A mid-sized GPT-2 variant with 345 million parameters, this model sits in a practical middle ground — larger than the base GPT-2 but far more accessible than modern large-scale models. It handles text generation tasks with the characteristic style of the Megatron-trained GPT-2 family, though its age and scale mean it will struggle with complex reasoning or instruction-following compared to contemporary models.