waymax-rl
Operate PARL's optional all-GPU Waymax-RL autonomous-driving workflow, validate Hydra configs, and handle JAX/Waymax data and GPU prerequisites.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- waymax-rl
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Operate PARL's optional all-GPU Waymax-RL autonomous-driving workflow, validate Hydra configs, and handle JAX/Waymax data and GPU prerequisites.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/configuration.md5.3 KB
- 参考文档references/setup-and-training.md5.4 KB
- 参考文档references/troubleshooting.md5.9 KB
- 脚本scripts/validate_waymax_config.py13.5 KB
03 SKILL.md 正文
Waymax-RL
Use this sub-skill when a task involves PARL's optional Waymax-RL autonomous-driving workflow, Waymax simulator integration, all-GPU RL training, Hydra ppo_config files, JAX CUDA setup, Waymo Open Dataset TFRecord paths, or rl-games/PyTorch runner settings for the Waymax environment.
Verification boundary
This Waymax-RL guidance is source-distilled only. The workflow was not runtime-verified during this skill production run because it requires a CUDA-capable GPU, a matching JAX CUDA install, a Waymax checkout at the required revision, and local Waymo-format data. Treat every runtime claim as a checklist to verify in the user's environment before training.
Safety gate
Do not start training until all of these are true:
- A CUDA GPU is available to JAX, not just to the OS or to PyTorch.
- Waymax is installed at the required revision and imports cleanly.
- The config's
params.config.env_config.backendisgpu. - The config's
data_cfg.data_pathpoints to a real TFRecord file or a directory containing the intended files; the default placeholder path is replaced. - The user explicitly accepts that this is an all-GPU loop with no CPU substitute and that training can be long-running and memory-intensive.
Run the bundled static validator before any launch:
python scripts/validate_waymax_config.py <path-to-hydra-yaml>
The validator is intentionally static: it does not import JAX, Waymax, TensorFlow, Torch, Hydra, or rl-games, and it never starts training.
Route map
- Read
references/setup-and-training.mdfor the installation sequence, runtime architecture, training entry point, and GPU/data readiness checklist. - Read
references/configuration.mdfor the Hydra config schema, important defaults, shape/path implications, and safe edits. - Read
references/troubleshooting.mdfor CUDA JAX failures, missing Waymax revision, TensorFlow conflicts, placeholder data paths, Colab/local differences, and memory issues. - Use
scripts/validate_waymax_config.pyto statically inspect a config and surface placeholder paths, backend expectations, andmixed_precision/multi_gpuflags. - Load sibling
xparl-distributedonly to contrast PARL's CPU/xparl distributed actor model with this optional Waymax all-GPU loop. Do not use xparl to replace Waymax-RL's GPU-resident simulator/trainer path.
Non-goals
This sub-skill does not install CUDA, download Waymo data, clone external repositories, run Colab notebooks, verify a GPU, tune a production driving policy, or claim vehicle-readiness. It only captures the repo-local Waymax-RL operating workflow and the preflight checks needed before a user deliberately runs that 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).