diff --git a/tasks/ssl/swav/README.md b/tasks/ssl/swav/README.md
index e573b84c..7f51859c 100644
--- a/tasks/ssl/swav/README.md
+++ b/tasks/ssl/swav/README.md
@@ -62,7 +62,8 @@ python -m paddle.distributed.launch \
     --master=$PADDLE_MASTER \
     --devices=$CUDA_VISIBLE_DEVICES \
     passl-train \
-    -c ./configs/swav_resnet50_224_lp_in1k_1n8c_dp_fp16o1.yaml
+    -c ./configs/swav_resnet50_224_lp_in1k_1n8c_dp_fp32.yaml
+
 ```
 
 ## How to End-to-End Fine-tuning
@@ -96,8 +97,8 @@ To perform end-to-end fine-tuning for SwAV:
         --master=$PADDLE_MASTER \
         --devices=$CUDA_VISIBLE_DEVICES \
         passl-train \
-        -c ./configs/swav_resnet50_224_ft_in1k_1n4c_dp_fp16o1.yaml
-        -o Global.pretrained_model=./pretrained/swav/swav_resnet50_in1k_800ep_pretrained
+        -c ./configs/swav_resnet50_224_ft_in1k_1n4c_dp_fp32.yaml
+        -o Global.pretrained_model=./pretrained/swav/swav_resnet50_in1k_800ep_bz4096_pretrained
     ```
 
 ## Other Configurations
@@ -109,10 +110,10 @@ We provide more directly runnable configurations, see [SwAV Configurations](./co
 | Model         | Phase       | Dataset      | Configs  | GPUs       | Epochs | Top1 Acc (%) | Links                                                   |
 | ------------- | ----------- | ------------ | ------------------------------------------------------------ | ---------- | ------ | -------- | ------------------------------------------------------------ |
 | resnet50 | pretrain    | ImageNet2012 | [config](./configs/swav_resnet50_224_pt_in1k_4n32c_dp_fp16o1.yaml) | A100*N2C16 | 800    | -        | [model](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_800ep_bz4096_pretrained.pdparams) \| [log](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_800ep_bz4096_pretrained.log) |
-| resnet50 | linear probe | ImageNet2012 | [config](./configs/swav_resnet50_224_lp_in1k_4n32c_dp_fp16o1.yaml) | A100*N1C8  |  100  | 75.3    |        [model](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_linearprobe.pdparams) \| [log](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_linearprobe.log) |
-| resnet50 | finetune-semi10    | ImageNet2012 | [config](./configs/swav_resnet50_224_ft_in1k_1n4c_dp_fp16o1.yaml) | A100*N1C4  | 20    | 69.0   | [model](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_10percent.pdparams) \| [log](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_10percent.log) |
-| resnet50 | finetune-semi10    | ImageNet2012 | [config](./configs/swav_resnet50_224_ft_in1k_1n4c_dp_fp16o1.yaml) | A100*N1C4  | 20    | 55.0   | [model](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_1percent.pdparams) \| [log](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_1percent.log) |
-## Citations
+| resnet50 | linear probe | ImageNet2012 | [config](./configs/swav_resnet50_224_lp_in1k_4n32c_dp_fp32.yaml) | A100*N1C8  |  100  | 75.3    |        [model](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_linearprobe.pdparams) \| [log](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_linearprobe.log) |
+| resnet50 | finetune-semi10    | ImageNet2012 | [config](./configs/swav_resnet50_224_ft_in1k_1n4c_dp_fp32.yaml) | A100*N1C4  | 20    | 69.0   | [model](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_10percent.pdparams) \| [log](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_10percent.log) |
+| resnet50 | finetune-semi10    | ImageNet2012 | [config](./configs/swav_resnet50_224_ft_in1k_1n4c_dp_fp32.yaml) | A100*N1C4  | 20    | 55.0   | [model](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_1percent.pdparams) \| [log](https://passl.bj.bcebos.com/models/swav/swav_resnet50_in1k_finetune_1percent.log) |
+
 
 ```bibtex
 @misc{caron2021unsupervised,