Contributing — Mens native training
Contributing — Mens training (native)
Section titled “Contributing — Mens training (native)”This guide covers how to contribute to the vox mens train pipeline and how to ensure your code changes feed the MENS training loop correctly.
Read first
Section titled “Read first”Entrypoints
Section titled “Entrypoints”| Surface | Location |
|---|---|
| CLI | vox mens train → crates/vox-cli/src/commands/schola/train/ |
| Library | vox_populi::mens::tensor::run_mens_training (lora_train.rs) |
| Contract | FineTuneContract, ExecutionPlanner, preflight_train |
The Corpus Validate Flow
Section titled “The Corpus Validate Flow”MENS training relies on high-quality .vox examples. When you add or modify code in examples/golden/*.vox, it must pass validation before being ingested.
The flow is:
examples/golden/*.vox(human or agent written)vox corpus validate-batch(CI pipeline check)- Output:
golden_validated.jsonl(ready for training)
What Training-Eligible Code Looks Like
Section titled “What Training-Eligible Code Looks Like”To ensure your code becomes a positive training example:
- Parse Rate: The code must pass the parser 100% cleanly.
vox corpus eval --mode astmust succeed. - Test Blocks: Use
@testblocks to validate logic. The AST coverage and test pass rates will be used in future GRPO reward shaping. - No Stubs: Zero
todo!(),unimplemented!(), or empty function bodies. - Rich Constructs: Use idiomatically correct Vox patterns.
Commands
Section titled “Commands”Run the training planner tests:
cargo check -p vox-populi --features mens-traincargo test -p vox-populi --features mens-train execution_plannerValidate the golden corpus locally:
cargo run -p vox-cli -- corpus eval --mode ast examples/golden/Definition of Done
Section titled “Definition of Done”A PR contributing to the MENS pipeline is “done” when:
cargo test -p vox-populi --features mens-trainis green.- No new parse failures are introduced to the golden corpus.
- TOESTUB reports zero stubs or god-object violations in the touched code.
- Any new CLI flags are documented in the SSOT.
SSOT rule
Section titled “SSOT rule”Candle QLoRA is the active vox mens train backend; keep docs and error messages aligned (lora_train.rs is authoritative when in doubt).