policy-training
Routes ACT, CNNMLP, Diffusion Policy, and latent-model training or evaluation workflows for ACT++ checkpoints and datasets.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- policy-training
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Routes ACT, CNNMLP, Diffusion Policy, and latent-model training or evaluation workflows for ACT++ checkpoints and datasets.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/model-overview.md2.3 KB
- 参考文档references/troubleshooting.md2.4 KB
- 参考文档references/workflows.md3.7 KB
- 脚本scripts/check_policy_stack.py2.4 KB
03 SKILL.md 正文
policy-training
Use this sub-skill when the task is about training or evaluating ACT++, CNNMLP, Diffusion Policy, or the VQ latent model that follows ACT training.
Typical triggers
- "Train ACT on sim_transfer_cube_scripted"
- "Evaluate the best checkpoint"
- "Why does policy.py fail to import?"
- "What does --temporal_agg do?"
- "How do I run the latent model training pass?"
What this sub-skill covers
- Step-based training and evaluation through the
imitate_episodesworkflow. - ACT, CNNMLP, and Diffusion policy wrappers.
- The DETR-derived backbone / transformer / latent model internals that drive those wrappers.
- Dataset loading, normalization, and evaluation statistics.
- VQ latent model training from a previously trained ACT checkpoint.
What it excludes
- Simulated data generation, replay, and rendering -> simulation-data.
- VINN feature cache and k-selection -> vinn-offline.
- Real robot deployment or ROS/servo control -> root troubleshooting only.
- Experimental actuator-network training with hard-coded paths -> reference-only note only.
Read these first
Run this helper first
Before trying to train, use check_policy_stack.py to confirm the repo checkout imports, the CUDA backend is visible, and the policy wrapper surface is present.
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).