Target
Increase conditional capacity while keeping active compute bounded.
Failure this recipe addresses
A dense model lacks capacity and a true sparse execution path is available.
Data contract
A deterministic routing fixture plus task data and expert-load diagnostics.
Method or policy
Train a router and experts, then execute only selected experts with sparse gather/scatter or grouped matmul.
Evaluation contract
Task quality and real wall-clock/energy versus a dense parameter-matched baseline.
- expert balance
- router collapse
- memory
- wall time
- quality
Budget and stop rule
Dense-compute smoke completed; no further project run without a real sparse kernel and a fresh target.
Stop if all experts still execute or the router collapses.
Decision rule
Treat as capacity-only until sparse execution produces a measured compute win.
Learning exercise
Calculate active versus total parameters for top-k expert routing and identify why dense execution saves nothing.
Explain the difference between sparse parameters and sparse compute.
How to read this dossier
This canonical page separates the retained repository decision from the material that informed it. A source may describe an external claim, historical measurement, local observation, or planned procedure; those evidence classes are not interchangeable. Follow the provenance links before reusing a number or method.
Status and disposition describe what PostTrainLLM retained when this record was indexed. They are not a live product promise, a newly run benchmark, or permission to restart historical work. Unknown or unmeasured fields remain unknown rather than being treated as zero.
To reuse the record, first name the exact claim you need and trace it to the linked source. Then check whether the original environment, model revision, data split, evaluator, hardware, and budget match the proposed use. If they do not, treat the record as a hypothesis or design reference and run the smallest fresh comparison that can falsify it. Preserve negative outcomes and regressions beside any improvement; a local win on one slice does not silently become a general capability claim.
Source provenance
The normalized record comes from docs/recipes/registry.json. The links below are the tracked evidence and explanatory sources preserved with the record.