ppo-pytorch
Routes PPO-PyTorch training, evaluation, and visualization workflows for the repository's PPO implementation.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- ppo-pytorch
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Routes PPO-PyTorch training, evaluation, and visualization workflows for the repository's PPO implementation.
- disable-model-invocation
- true
02 资源文件 8
- 参考文档references/api-reference.md2.9 KB
- 参考文档references/dependencies-and-environments.md2.3 KB
- 参考文档references/repo-provenance.md1.7 KB
- 参考文档references/repo-routing-metadata.json443 B
- 参考文档references/troubleshooting.md2.7 KB
- 参考文档references/workflow-overview.md2.0 KB
- 脚本scripts/check_ppo_setup.py3.9 KB
- 脚本scripts/ppo_core.py9.3 KB
03 SKILL.md 正文
PPO-PyTorch
Use this skill when the user asks about PPO training, pretrained checkpoint evaluation, reward plots, GIF creation, or the repository's shared PPO implementation.
Start here
- Read workflow overview to choose the right route.
- Read dependencies and environments before installing packages or choosing a legacy Gym variant.
- Run
python scripts/check_ppo_setup.py --helpfor a safe import check. - Import the bundled core module from
scripts/ppo_core.pywhen you need the shared PPO classes.
Route map
sub-skills/training/— start or configure PPO training runs, log paths, checkpoints, and action-std schedules.sub-skills/evaluation/— load pretrained checkpoints, evaluate episodes, and diagnose checkpoint mismatches.sub-skills/visualization/— plot reward logs and compose GIFs from saved frames.
Quick install
Install the core scientific stack first, then add the environment package that matches the route you want to use:
python -m pip install torch numpy
python -m pip install pandas matplotlib pillow
python -m pip install gym
Add a legacy environment package only when the chosen route needs it:
roboschoolfor the shipped Roboschool checkpoints and scripts.gym[box2d]or a compatible Box2D package forLunarLander-v2andBipedalWalker-v2.pybulletonly when you intentionally follow the notebook's optional alternative environment stack.
Minimal checks
python scripts/check_ppo_setup.py
python scripts/check_ppo_setup.py --help
For checkpoint inspection, pass --checkpoint-path to the same helper or use the evaluation sub-skill helper.
When to read the provenance file
Read references/repo-provenance.md before deciding whether this skill is current for a checkout or before running a refresh workflow.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).