Update RK3562/RK3566/RK3568/RK3576/RK3588/RV1103B/RV1103/RV1106/RK211…

…8 NPU SDK to V2.2.0 Signed-off-by: Randall Zhuo <[email protected]>
fkiller · Sep 18, 2024 · 7efa763 · 7efa763
1 parent deaba85
commit 7efa763
Show file tree

Hide file tree

Showing 95 changed files with 803 additions and 312 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,12 @@
 # CHANGELOG
 
+## v2.2.0
+
+- Support installation via pip
+- Optimize transformer model performance
+- Support Python 3.12
+- Operator optimization, such as softmax, hardmax, MatMul, etc.
+
 ## v2.1.0
 
 - Support RV1103B (Beta)

diff --git a/README.md b/README.md
@@ -21,7 +21,7 @@
   - RK3566/RK3568 Series
   - RK3562 Series
   - RV1103/RV1106
-  - RV1103B
+  - RV1103B/RV1106B
   - RK2118
 
 
@@ -42,11 +42,15 @@ Note:
 
 # Notes
 - RKNN-Toolkit2 is not compatible with [RKNN-Toolkit](https://github.com/airockchip/rknn-toolkit)
-- Currently only support on:
-  - Ubuntu 18.04 python 3.6/3.7
-  - Ubuntu 20.04 python 3.8/3.9
-  - Ubuntu 22.04 python 3.10/3.11
-- Latest version:v2.1.0
+- The supported Python versions are:
+  - Python 3.6
+  - Python 3.7
+  - Python 3.8
+  - Python 3.9
+  - Python 3.10
+  - Python 3.11
+  - Python 3.12
+- Latest version:v2.2.0
 
 
 
@@ -60,13 +64,11 @@ https://github.com/airockchip/rknn-llm
 
 # CHANGELOG
 
-## v2.1.0
- - Support RV1103B (Beta)
-- Support RK2118 (Beta)
-- Support Flash Attention (Only RK3562 and RK3576)
-- Improve MatMul API
-- Improve support for int32 and int64
-- Support more operators and operator fusion
+## v2.2.0
+- Support installation via pip
+- Optimize transformer model performance
+- Support Python 3.12
+- Operator optimization, such as softmax, hardmax, MatMul, etc.
 
  for older version, please refer [CHANGELOG](CHANGELOG.md)
 

