TokenPilot Experiments
July 5, 2026 ยท View on GitHub
This directory is the experiment hub for the current TokenPilot component in LightMem2.
Use the root README.md first if you still need to:
- install the repository
- install the component
- verify the runtime path in a real session
Use ../README.md for the top-level experiment entry across the repository.
Benchmarks
The current public experiment surface includes:
| Benchmark | Scope | Docs |
|---|---|---|
PinchBench | runtime cost and quality evaluation in isolated and continuous session modes | pinchbench/README.md |
Claw-Eval | OpenClaw task execution evaluation in isolated and continuous session modes | claw-eval/README.md |
Recommended Workflow
- Finish the installation and quick-start flow in the root README.md.
- Confirm the current TokenPilot runtime path works in a real session on the host you plan to use for the benchmark.
- Pick the benchmark you want from the table above.
- Follow that benchmark README for assets, environment setup, and official commands.
Notes
- This directory is a landing page, not a duplicate of the benchmark manuals.
- Large benchmark datasets and experiment outputs are stored outside Git.
- Use experiments/README.md for the shared Google Drive layout and external storage policy.
- Exact benchmark scripts, assets, and caveats stay inside each benchmark subtree.