Search papers, labs, and topics across Lattice.
This paper introduces Weak-to-Strong On-Policy Distillation (W2S-OPD), a novel framework that enhances the capabilities of a student model by distilling knowledge from multiple weaker models instead of relying on a single, stronger teacher. By constructing a proxy teacher in logit space from contrast pairs of positive and negative models, W2S-OPD effectively isolates and transfers specific capability directions to the student, allowing it to surpass even its stronger counterparts. The method demonstrates significant performance improvements across various math and code benchmarks, showcasing its ability to leverage weaker models for effective learning and continuous enhancement.
Distilling from weaker models can enable a student to outperform its stronger counterparts, challenging the conventional wisdom of model hierarchy in AI training.
On-policy distillation (OPD), which aligns a student with the teacher's token-level distribution on the student's own rollouts, is an effective paradigm for transferring capabilities across LLMs. Prevailing approaches assume a teacher at least as capable as the student: they either distill a larger model into a smaller one, which fails at the frontier where no larger teacher exists, or consolidate multiple domain experts trained from a shared base, which requires costly training at the student's scale. We introduce Weak-to-Strong On-Policy Distillation (W2S-OPD), a simple yet effective OPD framework that improves the strong student by distilling from multiple weak models. W2S-OPD constructs a proxy teacher in logit space from a contrast pair of a positive and a negative model, both smaller than the student and cheap to obtain. Their logit difference isolates the capability direction, which is added to the student's own base model, yielding a proxy teacher that couples this direction while staying distributionally adjacent to the student. The student then distills it by minimizing the per-token reverse KL on its own rollouts. We instantiate the contrast pair as i) a post-RL expert against its pre-RL initialization, isolating the skill RL instills, ii) a larger against a smaller base model, isolating the capability from scale, and iii) a small base model with correct versus wrong hints, isolating the instance-level direction toward the solution. Across four math and three code benchmarks, W2S-OPD outperforms OPD, enables the student to surpass the domain teacher, and keeps improving the student even when every supervision source is weaker. Analysis shows different contrasts yield distinct signals: the post-RL and hint contrasts emphasize reasoning frameworks, while the scale contrast emphasizes the solving procedure. Our code will be available at https://github.com/Yu-Fangxu/W2S-OPD.
Microsoft Research