Post

One of the questions we debated while training ModernBERT was whether a modern trained encoder would unlock zero-shot reasoning using only it's generative head? Spoilers: the answer is yes.

With @bclavie.bsky.social and @ncoop57.bsky.social, we tried to answer two questions: - Can an instruction-tuned ModernBERT zero-shot tasks using the MLM-head? - Could we then fine-tune instruction-tuned ModernBERT to complete any task? Detailed answers: arxiv.org/abs/2502.03793

After instruction tuning on Flan, ModernBERT-Large-Instruct outperforms similarly sized LLMs on MMLU & MMLU-Pro, and achieves ~90 percent of Llama 3.2 1B's performance with ~65 percent fewer parameters.

When we finetune ModernBERT-Large-Instruct on task specific datasets, the generative MLM head is better or nearly equal to standard classification heads.

Can all encoders be instruction-tuned? Can we replicate ModernBERT's results with an older model like RoBERTa or peer model like GTE-en-MLM? No. And it's not close.