Comparing Llama-2 and GPT-3 LLMs for HPC kernels generation