a1×b0 + 19×(a2×b4 + a3×b3 + a4×b2) r1 := mul64(a0, b1) r1 = addMul64(r1, a1, b0) r1 = addMul64(r1, a2_19, b4) r1 = addMul64(r1, a3_19, b3) r1 = addMul64(r1, a4_19, b2) // r2 = a0×b2 + a1×b1 + a2×b0 + 19×(a3×b4 + a4×b3) r2 := mul64(a0, b2) r2 = addMul64(r2, a1, b1) r2 = addMul64(r2, a2, b0) r2 = addMul64(r2, a3_19, b4) r2 = addMul64(r2, a4_19, b3) // r3 = a0×b3 + a1×b2 + a2×b1 + a3×b0 + 19×a4×b4 r3 := mul64(a0, b3) r3 = addMul64(r3, a1, b2) r3 = addMul64(r3, a2, b1) r3 = addMul64(r3, a3, b0) r3 = addMul64(r3,...

        [`experimental_io_device`](https://www.tensorflow.org/versions/r2.3/api_docs/python/tf/saved_model/LoadOptions?hl=en)
        as arg with default value `None` to choose the I/O device for loading
        models and weights.
    *   Update `tf.saved_model.SaveOptions` with
        [`experimental_io_device`](https://www.tensorflow.org/versions/r2.3/api_docs/python/tf/saved_model/SaveOptions?hl=en)