diff --git a/...RK2118_Quick_Start_RKNN_SDK_V2.1.0_CN.pdf → ...RK2118_Quick_Start_RKNN_SDK_V2.2.0_CN.pdf b/...RK2118_Quick_Start_RKNN_SDK_V2.1.0_CN.pdf → ...RK2118_Quick_Start_RKNN_SDK_V2.2.0_CN.pdf
diff --git a/...RK2118_Quick_Start_RKNN_SDK_V2.1.0_EN.pdf → ...RK2118_Quick_Start_RKNN_SDK_V2.2.0_EN.pdf b/...RK2118_Quick_Start_RKNN_SDK_V2.1.0_EN.pdf → ...RK2118_Quick_Start_RKNN_SDK_V2.2.0_EN.pdf
diff --git a/..._RKNPU_Quick_Start_RKNN_SDK_V2.1.0_CN.pdf → ..._RKNPU_Quick_Start_RKNN_SDK_V2.2.0_CN.pdf b/..._RKNPU_Quick_Start_RKNN_SDK_V2.1.0_CN.pdf → ..._RKNPU_Quick_Start_RKNN_SDK_V2.2.0_CN.pdf
diff --git a/..._RKNPU_Quick_Start_RKNN_SDK_V2.1.0_EN.pdf → ..._RKNPU_Quick_Start_RKNN_SDK_V2.2.0_EN.pdf b/..._RKNPU_Quick_Start_RKNN_SDK_V2.1.0_EN.pdf → ..._RKNPU_Quick_Start_RKNN_SDK_V2.2.0_EN.pdf
diff --git a/...RV1103_Quick_Start_RKNN_SDK_V2.1.0_CN.pdf → ...RV1103_Quick_Start_RKNN_SDK_V2.2.0_CN.pdf b/...RV1103_Quick_Start_RKNN_SDK_V2.1.0_CN.pdf → ...RV1103_Quick_Start_RKNN_SDK_V2.2.0_CN.pdf
diff --git a/...RV1103_Quick_Start_RKNN_SDK_V2.1.0_EN.pdf → ...RV1103_Quick_Start_RKNN_SDK_V2.2.0_EN.pdf b/...RV1103_Quick_Start_RKNN_SDK_V2.1.0_EN.pdf → ...RV1103_Quick_Start_RKNN_SDK_V2.2.0_EN.pdf
diff --git a/...p_RKNPU_User_Guide_RKNN_SDK_V2.1.0_CN.pdf → ...p_RKNPU_User_Guide_RKNN_SDK_V2.2.0_CN.pdf b/...p_RKNPU_User_Guide_RKNN_SDK_V2.1.0_CN.pdf → ...p_RKNPU_User_Guide_RKNN_SDK_V2.2.0_CN.pdf
diff --git a/...p_RKNPU_User_Guide_RKNN_SDK_V2.1.0_EN.pdf → ...p_RKNPU_User_Guide_RKNN_SDK_V2.2.0_EN.pdf b/...p_RKNPU_User_Guide_RKNN_SDK_V2.1.0_EN.pdf → ...p_RKNPU_User_Guide_RKNN_SDK_V2.2.0_EN.pdf
diff --git a/...API_Reference_RKNN_Toolkit2_V2.1.0_CN.pdf → ...API_Reference_RKNN_Toolkit2_V2.2.0_CN.pdf b/...API_Reference_RKNN_Toolkit2_V2.1.0_CN.pdf → ...API_Reference_RKNN_Toolkit2_V2.2.0_CN.pdf
diff --git a/...API_Reference_RKNN_Toolkit2_V2.1.0_EN.pdf → ...API_Reference_RKNN_Toolkit2_V2.2.0_EN.pdf b/...API_Reference_RKNN_Toolkit2_V2.1.0_EN.pdf → ...API_Reference_RKNN_Toolkit2_V2.2.0_EN.pdf
diff --git a/..._RKNPU_API_Reference_RKNNRT_V2.1.0_CN.pdf → ..._RKNPU_API_Reference_RKNNRT_V2.2.0_CN.pdf b/..._RKNPU_API_Reference_RKNNRT_V2.1.0_CN.pdf → ..._RKNPU_API_Reference_RKNNRT_V2.2.0_CN.pdf
diff --git a/..._RKNPU_API_Reference_RKNNRT_V2.1.0_EN.pdf → ..._RKNPU_API_Reference_RKNNRT_V2.2.0_EN.pdf b/..._RKNPU_API_Reference_RKNNRT_V2.1.0_EN.pdf → ..._RKNPU_API_Reference_RKNNRT_V2.2.0_EN.pdf
diff --git a/...Compiler_Support_Operator_List_V2.1.0.pdf → ...Compiler_Support_Operator_List_V2.2.0.pdf b/...Compiler_Support_Operator_List_V2.1.0.pdf → ...Compiler_Support_Operator_List_V2.2.0.pdf
diff --git a/doc/RKNNToolKit2_OP_Support-v2.1.0.md → doc/RKNNToolKit2_OP_Support-2.2.0.md b/doc/RKNNToolKit2_OP_Support-v2.1.0.md → doc/RKNNToolKit2_OP_Support-2.2.0.md
diff --git a/doc/Using RKNN-ToolKit2 in WSL.md b/doc/Using RKNN-ToolKit2 in WSL.md
@@ -4,8 +4,8 @@
 - You can refer to the following link for installing WSL: https://learn.microsoft.com/en-us/windows/wsl/install
 
 ## 2. Using RKNN-Toolkit2 in WSL
-1. Refer to the document[《Rockchip_RKNPU_Quick_Start》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/01_Rockchip_RKNPU_Quick_Start_RKNN_SDK_V2.0.0beta0_EN.pdf) to install the RKNN-ToolKit2 environment in WSL
-2. Refer to the document[《Rockchip_RKNPU_User_Guide》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/02_Rockchip_RKNPU_User_Guide_RKNN_SDK_V2.0.0beta0_EN.pdf) for model conversion, quantization, and other operations in WSL.
+1. Refer to the document[《Rockchip_RKNPU_Quick_Start》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/) to install the RKNN-ToolKit2 environment in WSL
+2. Refer to the document[《Rockchip_RKNPU_User_Guide》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/) for model conversion, quantization, and other operations in WSL.
 
