parallel-actor-critic
Guides minimalRL A2C and A3C multiprocessing actor-critic workflows, Gym API migration, and safe process debugging.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- parallel-actor-critic
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guides minimalRL A2C and A3C multiprocessing actor-critic workflows, Gym API migration, and safe process debugging.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md2.9 KB
- 参考文档references/troubleshooting.md2.6 KB
- 参考文档references/workflows.md2.6 KB
- 脚本scripts/smoke_parallel_actor_critic.py3.0 KB
03 SKILL.md 正文
Parallel Actor-Critic Workflows
Use this sub-skill when a task asks about minimalRL's A2C or A3C scripts, especially synchronous vector rollouts, asynchronous workers, shared models, process lifecycle, or Gym API modernization.
Route by task
- A2C / synchronous vector rollout: read references/api-reference.md for
ParallelEnv,worker, andcompute_target; read references/workflows.md for the update sequence. - A3C / asynchronous global-local training: use references/workflows.md for shared model and local gradient flow, then references/troubleshooting.md for worker hangs or missing gradients.
- Gym 0.26 migration: read the migration section in references/workflows.md before running either full native script.
- Safe preflight checks: run scripts/smoke_parallel_actor_critic.py, which avoids full training and unbounded process spawning.
Do not use this sub-skill for
- Single-process REINFORCE, actor-critic, PPO, PPO-LSTM, or V-trace network/update basics; use ../on-policy-discrete/SKILL.md.
- DQN, ACER, or replay/off-policy correction; use ../off-policy-value/SKILL.md.
- DDPG, continuous PPO, or SAC; use ../continuous-control/SKILL.md.
Quick operating workflow
-
Decide whether the user wants A2C's synchronous vectorized rollout or A3C's asynchronous worker updates.
-
Check references/api-reference.md for exact actor-critic, worker, and target-return contracts.
-
If using modern Gym, apply the reset/step/seed migration in references/workflows.md before running full training.
-
Run a bounded smoke check:
python sub-skills/parallel-actor-critic/scripts/smoke_parallel_actor_critic.py --check all -
For deadlocks, tuple arity failures, gradient-copy bugs, or CPU oversubscription, read references/troubleshooting.md.
Bundled runtime files
- references/api-reference.md records A2C/A3C classes, functions, hyperparameters, worker protocol, and tensor shapes.
- references/workflows.md explains synchronous and asynchronous actor-critic workflows plus Gym migration.
- references/troubleshooting.md gives conservative process-debugging guidance.
- scripts/smoke_parallel_actor_critic.py validates model, target-return, and bounded protocol contracts without full training.
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).