At the L2 and L3 layers, bits are either assigned into "fields" of a fixed number of bits or such a "fixed" number of bits provides the "size" of a subsequent grouping of bits. That's how a network host knows how to "parse" a frame's contents, and how the same device knows how, for example, to parse a data portion of a frame into a packet.
Unsure whether you question also encompasses L1, but how does a network device "know" where these "fields" start, or even, how do they "know" when a bit is actually being received and whether it's on or off?
At the high level, at L1, there's generally some special "signal", on the media, that when it ends/stops indicates bits are are now being sent. I.e. the receiver can then "know" here's the first bit, and then it just keeps counting them, and by their positional count, "knows" what field they are part of.
Further, on many media, bits are not sent individually, instead of group of bits are sent, concurrently as a specific signal which allows the receiver to "decode" the bits.
Generally different media standards, have their own way of doing L1. For example, although fastEthernet is 10x faster than 10 Mbps Ethernet, each bit might not be transmitted 10x faster, rather it's often some combination of how fast signals are being modulated and how much information each signal contains.