 ## 3. Using RKNN-Toolkit2 for Board Debugging in WSL
 
@@ -45,7 +45,7 @@ adb connect <IP address:port>
 - `IP address` is the IP address of the Windows host.
 
 ### 3.3 Using RKNN-Toolkit2 for Board Debugging
-Refer to the document [《Rockchip_RKNPU_User_Guide》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/02_Rockchip_RKNPU_User_Guide_RKNN_SDK_V2.0.0beta0_EN.pdf) for board inference, accuracy analysis, and other operations in WSL.
+Refer to the document [《Rockchip_RKNPU_User_Guide》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/) for board inference, accuracy analysis, and other operations in WSL.
 
 
 

diff --git a/doc/WSL中使用RKNN_ToolKit2.md b/doc/WSL中使用RKNN_ToolKit2.md
@@ -4,8 +4,8 @@
 - 可参考 https://learn.microsoft.com/zh-cn/windows/wsl/install
 
 ## 2. WSL 中使用RKNN-Toolkit2
-1. 参考[《Rockchip_RKNPU_Quick_Start手册》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/01_Rockchip_RKNPU_Quick_Start_RKNN_SDK_V2.0.0beta0_CN.pdf)在 WSL 中安装RKNN-ToolKit2环境
-2. 参考[《Rockchip_RKNPU_User_Guide手册》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/02_Rockchip_RKNPU_User_Guide_RKNN_SDK_V2.0.0beta0_CN.pdf)在 WSL 进行模型转换、量化等操作
+1. 参考[《Rockchip_RKNPU_Quick_Start手册》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/)在 WSL 中安装RKNN-ToolKit2环境
+2. 参考[《Rockchip_RKNPU_User_Guide手册》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/)在 WSL 进行模型转换、量化等操作
 
 ## 3. WSL 中使用RKNN-Toolkit2进行连板调试
 
@@ -45,7 +45,7 @@ adb connect <IP地址:端口号>
 - `IP地址` 为 Windows主机 的IP地址
 
 ### 3.3 使用RKNN-ToolKit2进行连板调试
-参考[《Rockchip_RKNPU_User_Guide手册》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/02_Rockchip_RKNPU_User_Guide_RKNN_SDK_V2.0.0beta0_CN.pdf)在 WSL 进行连板推理、连板精度分析等操作
+参考[《Rockchip_RKNPU_User_Guide手册》](https://github.com/airockchip/rknn-toolkit2/blob/master/doc/)在 WSL 进行连板推理、连板精度分析等操作
 
 
 

diff --git a/rknn-toolkit-lite2/CHANGELOG.txt b/rknn-toolkit-lite2/CHANGELOG.txt
@@ -1,3 +1,7 @@
+2024-09-14
+版本: v2.2.0
+1. 修复已知BUG。
+
 2024-08-08
 版本: v2.1.0
 1. 修复已知BUG。

