Instances
These are the instances tested in:
Lower and upper bounds for the nonlinear generalized assignment problem
by Claudia D’Ambrosio, Silvano Martello, and Michele Monaci
The name of each zipfile gives the characterization of

weights: linear (L), nonlinear (N);

capacities: similar (S), dissimilar (D).
Each zipfile contains one folder with 300 instances (.dat format).
Each file contains the following information

N: number of tasks;

M: number of agents;

Ftype: type of profit functions (always = 0);

Gtype: type of weight functions (=0 for nonlinear function, =1 for linear function);

c_i: capacity of each agent i (i=1, ..., m);

obja_{ji}, objb_{ji}, objc_{ji}, objd_{ji}: parameters a, b, c, and d in the definition of the profit functions (j = 1, ..., n; i=1, ..., m);

weip_{ji}, weiq_{ji}, and weiw_{ji}: parameters \pi and \tau (for the nonlinear case), and \pi (for the linear case) in the definition of the weight functions (j = 1, ..., n; i=1, ..., m);

U_j: availability of each task j (j = 1, ..., n).