update readme

Author: Luciennnnnnn

README.md

@@ -26,6 +26,7 @@
 - **Versatile Applications**: Ready to use as a best-in-class reranker to improve editing outputs, or as a high-fidelity reward signal for **stable and effective Reinforcement Learning (RL) fine-tuning**.
 
 ## 🔥 News
+- **2025-09-30**: LoRA weights for OmniGen2, trained via Reinforcement Learning with EditScore-7B (Avg8) as the reward model, have been released on [Hugging Face]() and [ModelScope]().
 - **2025-09-30**: We are excited to release **EditScore** and **EditReward-Bench**! Model weights and the benchmark dataset are now publicly available. You can access them on Hugging Face: [Models Collection](https://huggingface.co/collections/EditScore/editscore-68d8e27ee676981221db3cfe) and [Benchmark Dataset](https://huggingface.co/datasets/EditScore/EditReward-Bench), and on ModelScope: [Models Collection](https://www.modelscope.cn/collections/EditScore-8b0d53aa945d4e) and [Benchmark Dataset](https://www.modelscope.cn/datasets/EditScore/EditReward-Bench).
 
 ## 📖 Introduction

editscore/__init__.py

@@ -129,6 +129,8 @@ def evaluate(self, image_prompts, text_prompt):
                     "consistency": np.mean([output_per_pass["consistency"] for output_per_pass in outputs_multi_pass]),
                     "perceptual_quality": np.mean([output_per_pass["perceptual_quality"] for output_per_pass in outputs_multi_pass]),
                     "overall": np.mean([output_per_pass["overall"] for output_per_pass in outputs_multi_pass]),
+                    "SC_reasoning": SC_dict["reasoning"],
+                    "PQ_reasoning": PQ_dict["reasoning"],
                 }
         if self.reduction == "average_first":
             output["overall"] = math.sqrt(output["prompt_following"] * output["perceptual_quality"])

editscore/mllm_tools/qwen25vl.py

@@ -43,7 +43,7 @@ def __init__(
         self.lora_path = lora_path
 
         self.model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-            vlm_model, dtype=torch.bfloat16, device_map="auto"
+            vlm_model, torch_dtype=torch.bfloat16, device_map="auto"
         )
         if enable_lora:
             self.model = PeftModel.from_pretrained(self.model, lora_path)

evaluate.sh

@@ -2,6 +2,9 @@
 SHELL_FOLDER=$(cd "$(dirname "$0")";pwd)
 cd $SHELL_FOLDER
 
+source "$(dirname $(which conda))/../etc/profile.d/conda.sh"
+conda activate py3.12+pytorch2.7.1+cu126
+
 python evaluation.py \
 --benchmark_dir EditScore/EditReward-Bench \
 --result_dir results/EditScore-7B \

evaluate_vllm.sh

@@ -2,6 +2,9 @@
 SHELL_FOLDER=$(cd "$(dirname "$0")";pwd)
 cd $SHELL_FOLDER
 
+source "$(dirname $(which conda))/../etc/profile.d/conda.sh"
+conda activate py3.12+pytorch2.7.1+cu126
+
 python evaluation.py \
 --benchmark_dir EditScore/EditReward-Bench \
 --result_dir results/EditScore-7B \

evaluation.py

@@ -228,7 +228,7 @@ def main(args):
     print("Writing results...", flush=True)
 
     start_time = time.time()
-    dataset = dataset.remove_columns(["input_image", "output_images"])
+    # dataset = dataset.remove_columns(["input_image", "output_images"])
     for idx, data in enumerate(dataset):
         key1, key2 = data["key"]
         task_type = data["task_type"]
@@ -238,13 +238,22 @@ def main(args):
         score2 = all_scores[key2][dimension]
         data["score"] = [score1, score2]
 
+        input_image_path = os.path.join(args.result_dir, "input_images", f"{key1}_input.png")
+        output_image_path1 = os.path.join(args.result_dir, "output_images", f"{key1}_output0.png")
+        output_image_path2 = os.path.join(args.result_dir, "output_images", f"{key2}_output1.png")
+
+        data['input_image'].save(input_image_path)
+        data['output_images'][0].save(output_image_path1)
+        data['output_images'][1].save(output_image_path2)
+
         json_line = {
             "key": (key1, key2),
             "idx": idx,
-            "score": [
-                score1,
-                score2,
-            ],
+            "score": [score1, score2],
+            "SC_reasoning": [all_scores[key1]["SC_reasoning"], all_scores[key2]["SC_reasoning"]],
+            "PQ_reasoning": [all_scores[key1]["PQ_reasoning"], all_scores[key2]["PQ_reasoning"]],
+            "input_image": input_image_path,
+            "output_images": [output_image_path1, output_image_path2],
         }
 
         save_file = os.path.join(