diff --git a/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3562.rknn b/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3562.rknn
diff --git a/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3566_rk3568.rknn b/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3566_rk3568.rknn
diff --git a/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3576.rknn b/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3576.rknn
diff --git a/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3588.rknn b/rknn-toolkit-lite2/examples/dynamic_shape/mobilenet_v2_for_rk3588.rknn
diff --git a/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3562.rknn b/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3562.rknn
diff --git a/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3566_rk3568.rknn b/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3566_rk3568.rknn
diff --git a/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3576.rknn b/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3576.rknn
diff --git a/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3588.rknn b/rknn-toolkit-lite2/examples/resnet18/resnet18_for_rk3588.rknn
diff --git a/rknn-toolkit-lite2/packages/rknn-toolkit-lite2-v2.2.0_packages.md5sum b/rknn-toolkit-lite2/packages/rknn-toolkit-lite2-v2.2.0_packages.md5sum
@@ -0,0 +1,6 @@
+a69816d9d726afb408a14e6b1670b0cb  rknn_toolkit_lite2-2.2.0-cp310-cp310-linux_aarch64.whl
+ee32d72bbf968133b5314c68d731b661  rknn_toolkit_lite2-2.2.0-cp311-cp311-linux_aarch64.whl
+c0986325c9af139916e2383c4a05079c  rknn_toolkit_lite2-2.2.0-cp312-cp312-linux_aarch64.whl
+c914cc293bbb4a2c1eccfa267bfd727c  rknn_toolkit_lite2-2.2.0-cp37-cp37m-linux_aarch64.whl
+46a239929fbf7b820646f0d54ec6ed09  rknn_toolkit_lite2-2.2.0-cp38-cp38-linux_aarch64.whl
+7c8a01028d0b496c2c805f4b14749a99  rknn_toolkit_lite2-2.2.0-cp39-cp39-linux_aarch64.whl
diff --git a/...lite2-2.1.0-cp310-cp310-linux_aarch64.whl → ...lite2-2.2.0-cp310-cp310-linux_aarch64.whl b/...lite2-2.1.0-cp310-cp310-linux_aarch64.whl → ...lite2-2.2.0-cp310-cp310-linux_aarch64.whl
diff --git a/...lite2-2.1.0-cp311-cp311-linux_aarch64.whl → ...lite2-2.2.0-cp311-cp311-linux_aarch64.whl b/...lite2-2.1.0-cp311-cp311-linux_aarch64.whl → ...lite2-2.2.0-cp311-cp311-linux_aarch64.whl
diff --git a/...lite2-2.1.0-cp312-cp312-linux_aarch64.whl → ...lite2-2.2.0-cp312-cp312-linux_aarch64.whl b/...lite2-2.1.0-cp312-cp312-linux_aarch64.whl → ...lite2-2.2.0-cp312-cp312-linux_aarch64.whl
diff --git a/..._lite2-2.1.0-cp37-cp37m-linux_aarch64.whl → ..._lite2-2.2.0-cp37-cp37m-linux_aarch64.whl b/..._lite2-2.1.0-cp37-cp37m-linux_aarch64.whl → ..._lite2-2.2.0-cp37-cp37m-linux_aarch64.whl
diff --git a/...t_lite2-2.1.0-cp38-cp38-linux_aarch64.whl → ...t_lite2-2.2.0-cp38-cp38-linux_aarch64.whl b/...t_lite2-2.1.0-cp38-cp38-linux_aarch64.whl → ...t_lite2-2.2.0-cp38-cp38-linux_aarch64.whl
diff --git a/...t_lite2-2.1.0-cp39-cp39-linux_aarch64.whl → ...t_lite2-2.2.0-cp39-cp39-linux_aarch64.whl b/...t_lite2-2.1.0-cp39-cp39-linux_aarch64.whl → ...t_lite2-2.2.0-cp39-cp39-linux_aarch64.whl
diff --git a/rknn-toolkit2/CHANGELOG.txt b/rknn-toolkit2/CHANGELOG.txt
@@ -1,3 +1,47 @@
+2024-9-2:
+版本: v2.2.0:
+更新内容:
+1. 增加对bool输出的连扳推理支持
+2. 增加rv1106b支持
+3. whl包标准化(manylinux)
+4. 优化pad和mul相关的消融规则
+5. 优化TP的图级别消除
+6. 优化QAT模型支持
+7. 优化Gather性能
+8. 提高连板稳定性
+9. 更新动态shape功能
+10. 修复大模型混合量化报错
+11. 修复Caffe的inplace layer报错
+12. 完善带broadcast的Less/Greater/LessOrEqual/GreaterOrEqual/And/Or/Max/Min/Mean/Sum算子情况支持
+13. 增加RK3576 exSoftmaxMask(softmax+where)算子支持
+14. 完善Softmax算子支持
+15. 增加RK3562、RK3588、RK1106 Hardmax支持
+16. 完善RK3576 ConvDepthwise + Eltwise OP融合
+17. 增加RK3576 FP16类型Conv稀疏化推理
+18. 修复多线程运行失败问题
+19. 修复RKLLM模型和RKNN模型混合运行失败问题
+
+2024-8-22
+版本: v2.1.1b1:
+更新内容:
+1. 增加Transpose NPU算子支持，例如放宽perm=[1,2,0,3]规格NPU算子限制。
+2. 增强mode=bilinear Resize CPU算子浮点类型性能。
+3. 增加pixel-norm算子CPU实现。
+4. 增加python3.12的支持
+5. 增加动态权重conv的部分支持
+6. 增加PixelNorm/Sum支持
+7. 优化split&elements子图结构性能
+8. 优化exNorm前后Reshape开销
+9. 完善 einsum 支持
+10. 修复超2G模型报错
+
+2024-8-14
+版本: v2.1.1b0:
+更新内容:
+1. rk3576增加softmaxmask NPU算子支持。
+2. 增加lpNormalization非4维支持
+3. 优化Pool/GatherElements性能
+
 2024-8-1
 版本: v2.1.0:
 更新内容:

diff --git a/rknn-toolkit2/docker/docker_file/ubuntu_20_04_cp38/Dockerfile_ubuntu_20_04_for_cp38 b/rknn-toolkit2/docker/docker_file/ubuntu_20_04_cp38/Dockerfile_ubuntu_20_04_for_cp38
@@ -22,6 +22,7 @@ RUN pip3 config set install.trusted-host pypi.tuna.tsinghua.edu.cn
 
