on CPU.
    *   Add nominal support for unsigned 16-bit integer tensor types. Note that
        very few TFLite kernels support this type natively, so its use in mobile
        ML authoring is generally discouraged.
    *   Add support for unsigned 16-bit integer tensor types in cast op.
    *   Experimental support for lowering `list_ops.tensor_list_set_item` with

int8(window) } else { carry = 1 naf[pos] = int8(window) - int8(width) } pos += w } return naf } func (s *Scalar) signedRadix16() [64]int8 { b := s.Bytes() if b[31] > 127 { panic("scalar has high bit set illegally") } var digits [64]int8 // Compute unsigned radix-16 digits: for i := 0; i < 32; i++ { digits[2*i] = int8(b[i] & 15) digits[2*i+1] = int8((b[i] >> 4) & 15) } // Recenter coefficients: for i := 0; i < 63; i++ { carry := (digits[i] + 8) >> 4 digits[i] -= carry << 4 digits[i+1] += carry } return...

- Fix bug in the port allocation logic that caused that the NodePort creation with statically assigned portNumber collide in multi-master HA cluster ([#89937](https://github.com/kubernetes/kubernetes/pull/89937), [@aojea](https://github.com/aojea)) [SIG Network and Testing]

This was designed this way mainly to allow using the same objects "yielded" by dependencies inside of background tasks, because the exit code would be executed after the background tasks were finished.