Do two beams of light attract each other in general theory of relativity?
The general answer is "it depends." Light has energy, momentum, and puts a pressure in the direction of motion, and these are all equal in magnitude (in units of c = 1). All of these things contribute to the stress-energy tensor, so by the Einstein field equation, it is unambiguous to say that light produces gravitational effects.
However, the relationship between energy, momentum, and pressure in the direction of propagation leads to some effects which might not otherwise be expected. The most famous is that the deflection of light by matter happens at exactly twice the amount predicted by a massive particle, at least in the sense that in linearized GTR, ignoring the pressure term halves the effect (one can also compare it a naive model of a massive particle at the speed of light in Newtonian gravity, and again the GTR result is exactly twice that).
Similarly, antiparallel (opposite direction) light beams attract each other by four times the naive (pressureless or Newtonian) expectation, while parallel (same direction) light beams do not attract each other at all. A good paper to start with is: Tolman R.C., Ehrenfest P., and Podolsky B., Phys. Rev. 37 (1931) 602. Something one might worry about is whether the result is true to higher orders as well, but the light beams would have to be extremely intense for them to matter. The first order (linearized) effect between light beams is already extremely small.
According to general relativity, yes, two beams of light would gravitationally attract each other. Einstein's equation says that
$$R^{\mu\nu} - \frac{1}{2}R g^{\mu\nu} = 8\pi T^{\mu\nu}$$
The terms on the left represent the distortion ("curvature") of spacetime, and the term on the right represents matter and energy, including light. As long as $T^{\mu\nu}$ is nonzero, there will have to be some sort of induced distortion a.k.a. gravity, since $R^{\mu\nu} - \frac{1}{2}R g^{\mu\nu} = 0$ in flat spacetime.
In case you're interested, the relevant equations are the definition of the stress-energy tensor for electromagnetism,
$$T^{\mu\nu} = -\frac{1}{\mu_0}\biggl(F^{\mu\rho}F_{\rho}^{\ \nu} + \frac{1}{4}g^{\mu\nu}F^{\rho\sigma}F_{\rho\sigma}\biggr)$$
where $F$ is the electromagnetic field tensor, and the electromagnetic wave equation
$$D_{\alpha}D^{\alpha}F^{\mu\nu} = 0$$
where $D_{\alpha}$ is the covariant derivative operator. In principle, to calculate the gravitational attraction between two beams of light, you would identify the functions $F^{\mu\nu}$ that correspond to your beams (they would have to satisfy the wave equation), and then plug them in to calculate $T^{\mu\nu}$. When you put that into Einstein's equation, it places a constraint on the possible values of the metric $g_{\mu\nu}$ and its derivatives, and you could use that constraint to determine the geodesic deviation between the two beams of light, which, in a sense, corresponds to their gravitational attraction.
Yes. The energy-momentum tensor (which is on the right hand side of Einstein's equation) is non-zero in the presence of any kind of energy density, including radiation. This means that the light beams will curve spacetime (measured by the left hand side of Einstein's equation) and hence affect the path that the light takes. But for typical light beams, this is very small and hence neglectable.