Are action-based preferences essential? One of many key elements of ACT is that the contrastive pairs spotlight variations between conversational actions. In “ACT w/ Random Actions”, we moreover study the significance of motion choice by randomly sampling each the profitable and shedding motion when establishing the desire pair, and observe this underperforms regular ACT.
Do we’d like on-policy sampling? In “ACT w/o on-policy sampling”, we study the significance of on-policy sampling by evaluating regular off-policy DPO on the dataset as constructed in Section 1. Whereas we do observe some enhancements over SFT (e.g., from 69.0 to 74.8 Macro F1), the general enhancements are a lot bigger when utilizing on-policy sampling as with full ACT. This can be because of the truth that the off-policy adverse responses will not be assured to lie within the language manifold of the coverage mannequin, and distribution shift could also be too troublesome to beat with off-policy studying.
Is trajectory simulation essential? ACT is better-aligned with multi-turn conversations because of its trajectory simulation. With out multi-turn simulation, our strategy may be considered equally to on-policy DPO variants like IRPO, however with a conversation-specific reward sign which accounts for dialog actions and process heuristics. In “ACT w/ sampling w/o simulation”, we discover that this trajectory-level simulation is important to enhancing multi-turn efficiency, particularly the coverage mannequin’s capability to purpose about its personal clarification questions.
Is ACT mannequin agnostic? The bottom mannequin in our most important experiments, Zephyr, is obtained by aligning Mistral. In “ACT with unaligned basis fashions” we observe a efficiency hole of 6.5 Motion F1 and 4.3 Trajectory F1 after ACT tuning for the 2 fashions. Nonetheless, our outcomes exhibit ACT can enhance efficiency no matter pre-existing alignment with human suggestions, though it may assist as an improved mannequin initialization. General, we discover that enhancing base mannequin efficiency with ACT is mannequin agnostic.
