cuDNN integration enhancement and bug fixes
- [**bug**] Fix cudnn conv math type fallback [3460](https://github.com/Oneflow-Inc/oneflow/pull/3460)
设置 cudnn conv math type
- [**enhancement**] Add cuDNN implementation for bias_add [3489](https://github.com/Oneflow-Inc/oneflow/pull/3489)
增加 bias add 调用 cudnn 的实现
- [**enhancement**] Remove useless code in cudnn conv util [3480](https://github.com/Oneflow-Inc/oneflow/pull/3480)
删除 cudnn conv util 中的冗余代码
Eager execution enhancement
- [**feature**] manipulate lazy interface blobs in eager [3226](https://github.com/Oneflow-Inc/oneflow/pull/3226)
在 eager 中操作 lazy blob
CPU-only support
纯 CPU 支持的相关改动
- [**feature**] Dev pure cpu [3398](https://github.com/Oneflow-Inc/oneflow/pull/3398)
- [**enhancement**] Dev bn ln cpu [3431](https://github.com/Oneflow-Inc/oneflow/pull/3431)
- [**feature**] bert cpu [3476](https://github.com/Oneflow-Inc/oneflow/pull/3476)
- [**enhancement**] Dev Layer Norm CPU and Batch Norm CPU training [3433](https://github.com/Oneflow-Inc/oneflow/pull/3433)
- [**bug**] fix some bug in dev pure cpu [3410](https://github.com/Oneflow-Inc/oneflow/pull/3410)
- [**enhancement**] conv2d cpu support groups [3488](https://github.com/Oneflow-Inc/oneflow/pull/3488)
- [**enhancement**] Dev bn training cpu [3445](https://github.com/Oneflow-Inc/oneflow/pull/3445)
Preprocessing
数据预处理的改动
- [**feature**] image.random_crop [3442](https://github.com/Oneflow-Inc/oneflow/pull/3442)
- [**enhancement**] Refactor image resize [3450](https://github.com/Oneflow-Inc/oneflow/pull/3450)
Op/Kernel optimizations
- [**enhancement**] Optimize transpose performance [3487](https://github.com/Oneflow-Inc/oneflow/pull/3487)
优化 transpose 性能
- [**enhancement**] Refactor slice op [3444](https://github.com/Oneflow-Inc/oneflow/pull/3444)
优化 slice op
- [**feature**] Fused softmax kernel [3496](https://github.com/Oneflow-Inc/oneflow/pull/3496)
融合版本的 softmax
- [**enhancement**] Optimize conv performance [3477](https://github.com/Oneflow-Inc/oneflow/pull/3477)
优化 conv 性能
System
- [**bug**] Add B21SubTskGphBuilder [3360](https://github.com/Oneflow-Inc/oneflow/pull/3360)
修复多机消耗更多内存的 bug
Python front end
- [**enhancement**] Add typing.List global function return annotation [3443](https://github.com/Oneflow-Inc/oneflow/pull/3443)
增加 oneflow function 的返回值签名
Documentation
增加/修订文档
- [**enhancement**] add api docs [3483](https://github.com/Oneflow-Inc/oneflow/pull/3483)
- [**enhancement**] Dev add docstring2 [3305](https://github.com/Oneflow-Inc/oneflow/pull/3305)
Development and tooling
开发 / Debug 工具相关的修改
- [**feature**] Dev boxing log [3466](https://github.com/Oneflow-Inc/oneflow/pull/3466)
增加 boxing 日志
- [**feature**] add dqn to model tests [3473](https://github.com/Oneflow-Inc/oneflow/pull/3473)
DQN 加入集成测试
- [**enhancement**] py fmt respects gitignore [3492](https://github.com/Oneflow-Inc/oneflow/pull/3492)
- [**feature**] Add OF_CUDA_CHECK/OF_CUDNN_CHECK/OF_CUBLAS_CHECK/OF_CURAND_CHECK [3446](https://github.com/Oneflow-Inc/oneflow/pull/3446)
优化 CUDA CHECK 的错误信息
0.1.8-benchmark1
0.1.8-benchmark