AIbin

Results 7 issues of AIbin

Paddle Inference 支持动态图 & 静态图混合推理装饰器使用方法

Latest optimization: Re-network DIT, simplify the original model dynamic graph into a high-performance model network, 1. For the core that consumes more time: the transformer part uses `paddle.incubate.jit.inference` to do...

add to_static & new_ir & adaptive_layer_norm op & QKV op

### PR types ### PR changes ### Description 这个PR是为了解决Stable Diffusion 3 模型优化中的 Clip模型动转静时出现的参数列表不对应的问题。修改之前,转为静态图之后会出现参数类型不匹配的问题,生成的静态图会将输入输出的所有

we add SD3 batch parallel, Improve the single inference speed of SD3 (512*512, 50 steps) to about 0.86s