KoalaTrapLord
1f3b17d173
ForgeEpisodeQueue does not exist in rollout_buffer.py yet. Removing premature import to unblock supervised training pipeline.
29 lines
873 B
Python
29 lines
873 B
Python
"""
|
|
Commander AI Lab - Training Subpackage
|
|
==========================================
|
|
Provides RL training infrastructure including PPO, rollout buffers,
|
|
Forge episode generation, self-play, and distillation.
|
|
"""
|
|
|
|
from ml.training.rollout_buffer import RolloutBuffer
|
|
from ml.training.reward import compute_intermediate_reward, compute_terminal_reward, RewardConfig
|
|
from ml.training.decision_exporter import DecisionExporter
|
|
from ml.training.forge_episode_generator import (
|
|
ForgeEpisodeGenerator,
|
|
AsyncForgeEpisodeGenerator,
|
|
)
|
|
|
|
__all__ = [
|
|
# Rollout & Buffers
|
|
"RolloutBuffer",
|
|
# Rewards
|
|
"compute_intermediate_reward",
|
|
"compute_terminal_reward",
|
|
"RewardConfig",
|
|
# Decision export
|
|
"DecisionExporter",
|
|
# Forge episode generation (Phase 3)
|
|
"ForgeEpisodeGenerator",
|
|
"AsyncForgeEpisodeGenerator",
|
|
]
|