaming

The checkpoint already supports streaming: pass streaming=True to apply_chat_template. The runtime library for real-time online inference is coming soon.

License

Apache-2.0.

Downloads last month
75
Safetensors
Model size
5B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support