Improve model card for FIRM-Edit-8B

README.md

@@ -1,39 +1,42 @@
 ---
+base_model: Qwen/Qwen3-VL-8B-Instruct
 library_name: transformers
 license: other
-base_model: Qwen/Qwen3-VL-8B-Instruct
+pipeline_tag: image-text-to-text
 tags:
+- reward-model
+- image-editing
+- FIRM
 - llama-factory
-- full
 - generated_from_trainer
 model-index:
-- name: edit_evaluation_sft_202602030104
+- name: FIRM-Edit-8B
   results: []
 ---
 
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
+# FIRM-Edit-8B
 
-# edit_evaluation_sft_202602030104
+[**Project Page**](https://firm-reward.github.io/) | [**Paper**](https://arxiv.org/abs/2603.12247) | [**GitHub**](https://github.com/VisionXLab/FIRM-Reward)
 
-This model is a fine-tuned version of [Qwen/Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct) on the instruction_following_train_v3 and the consistency_train_v3 datasets.
-It achieves the following results on the evaluation set:
-- Loss: 0.5041
+**FIRM-Edit-8B** is a robust reward model (critic) designed for faithful image editing. It is a fine-tuned version of [Qwen/Qwen3-VL-8B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct) on the **FIRM-Edit-370K** dataset. The model is part of the **FIRM (Faithful Image Reward Modeling)** framework, which provides accurate and reliable guidance for visual reinforcement learning pipelines.
 
-## Model description
+## Model Description
 
-More information needed
+Conventional reward models used for image editing often suffer from hallucinations and assign noisy scores, misguiding the optimization process. FIRM-Edit-8B addresses these issues by evaluating edits through two competing objectives:
+1. **Execution**: Adherence to the editing instruction.
+2. **Consistency**: Preservation of original content in unedited regions.
 
-## Intended uses & limitations
+By formulating a "Consistency-Modulated Execution" (CME) reward strategy, this model acts as a stable critic that mitigates hallucinations and helps establish a new standard for fidelity in image editing.
 
-More information needed
+## Intended Uses & Limitations
 
-## Training and evaluation data
-
-More information needed
+- **Reward Modeling**: To be used as a reward signal in Reinforcement Learning (RL) pipelines for image editing.
+- **Evaluation**: To serve as a metric for benchmarking the performance of image editing models.
 
 ## Training procedure
 
+The model was fine-tuned using the [LLaMA Factory](https://github.com/hiyouga/LLaMA-Factory) framework.
+
 ### Training hyperparameters
 
 The following hyperparameters were used during training:
@@ -41,12 +44,7 @@ The following hyperparameters were used during training:
 - train_batch_size: 10
 - eval_batch_size: 2
 - seed: 42
-- distributed_type: multi-GPU
-- num_devices: 8
 - gradient_accumulation_steps: 2
-- total_train_batch_size: 160
-- total_eval_batch_size: 16
-- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.1
 - num_epochs: 1.0
@@ -60,10 +58,24 @@ The following hyperparameters were used during training:
 | 0.5252        | 0.6546 | 1500 | 0.5199          |
 | 0.5075        | 0.8728 | 2000 | 0.5055          |
 
+## Usage
+
+To use the model as a reward server for RL training, you can use the script provided in the official repository:
+
+```bash
+# Launch the reward server
+python editing/reward_server/reward_server_qwen3_vl_8b_sft.py
+```
+
+## Citation
 
-### Framework versions
+If you find this work useful, please cite:
 
-- Transformers 4.57.3
-- Pytorch 2.7.1+cu128
-- Datasets 4.0.0
-- Tokenizers 0.22.2
+```bibtex
+@article{zhao2026trust,
+  title={Trust Your Critic: Robust Reward Modeling and Reinforcement Learning for Faithful Image Editing and Generation},
+  author={Zhao, Xiangyu and Zhang, Peiyuan and Lin, Junming and Liang, Tianhao and Duan, Yuchen and Ding, Shengyuan and Tian, Changyao and Zang, Yuhang and Yan, Junchi and Yang, Xue},
+  journal={arXiv preprint arXiv:2603.12247},
+  year={2026}
+}
+```