pql_parallel_topology
Build and validate the process topology for Parallel Q-Learning with actors, replay, policy learning, and value learning.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- pql_parallel_topology
- description
- Build and validate the process topology for Parallel Q-Learning with actors, replay, policy learning, and value learning.
02 资源文件 2
- 脚本scripts/topology.py3.4 KB
- 脚本tests/test_topology.py620 B
03 SKILL.md 正文
PQL Parallel Topology
Use this skill when implementing or auditing a Parallel Q-Learning style recovery or experiment. It is appropriate when an off-policy learner should exploit many parallel simulation actors and separate data collection from policy and value updates. Do not use it for purely on-policy rollouts where learner updates must consume only the latest trajectories.
Inputs
num_envs: positive integer count of parallel actors or environments.replay_capacity: positive integer replay capacity.actor_steps_per_tick: positive integer rollout work per actor tick.policy_updates_per_tickandvalue_updates_per_tick: non-negative update counts.sync_interval: positive integer policy synchronization cadence.
Outputs
- A topology JSON object with process roles, dataflow edges, and invariants.
- Validation errors when the configuration cannot exercise the PQL mechanism.
Workflow
- Create an actor role that writes transition batches to replay.
- Create a replay role that stores off-policy transitions independently of the latest policy.
- Create a value learner role that samples replay and updates Q estimates.
- Create a policy learner role that improves actions using the value learner's Q estimates.
- Add synchronization edges from policy learner to actors and value learner to policy learner.
- Validate that the topology contains all required PQL roles before using it as recovery evidence.
Validation
Run:
python scripts/topology.py --num-envs 8 --replay-capacity 64 --actor-steps-per-tick 2 --policy-updates-per-tick 1 --value-updates-per-tick 2 --sync-interval 4
python tests/test_topology.py
Limitations
This skill defines the topology and invariants; it does not run an RL optimizer by itself. Pair it with a recovery harness or training skill for executable results.
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).