 RUN python3 --version
 RUN pip3 --version
-COPY rknn_toolkit2-2.1.0+708089d1-cp38-cp38-linux_x86_64.whl rknn_toolkit2-2.1.0+708089d1-cp38-cp38-linux_x86_64.whl
+COPY rknn_toolkit2-2.2.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl rknn_toolkit2-2.2.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
 RUN pip3 install torch==1.10.1
-RUN pip3 install rknn_toolkit2-2.1.0+708089d1-cp38-cp38-linux_x86_64.whl
+RUN pip3 install rknn_toolkit2-2.2.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+RUN pip3 cache purge
diff --git a/...ker/docker_file/ubuntu_20_04_cp38/rknn_toolkit2-2.1.0+708089d1-cp38-cp38-linux_x86_64.whl b/...ker/docker_file/ubuntu_20_04_cp38/rknn_toolkit2-2.1.0+708089d1-cp38-cp38-linux_x86_64.whl
diff --git a/....1.0+708089d1-cp36-cp36m-linux_x86_64.whl → ...inux_2_17_x86_64.manylinux2014_x86_64.whl b/....1.0+708089d1-cp36-cp36m-linux_x86_64.whl → ...inux_2_17_x86_64.manylinux2014_x86_64.whl
diff --git a/rknn-toolkit2/docker/docker_file/ubuntu_20_04_cp38/sources_bionic.list b/rknn-toolkit2/docker/docker_file/ubuntu_20_04_cp38/sources_bionic.list
@@ -1,14 +1,8 @@
-deb http://mirrors.aliyun.com/ubuntu/ focal main restricted universe multiverse
-deb-src http://mirrors.aliyun.com/ubuntu/ focal main restricted universe multiverse
-
-deb http://mirrors.aliyun.com/ubuntu/ focal-security main restricted universe multiverse
-deb-src http://mirrors.aliyun.com/ubuntu/ focal-security main restricted universe multiverse
-
-deb http://mirrors.aliyun.com/ubuntu/ focal-updates main restricted universe multiverse
-deb-src http://mirrors.aliyun.com/ubuntu/ focal-updates main restricted universe multiverse
-
-deb http://mirrors.aliyun.com/ubuntu/ focal-proposed main restricted universe multiverse
-deb-src http://mirrors.aliyun.com/ubuntu/ focal-proposed main restricted universe multiverse
-
-deb http://mirrors.aliyun.com/ubuntu/ focal-backports main restricted universe multiverse
-deb-src http://mirrors.aliyun.com/ubuntu/ focal-backports main restricted universe multiverse
+deb http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal main restricted universe multiverse
+# deb-src http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal main restricted universe multiverse
+deb http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal-updates main restricted universe multiverse
+# deb-src http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal-updates main restricted universe multiverse
+deb http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal-backports main restricted universe multiverse
+# deb-src http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal-backports main restricted universe multiverse
+deb http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal-security main restricted universe multiverse
+# deb-src http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ focal-security main restricted universe multiverse multiverse
diff --git a/rknn-toolkit2/packages/md5sum.txt b/rknn-toolkit2/packages/md5sum.txt
@@ -1,6 +1,7 @@
-49dc6e0b6c276db1db44ba38e6f4a277  rknn_toolkit2-2.1.0+708089d1-cp310-cp310-linux_x86_64.whl
-3ed72827476a2b2905471c8dddd33a04  rknn_toolkit2-2.1.0+708089d1-cp311-cp311-linux_x86_64.whl
-5d8441cb5951cff671cb678584c83d93  rknn_toolkit2-2.1.0+708089d1-cp36-cp36m-linux_x86_64.whl
-004c4d1eebc3f55572858020145ed147  rknn_toolkit2-2.1.0+708089d1-cp37-cp37m-linux_x86_64.whl
-f819ebb2a3badbc44b21924969c7f588  rknn_toolkit2-2.1.0+708089d1-cp38-cp38-linux_x86_64.whl
-f84006e40bf8039757a2ae67013c8a11  rknn_toolkit2-2.1.0+708089d1-cp39-cp39-linux_x86_64.whl
+c251f5bc4f225ef6b51063a049c25c49  rknn_toolkit2-2.2.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+a977e101a4442e2318ad65346a1ffc48  rknn_toolkit2-2.2.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+f696ddda34999ded3c678fc71a6cf96e  rknn_toolkit2-2.2.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+c90dc194440ff14c5e64634eb89a0c10  rknn_toolkit2-2.2.0-cp36-cp36m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+8008bf8d3d37d7646e7e68d60e6657cc  rknn_toolkit2-2.2.0-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+6894f32075a517b78e5f3cab1608dd8c  rknn_toolkit2-2.2.0-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
+f6b2ca1a63b1ef7ec4afadcfde623af8  rknn_toolkit2-2.2.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
diff --git a/...it2/packages/requirements_cp310-2.1.0.txt → ...it2/packages/requirements_cp310-2.2.0.txt b/...it2/packages/requirements_cp310-2.1.0.txt → ...it2/packages/requirements_cp310-2.2.0.txt
@@ -14,7 +14,6 @@ numpy<=1.26.4
 
