4 - Compiler supports more operations
5 - New command line interface for user interface consistancy
7 - Runtime CPU backend supports more operations
8 - Runtime CPU backend supports more quant8 operations
14 ### Compiler supports more operations
16 - MatrixDiag, MatrixSetDiag, ReverseSequence, ReverseV2, SegmentSum, SelectV2, SparseToDense, Where
18 ### New command line interface for user interface consistancy
20 - one-import: imports conventional model files to circle
21 - one-import-tf: imports TensorFlow model to circle
22 - one-import-tflite: imports TensorFlow lite model to circle
23 - one-optimize: circle optimize command
24 - one-quantize: circle quantize command
25 - supports float32 to uint8, layer wise (for Conv series)
26 - one-pack: package command
27 - one-prepare-venv: prepares python virtual environment for importing TensorFlow model
28 - one-codegen: backend(if available) code generator
32 ### Runtime CPU backend supports more operations
34 - LogSoftmax, SpaceToBatchND
36 ### Runtime CPU backend supports more quant8 operations
38 - Logistic, Mul, Tanh, SpaceToBatchND, Transpose, Sub, Max, Min, Less, Greater, GreaterEqual, LessEqual, Equal, NotEqual
42 - Introduce basic asynchronous execution API
46 - Remove dynamic tensor overhead from static models