Instructions to use chibifire/EditScore-7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use chibifire/EditScore-7B with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("/share/project/luoxin/huggingface/hub/models--Qwen--Qwen2.5-VL-7B-Instruct/snapshots/cc594898137f460bfe9f0759e9844b3ce807cfb5")
model = PeftModel.from_pretrained(base_model, "chibifire/EditScore-7B")

Transformers

How to use chibifire/EditScore-7B with Transformers:

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

pipe = pipeline("text-generation", model="chibifire/EditScore-7B")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("chibifire/EditScore-7B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use chibifire/EditScore-7B with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/chibifire/EditScore-7B

SGLang

How to use chibifire/EditScore-7B 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 "chibifire/EditScore-7B" \
    --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": "chibifire/EditScore-7B",
		"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 "chibifire/EditScore-7B" \
        --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": "chibifire/EditScore-7B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use chibifire/EditScore-7B with Docker Model Runner:
```
docker model run hf.co/chibifire/EditScore-7B
```

EditScore-7B

Mirror of EditScore-7B, a LoRA adapter on Qwen2.5-VL-7B-Instruct that scores instruction-guided image edits.

What it is for

The repo holds the PEFT LoRA adapter (adapter_model.safetensors, adapter_config.json) and the card images from upstream. EditScore is a reward model for instruction-guided image editing. It takes an input image, an edited output image and the edit instruction, and returns a quality score. Use it to rerank candidate edits (best-of-N selection) or as a reward signal for reinforcement learning on image editing models.

Provenance

Upstream model: EditScore/EditScore-7B, part of the EditScore collection.
Base model: Qwen/Qwen2.5-VL-7B-Instruct.
Code: VectorSpaceLab/EditScore. Paper: arXiv 2509.23909. Project page: EditScore. Benchmark: EditReward-Bench.

Use

from PIL import Image
from editscore import EditScore

scorer = EditScore(
    backbone="qwen25vl",
    model_name_or_path="Qwen/Qwen2.5-VL-7B-Instruct",
    enable_lora=True,
    lora_path="chibifire/EditScore-7B",
    score_range=25,
    num_pass=1,
)
result = scorer.evaluate([Image.open("input.png"), Image.open("output.png")],
                         "Adjust the background to a glass wall.")
print(result["final_score"])

The editscore package comes from the upstream code repo. See upstream's card for setup and full usage.

Licence

Apache 2.0, the licence of the EditScore code repo. Terms: LICENSE.

Downloads last month: 24

Model tree for chibifire/EditScore-7B

Base model

Qwen/Qwen2.5-VL-7B-Instruct

Adapter

(367)

this model

Paper for chibifire/EditScore-7B

EditScore: Unlocking Online RL for Image Editing via High-Fidelity Reward Modeling

Paper • 2509.23909 • Published Sep 28, 2025 • 34