Comment by vblanco

This is done on Nemotron models + mistral, so its not very relevant to the current frontier of cheap chinese models + big models from Claude/GPT. Big miss not having qwen or deepseek in this research.