Elamur Groupantsgparse-test-training-speed-bounded-check
The groupant beside gparse/tests/training_speed.rs in the Mova repository: what the code computes, what it never does, how to go back from its outputs, its invariants, cost and checks.
Facts
Implementation material. A Mova code model describes this code at one source revision: partial models, finite checks and their stated scope. It is not a reviewed general exposition, and its labels are shown as written, never raised.
Components
| component | label | where |
|---|---|---|
| approximant | partial | approximant |
| ast | partial | ../syntax.json |
| cost | symbolic source derived | invariant |
| descriptant | present | descriptant |
| invariant | partial | invariant |
| tests | partial bounded checks | tests |
Obernants
The typed relations around the forward model: the behaviour complement (what the code never does), the reverse reconstruction (every input compatible with an output) and specialised views.
| id | type | label | where |
|---|---|---|---|
behavior_complement | behavior_complement | partial source derived model | behavior_complement |
reverse_reconstruction | reverse_reconstruction | partial source derived model | reverse_reconstruction |
head_transport | head_transport | partial source derived model | head_transport |
loss_weights | loss_weights | partial source derived model | loss_weights |
mutation_order | mutation_order | partial source derived model | mutation_order |
observation_scope | observation_scope | partial source derived model | observation_scope |
Model text
descriptant
Source-bound description
step forwards train=true, gathers real arc rows and local-to-flat gold heads, constructs arc/relation/tag losses, differentiates joint loss, optionally validates four gradient norms, updates AdamW, returns pre-update scalar loss.
Epoch 2026-10-31
Prospective model was recorded before edits. Added tests use bounded synthetic fixtures; production algorithms unchanged. Coverage is partial, not proof. No trained corpus, live teacher or throughput experiment was executed.
approximant
Forward approximant
step forwards train=true, gathers real arc rows and local-to-flat gold heads, constructs arc/relation/tag losses, differentiates joint loss, optionally validates four gradient norms, updates AdamW, returns pre-update scalar loss.
Domain: valid source-specific dimensions and resources stated in invariant; no automatic production substitution. Numerical observations are finite checks, not an end-to-end relative bound. Executable Rust surrogate: unsupported gap; existing exact tests are not surrogates. Future metric-specific caps may be 1/2/5/10%, never above 10%.
invariant
Invariant and cost
Flattened gold head is r-r%T+h; loss=La+Lr+0.5*(Lx+Lu). Inspect assertions precede opt.step.
Cost
One forward/backward plus O(parameters) optimizer update; storage includes gradients/moments and batch tensors.
Allocation, finite arithmetic, platform dimensions and external failures are not universally certified.
models
Typed model catalog
behavior_complement
Missing or nonpositive/nonfinite inspected gradient prevents optimizer update; loss return is not post-step reevaluation.
reverse_reconstruction
Loss admits many logits with equal cross entropy; post-step parameter values do not reconstruct absent AdamW moments.
head_transport
Subtract row remainder to get batch offset before adding gold head.
loss_weights
Both tag losses get weight0.5; arc and relation weight1.
mutation_order
gradients=total.backward then optional inspect then opt.step; norm failure has no update effect.
observation_scope
Return total.to_scalar after update reads existing pre-update loss tensor; it does not run network again.
tests
Checks and limits
Selected parent checks relevant to this fragment (not exhaustive): synthetic_single_optimizer_step
Selected commands run twice debug and twice release; exact receipts and source-to-test mapping retained in wave30 gparse external run directory. Tests do not certify allocation failure, all NaN/overflow inputs, trained-model accuracy or corpus timing. Compiled mutants cover selected boundaries; per-source coverage is recorded in the external map, not presumed.
This page in Markdown → https://md.groupants.elamur.ai/free/en/mova/gparse/tests/training_speed.rs.groupant/bounded-check.groupant/