Instructions to use ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128 with Transformers:

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

pipe = pipeline("image-text-to-text", model="ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128")

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128

SGLang

How to use ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128 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 "ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128 with Docker Model Runner:
```
docker model run hf.co/ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128
```

LLaVA-1.5-7B-GRACE-W4G128

This repository contains a GRACE-trained LLaVA-1.5-7B checkpoint with quantization-aware training (QAT), W4G128 group-wise INT4 quantization, and learned scales.

This model is associated with our ICML 2026 paper:

Gated Relational Alignment via Confidence-based Distillation for Efficient VLMs
Yanlong Chen, Amirhossein Habibian, Luca Benini, Yawei Li
Accepted to the International Conference on Machine Learning (ICML 2026)

Paper: https://arxiv.org/abs/2601.22709
DOI: https://doi.org/10.48550/arXiv.2601.22709
Code: https://github.com/ForeverBlue816/GRACE

Model Details

Base model: liuhaotian/llava-v1.5-7b
Model family: LLaVA-1.5
Method: GRACE
Quantization: W4G128 group-wise INT4 QAT
Scale setting: learned scales
Repository: ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128

Files

This repository includes the merged model checkpoint and GRACE-specific quantized weights:

model-00001-of-00003.safetensors
model-00002-of-00003.safetensors
model-00003-of-00003.safetensors
model.safetensors.index.json
qat_quantized_weights.bin
config.json
generation_config.json
tokenizer.model
tokenizer_config.json
special_tokens_map.json

Intended Use

This model is intended for research on efficient vision-language models, quantization-aware training, and multimodal model compression.

Potential use cases include:

evaluating INT4 quantized vision-language models;
studying quantization-aware training for multimodal models;
reproducing or extending GRACE-style efficient VLM experiments;
comparing quantized LLaVA-1.5 variants under multimodal benchmarks.

Loading

Please use a LLaVA-compatible inference environment together with the GRACE quantization-aware loading code.

from transformers import AutoTokenizer, AutoModelForCausalLM

repo_id = "ForeverBlue/LLaVA-1.5-7B-GRACE-W4G128"

tokenizer = AutoTokenizer.from_pretrained(repo_id, use_fast=False)
model = AutoModelForCausalLM.from_pretrained(
    repo_id,
    trust_remote_code=True,
    device_map="auto"
)

Important Notes

This checkpoint includes QAT-specific quantized weights in qat_quantized_weights.bin. Depending on the inference codebase, additional GRACE-specific quantization-aware loading logic may be required.

The standard from_pretrained call may load the model configuration and checkpoint files, but fully reproducing the intended INT4 QAT behavior may require the GRACE repository:

https://github.com/ForeverBlue816/GRACE

Limitations

This model is released for research purposes.
The quantized checkpoint may require custom loading logic for QAT-specific weights.
Performance may vary depending on the evaluation codebase, preprocessing, generation parameters, and multimodal benchmark implementation.
Users should follow the license and usage restrictions of the original LLaVA-1.5 base model.

Citation

If you use this model, please cite:

@article{chen2026gated,
  title={Gated Relational Alignment via Confidence-based Distillation for Efficient VLMs},
  author={Chen, Yanlong and Habibian, Amirhossein and Benini, Luca and Li, Yawei},
  journal={arXiv preprint arXiv:2601.22709},
  year={2026}
}