 # base
 onnx==1.14.1
-onnxoptimizer==0.3.8
 onnxruntime==1.16.0
 torch>=1.13.1,<=2.1.0
-tensorflow==2.8.1
+# tensorflow>=2.8.0rc0,<=2.14.0
diff --git a/...it2/packages/requirements_cp311-2.1.0.txt → ...it2/packages/requirements_cp311-2.2.0.txt b/...it2/packages/requirements_cp311-2.1.0.txt → ...it2/packages/requirements_cp311-2.2.0.txt
@@ -1,7 +1,7 @@
 # if install failed, please change the pip source to 'https://pypi.tuna.tsinghua.edu.cn/simple/'
 
 # base deps
-protobuf==3.20.3
+protobuf>=4.21.0,<=4.25.4
 
 # utils
 psutil>=5.9.0
@@ -10,12 +10,10 @@ scipy>=1.9.3
 tqdm>=4.64.1
 opencv-python>=4.5.5.64
 fast-histogram>=0.11
-Pillow>=10.0.1
 numpy<=1.26.4
 
 # base
 onnx==1.14.1
-onnxoptimizer==0.3.8
 onnxruntime==1.16.0
 torch>=1.13.1,<=2.1.0
-tensorflow==2.14.0
+# tensorflow>=2.12.0rc0,<=2.14.0
diff --git a/rknn-toolkit2/packages/requirements_cp312-2.2.0.txt b/rknn-toolkit2/packages/requirements_cp312-2.2.0.txt
@@ -0,0 +1,19 @@
+# if install failed, please change the pip source to 'https://pypi.tuna.tsinghua.edu.cn/simple/'
+
+# base deps
+protobuf>=4.21.0,<=4.25.4
+
+# utils
+psutil>=5.9.0
+ruamel.yaml>=0.17.21
+scipy>=1.9.3
+tqdm>=4.64.1
+opencv-python>=4.5.5.64
+fast-histogram>=0.11
+numpy<=1.26.4
+
+# base
+onnx==1.16.1
+onnxruntime==1.17.0
+torch>=1.13.1,<=2.2.0
+# tensorflow<=2.16.0rc0
diff --git a/...kit2/packages/requirements_cp36-2.1.0.txt → ...kit2/packages/requirements_cp36-2.2.0.txt b/...kit2/packages/requirements_cp36-2.1.0.txt → ...kit2/packages/requirements_cp36-2.2.0.txt
@@ -1,7 +1,7 @@
 # if install failed, please change the pip source to 'https://pypi.tuna.tsinghua.edu.cn/simple/'
 
 # base deps
-protobuf==3.12.2
+protobuf>=3.12.2,<=3.19.6
 
 # utils
 psutil>=5.9.0
@@ -14,7 +14,6 @@ numpy<=1.26.4
 
 # base
 onnx==1.10.0
-onnxoptimizer==0.2.7
 onnxruntime==1.10.0
 torch>=1.10.1,<=2.1.0
-tensorflow==2.6.2
+# tensorflow>=1.12.0,<=2.6.2