Update README.md

Update preconverted models

Delete onnx/vision_encoder.rknn

Split part of vision encoder to CPU and optimize Transpose ops. (Reupload to correct path)

Split part of vision encoder to CPU and optimize Transpose ops. (Reupload to correct path)

Delete rknnrun.py

Delete convert.py

Split part of vision encoder to CPU and optimize Transpose ops.

Fix RKNN data layout

Update README.md

Upload 21 files

initial commit