Reward-level distillation and FP16 quantization compress a 317M-parameter TD-MPC2 agent to 1M parameters, reaching 28.45 normalized score on MT30, though most of the gap over the original 18.93 comes from a longer training budget.
Model-agnostic meta-learning for fast adaptation of deep networks
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
TD-MPC-Opt: Distilling Model-Based Multi-Task Reinforcement Learning Agents
Reward-level distillation and FP16 quantization compress a 317M-parameter TD-MPC2 agent to 1M parameters, reaching 28.45 normalized score on MT30, though most of the gap over the original 18.93 comes from a longer training budget.