Instructions to use HoshiuZ/LLaVA-NeXT-8B-Unlearned with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use HoshiuZ/LLaVA-NeXT-8B-Unlearned with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="HoshiuZ/LLaVA-NeXT-8B-Unlearned")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("HoshiuZ/LLaVA-NeXT-8B-Unlearned")
model = AutoModelForMultimodalLM.from_pretrained("HoshiuZ/LLaVA-NeXT-8B-Unlearned", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use HoshiuZ/LLaVA-NeXT-8B-Unlearned with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "HoshiuZ/LLaVA-NeXT-8B-Unlearned"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "HoshiuZ/LLaVA-NeXT-8B-Unlearned",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/HoshiuZ/LLaVA-NeXT-8B-Unlearned

SGLang

How to use HoshiuZ/LLaVA-NeXT-8B-Unlearned with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "HoshiuZ/LLaVA-NeXT-8B-Unlearned" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "HoshiuZ/LLaVA-NeXT-8B-Unlearned",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "HoshiuZ/LLaVA-NeXT-8B-Unlearned" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "HoshiuZ/LLaVA-NeXT-8B-Unlearned",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use HoshiuZ/LLaVA-NeXT-8B-Unlearned with Docker Model Runner:
```
docker model run hf.co/HoshiuZ/LLaVA-NeXT-8B-Unlearned
```

LLaVA-NeXT-8B-Unlearned — 2026-11-01 base-reference version

This is a complete BF16 LLaVA-NeXT wrapper reconstructed from the pinned full multimodal base and the 2026-11-01 sparse overlay. The public checkpoint includes the language model, vision tower, projector, tokenizer, processor, configuration, and generation settings.

Source and version

Full multimodal base: llava-hf/llama3-llava-next-8b-hf
Frozen base revision: b041c0d0ea0dd0196d147206c210c8d1752fc2da
Wrapper class: LlavaNextForConditionalGeneration
New sparse overlay SHA-256: dd9b6bb0c01c2a40437c89c127540e83c7bd6554b1892984e2273ed946ddd1da (822,087,200 bytes)
Updated language-model MLP down_proj matrices: layers 4, 5, 6, 7, 8, 10, and 12.
Overlay and base were combined into a complete checkpoint; applying the overlay at load time is not required.

The previous public repository state remains available at revision 1c1b8d10c7ee3d4b364c1cd0cdd0316605eb5aae and tag hf1-2026-10-31. That version used a different overlay (627e161056ed640837eb007debe17ea225c23c781e5d154971b38a53822592f2). The current main card describes only the 2026-11-01 base-reference version.

Text-only experiment record

The source experiment records WMDP three-domain C = 80.4348%, other245 131/245, and MMLU285 145/285. These are experiment-record values and were not rescored during this publication preparation. An item-level score audit was not independently repeated here. S has not been measured for this version.

Evaluation used text-only inputs. Image inputs and broader multimodal safety were not evaluated for this version.

Loading

import torch
from transformers import AutoProcessor, LlavaNextForConditionalGeneration

repo = "HoshiuZ/LLaVA-NeXT-8B-Unlearned"
processor = AutoProcessor.from_pretrained(repo)
model = LlavaNextForConditionalGeneration.from_pretrained(
    repo, torch_dtype=torch.bfloat16, device_map="auto"
)

License and attribution

The selected source model is marked llama3. The Meta Llama 3 Community License Agreement and Acceptable Use Policy apply. This repository does not relicense the model as Apache-2.0 or MIT. See LICENSE, USE_POLICY.md, and Meta's Llama 3 license page.

Research source: HoshiuZ/multimodal-llm-safety. This model card documents the checkpoint version and text-only experiment record; it is not a claim that image or broader multimodal safety has been tested.

Downloads last month: 41

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for HoshiuZ/LLaVA-NeXT-8B-Unlearned

Base model

llava-hf/llama3-llava-next-8b-hf

Finetuned

(5)

this model