The development of antipsychotic drugs poses challenges due to the complex pharmacology of psychiatric drugs and the limited diversity of existing drugs. Recent advances in deep-generative models have enabled the exploration of vast chemical spaces for the design of novel drugs. In this study, we leverage the use of the Llama 3.2 1B model to predict potential antipsychotic drugs. We use the ligands and target information in the BindingDB dataset. The ligands were represented using the SELFIES and Group SELFIES notation. Initially, the model was trained for unconditional generation. We also studied the effect of relative attention on the model performance using both the SELFIES and Group SELFIES representation. Then we use a pre-trained model to fine-tune the ligand-target dataset in order to discover molecules that bind to Dopamine D2 and Serotonin receptors (the primary receptors for antipsychotics). We found 10 potential candidates. We then screen them for drug candidacy. Next, a docking analysis was conducted on the screened drugs to find the binding affinity of the drug to the different proteins within the receptors. Finally, we suggest 4 molecules as potential antipsychotic drugs.