runner-and-cli-workflows
Use Tensorforce Runner training/evaluation workflows, CLI-equivalent semantics, logging, parallel execution, and optional BOHB tuning.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- runner-and-cli-workflows
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use Tensorforce Runner training/evaluation workflows, CLI-equivalent semantics, logging, parallel execution, and optional BOHB tuning.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/cli-reference.md4.7 KB
- 参考文档references/runner-workflows.md7.8 KB
- 参考文档references/troubleshooting.md3.1 KB
- 参考文档references/tuning.md2.5 KB
- 脚本scripts/quickstart_cartpole_smoke.py4.4 KB
- 脚本scripts/tensorforce_runner_smoke.py9.4 KB
03 SKILL.md 正文
Runner and CLI Workflows
Use this sub-skill when the task is to run bounded Tensorforce training or evaluation through tensorforce.Runner, translate the historical runner CLI flags into public API calls, configure local/remote parallel execution, collect callback/logging metrics, or reason about optional BOHB hyperparameter tuning.
Route map
- Start with
references/runner-workflows.mdfor the publicRunnerconstructor/run signatures, safe train/evaluate recipes, callbacks, stopping criteria, and parallel modes. - Use
references/cli-reference.mdwhen a user provides command-line style flags such as--agent,--environment,--episodes,--num-parallel,--remote,--path, or--save-best-agent; translate them to a self-contained Python wrapper instead of relying on a repository script. - Use
references/tuning.mdonly for optional BOHB/Hyperband tuning. Treat tuning dependencies and long optimization runs as optional, not part of the minimum Tensorforce runtime. - Use
references/troubleshooting.mdfor non-terminating runs, evaluation placement errors, parallel/socket mistakes, missingmax_episode_timesteps, logging extras, or dependency-version issues surfaced while running workflows. - Run
scripts/quickstart_cartpole_smoke.py --helpfor a compact CartPole quickstart smoke, orscripts/tensorforce_runner_smoke.py --helpfor a CLI-like Runner smoke. Both scripts use the installed Tensorforce package and do not call any repository-level CLI script.
Boundaries
- For agent aliases, state/action specifications, action masking, manual
act/observe,experience, orupdateloops, useagents-and-specifications. - For network, objective, optimizer, memory, policy, preprocessing, and JSON/dict module configuration details, use
modules-and-configuration. - For custom
Environmentsubclasses, Gym/adapter setup, reward shaping, vectorized/multi-actor environment contracts, or simulator backends, useenvironments-and-interaction. - For checkpoints, TensorBoard summaries, recorder/pretraining,
Agent.save,Agent.load, exported SavedModels, or best-agent persistence details, usepersistence-export-and-recordingafter this sub-skill has established the runner workflow.
Minimum safe pattern
from tensorforce import Runner
runner = Runner(
agent='random',
environment=dict(environment='custom_cartpole'),
max_episode_timesteps=10
)
try:
runner.run(num_episodes=3, use_tqdm=False)
print(runner.episode_returns)
finally:
runner.close()
Always pass at least one bounded stopping criterion (num_episodes, num_timesteps, or num_updates). Prefer use_tqdm=False for automation and CI logs.
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).