diff --git a/assets/nextqa_loss_full.png b/assets/nextqa_loss_full.png
index 8f1b062..4a0a54a 100644
Binary files a/assets/nextqa_loss_full.png and b/assets/nextqa_loss_full.png differ
diff --git a/assets/refcoco_loss_full.png b/assets/refcoco_loss_full.png
index a4886ca..cc8ab8a 100644
Binary files a/assets/refcoco_loss_full.png and b/assets/refcoco_loss_full.png differ
diff --git a/examples/nextqa/README.md b/examples/nextqa/README.md
index dcb0cda..7770f4c 100644
--- a/examples/nextqa/README.md
+++ b/examples/nextqa/README.md
@@ -49,7 +49,7 @@ CUDA_VISIBLE_DEVICES=0 python examples/nextqa/evaluation.py \
 The `Accuracy`:
 | Aria                           | LoRA SFT               | Full Params SFT  |
 |:-------------------------------------:|:-------------------------:|:-------:|
-|78.14 | 80.80 | 81.42 |
+|78.14 | 80.80 | 80.08 |
 
 ## Loss Curve
 These are the loss curves of `LoRA SFT` and `Full Params SFT`:
diff --git a/examples/refcoco/README.md b/examples/refcoco/README.md
index 524247b..e1b5ee4 100644
--- a/examples/refcoco/README.md
+++ b/examples/refcoco/README.md
@@ -70,7 +70,7 @@ CUDA_VISIBLE_DEVICES=0 python examples/refcoco/evaluation.py \
 The `Precision@1`:
 | Aria                           | LoRA SFT               | Full Params SFT  |
 |:-------------------------------------:|:-------------------------:|:-------:|
-|2.27 | 88.68 | 88.85 |
+|2.27 | 88.68 | 88.69 |
 
 # Loss Curve
 These are the loss curves of `LoRA Finetuning` (left) and `Full Params Finetuning` (right):