We present VeriThoughts, a dataset for reasoning-based Verilog Code generation; we also present a new benchmark based on formal verification, and a suite of small-scale models specialized for Verilog generation.