Generating stable molecular conformations is crucial in several drug discovery applications, suchas estimating the binding affinity of a molecule toa target. Recently, generative machine learningmethods have emerged as a promising, more efficient method than molecular dynamics for sampling of conformations from the Boltzmann distribution. In this paper, we introduce Torsional-GFN, a conditional GFlowNet specifically designed to sample conformations of molecules proportionally to their Boltzmann distribution, usingonly a reward function as training signal. Conditioned on a molecular graph and its local structure(bond lengths and angles), Torsional-GFN samples rotations of its torsion angles. Our resultsdemonstrate that Torsional-GFN is able to sample conformations approximately proportional tothe Boltzmann distribution for multiple moleculeswith a single model, and allows for zero-shotgeneralization to unseen bond lengths and angles coming from the MD simulations for suchmolecules. Our work presents a promising avenuefor scaling the proposed approach to larger molecular systems, achieving zero-shot generalizationto unseen molecules, and including the generationof the local structure into the GFlowNet model.