While it is clear that Type Ia supernovae (SNe) are the result of thermonuclear explosions in C/O white dwarfs (WDs), a great deal remains uncertain about the binary companion that facilitates the explosive disruption of the WD. Here, we present a comprehensive analysis of a unique, and large, data set of 127 SNe Ia with exquisite coverage by the Zwicky Transient Facility (ZTF). High-cadence (6 observations per night) ZTF observations allow us to measure the SN rise time and examine its initial evolution. We develop a Bayesian framework to model the early rise as a power-law in time, which enables the inclusion of priors in our model. For a volume-limited subset of normal SNe Ia, we find the mean power-law index is consistent with 2 in the $r_mathrm{ztf}$-band ($alpha_r = 2.01pm0.02$), as expected in the expanding fireball model. There are, however, individual SNe that are clearly inconsistent with $alpha_r=2$. We estimate a mean rise time of 18.5$,$d (with a range extending from $sim$15$-$22$,$d), though this is subject to the adopted prior. We identify an important, previously unknown, bias whereby the rise times for higher redshift SNe within a flux-limited survey are systematically underestimated. This effect can be partially alleviated if the power-law index is fixed to $alpha=2$, in which case we estimate a mean rise time of 21.0$,$d (with a range from $sim$18$-$23$,$d). The sample includes a handful or rare and peculiar SNe Ia. Finally, we conclude with a discussion of lessons learned from the ZTF sample that can eventually be applied to Large Synoptic Survey Telescope observations.