(docs): update docs.

Author: PanAndy

README.md

@@ -41,7 +41,7 @@ Leveraging a multi-role distributed architecture with Ray for flexible resource
 | 📣   Updates                                                                                                                                                                                                                                                                                                                            |
 |:----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | **[10/23/2025]** 🎉 Our Papers released, see [Asymmetric Proximal Policy Optimization: mini-critics boost LLM reasoning](https://arxiv.org/abs/2510.01656) and [Attention Illuminates LLM Reasoning: The Preplan-and-Anchor Rhythm Enables Fine-Grained Policy Optimization](https://arxiv.org/abs/2510.13554).                         |
-| **[10/14/2025]** 🎉 Our Paper released, see [Part II: ROLL Flash -- Accelerating RLVR and Agentic Training with Asynchrony](https://arxiv.org/abs/2510.11345), the code will be released soon.                                                                                                                                          |
+| **[10/14/2025]** 🎉 Our Paper released, see [Part II: ROLL Flash -- Accelerating RLVR and Agentic Training with Asynchrony](https://arxiv.org/abs/2510.11345).                                                                                                                                                                          |
 | **[09/28/2025]** 🎉 Ascend NPU support — see [usage guide](https://alibaba.github.io/ROLL/docs/English/UserGuide/ascend/ascend_usage).                                                                                                                                                                                                  |
 | **[09/25/2025]** 🎉 Our Paper released, see [RollPacker: Mitigating Long-Tail Rollouts for Fast, Synchronous RL Post-Training](https://arxiv.org/abs/2509.21009)                                                                                                                                                                        |
 | **[09/24/2025]** 🎉 Support [Wan2_2 Reward FL pipeline](examples/wan2.2-14B-reward_fl_ds/reward_fl_config.yaml). Explore the new capabilities!                                                                                                                                                                                          |

docs_roll/docs/English/UserGuide/algorithms/Reward_FL.md

@@ -14,7 +14,7 @@ In ROLL, the Reward FL algorithm-specific configuration parameters are as follow
 
 ```yaml
 # reward fl
-learning_rate: 2.5e-6
+learning_rate: 2e-6
 lr_scheduler_type: constant
 per_device_train_batch_size: 1
 gradient_accumulation_steps: 1
@@ -73,7 +73,22 @@ The following parameters related to Wan2_2 are as follows:
 - `Distilled Wan2.2 DiT safetensors`: [lightx2v/Wan2.2-Lightning](https://huggingface.co/lightx2v/Wan2.2-Lightning/tree/main)
 - `Reward Model`: [deepinsight/insightface](https://github.com/deepinsight/insightface/releases/download/v0.7/antelopev2.zip) 
 
+## Preprocess checkpoints
+- Run `merge_model.py` to merge multiple files of `Official Wan2.2 pipeline` high noise model and low noise model into one file, respectively.
+- Run `merge_lora.py` to merge `Distilled Wan2.2 DiT safetensors` lora to the base model of `Official Wan2.2 pipeline` high noise model and low noise model, respectively.
+
+## Setup environments
+```
+pip install -r requirements_torch260_diffsynth.txt
+```
+
 ## Reference Example
 
 You can refer to the following configuration file to set up Reward FL training:
+
 - `./examples/docs_examples/example_reward_fl.yaml`
+
+Run `run_reward_fl_ds_pipeline.sh` to get start.
+
+## Reference
+[1]: Identity-Preserving Image-to-Video Generation via Reward-Guided Optimization. https://arxiv.org/abs/2510.14255