polish(pu): polish efficiency and performance on atari and dmc

opendilab / LightZero

[NeurIPS 2023 Spotlight] LightZero: A Unified Benchmark for Monte Carlo Tree Search in General Sequential Decision Scenarios (awesome MCTS)

Apache License 2.0

1.15k stars 120 forks source link

Closed puyuan1996 closed 5 days ago

puyuan1996 commented 2 weeks ago

Optimize efficiency, specifically by replacing deepcopy with a predefined share_kv_cache_pool.
Add muzero_segmentcollector, enable the buffer reanalyze option, and add the train***_segment.py entry point.
Refine the configurations for Atari and DeepMind Control environments.