Language-Conditioned Tasks on BabyAIGoToRedBall
0.92Mean Episodic ReturnFP16-PPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FP16-PPOPrecision=FP162026.04 | 0.92 | — | |
| INT8-PPOPrecision=INT82026.04 | 0.9 | — | |
| BitRL-1bitPrecision=1-bit2026.04 | 0.88 | 95.7 | |
| INT4-PPOPrecision=INT42026.04 | 0.87 | — | |
| MLP-PPOArchitecture=MLP2026.04 | 0.65 | — |