single mention · post-training
RL-based post-training framework with complementary rewards
An RL-based post-training framework that optimizes three complementary reward signals: reasoning quality, reasoning-action consistency, and trajectory quality.1 Unlike SFT, it provides explicit feedback on the model's own rollouts, aligning training with how the system is deployed.2
Trade-off Compared with SFT, it yields disproportionately larger gains in robustness and generalization for the same compute budget.3
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
an RL-based post-training framework shown in Fig. 6 that optimizes three complementary reward signals: reasoning quality (via large reasoning model feedback), reasoning-action consistency, and trajectory quality
— Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long TailRL provides explicit inference feedback on the model’s own rollouts, aligning the optimization objective with how the system is actually deployed
— Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tailyields disproportionately larger gains in robustness and generalization for the same compute budget
— Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail
Also called RL-based post-training framework.
Filed under post-training › reinforcement learning algorithm