Skip to content

Cooked#8

Open
RookieJoel wants to merge 4 commits into
sixteencharacter:mainfrom
RookieJoel:main
Open

Cooked#8
RookieJoel wants to merge 4 commits into
sixteencharacter:mainfrom
RookieJoel:main

Conversation

@RookieJoel

Copy link
Copy Markdown
Contributor

No description provided.

- Added STARFORMER optimization branch in training.py, including TD loss computation and logging.
- Introduced _compute_episode_rtg function for discounted return-to-go calculation.
- Enhanced train function to support STARFORMER context management and episode tracking.
- Updated select_action in utils.py to handle STARFORMER-specific action selection logic.
- Created AGENTS.md for documentation on project structure and key files.
- Added SKILL.md to outline coding patterns and conventions used in the project.
- Introduced unit tests for Transition dataclass and StarformerQNet in tests/test_datamodel.py and tests/test_starformer_net.py.
- Implemented tests for WindowPrioritizedReplayMemory in tests/test_window_replay.py to validate episode boundary and backfill logic.
- Added STARFORMER architecture integration with PPO in training.py.
- Introduced windowed tensor building and action selection for STARFORMER.
- Enhanced PPO optimization with AMP support and GAE computation.
- Updated action selection logic in utils.py to accommodate STARFORMER.
- Created new tests for STARFORMER network and PPO configurations.
- Added documentation for agents and coding patterns in AGENTS.md and SKILL.md.
- Established a test suite for datamodel and replay memory functionalities.
@Jirapat-K-F

Copy link
Copy Markdown
Contributor

cannot use PPO with random sampling it force to use uniform

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants