Target
Improve training signal or support speculative multi-token heads.
Failure this recipe addresses
Single-token supervision or decode throughput is the measured bottleneck.
Data contract
Sequence data with aligned future-token labels and a held-out language/task gate.
Method or policy
Add auxiliary future-token heads with explicit loss weights and verify the base next-token head remains stable.
Evaluation contract
Base-task quality plus accepted speculative tokens per step.
- base-head quality
- extra training cost
- head acceptance
- decode latency
Budget and stop rule
Closed as reference-only; future use requires a throughput-qualified candidate.
Stop if auxiliary loss harms the base head or acceptance cannot offset extra work.
Decision rule
Do not adopt without an end-to-end decode win at quality parity.
Learning exercise
Draw the label shift for two future-token heads and compute the added loss terms.
Explain why training more heads does not guarantee faster decoding.
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.