---
title: "gparse-cli-training — Mova code model"
description: "Mova code model of gparse/src/main.rs: components with their labels, obernants and the model text."
url: "https://groupants.elamur.ai/free/en/mova/gparse/src/main.rs.groupant/training.groupant/"
---
Elamur: [Mova](https://md.mova.elamur.ai/free/en/) · [Dictionary](https://md.dictionary.elamur.ai/free/en/) · [Research](https://md.research.elamur.ai/free/en/) · **Groupants**

This page as HTML: <https://groupants.elamur.ai/free/en/mova/gparse/src/main.rs.groupant/training.groupant/>

# gparse-cli-training

The groupant beside gparse/src/main.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

- **kind:** groupant_fragment
- **status:** partial source bound model
- **source:** [`gparse/src/main.rs`](https://github.com/ElamurAI/mova-free/blob/main/gparse/src/main.rs)
- **source SHA-256:** `e42b84f188b56efd…`
- **covers:** Training composes filtered data, optional initialization, sampled plans, joint cross entropy and AdamW; saves strict best full-dev mean. (bytes 3606–16378)
- **part of:** [gparse-cli](https://md.groupants.elamur.ai/free/en/mova/gparse/src/main.rs.groupant/)
- **repository folder:** [`gparse/src/main.rs.groupant/training.groupant`](https://github.com/ElamurAI/mova-free/tree/main/gparse/src/main.rs.groupant/training.groupant)
- **repository commit:** `48f2e6f74146`

> **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](#approximant) |
| **ast** | partial | `../syntax.json` |
| **cost** | symbolic source derived | [invariant](#invariant) |
| **descriptant** | present | [descriptant](#descriptant) |
| **invariant** | partial | [invariant](#invariant) |
| **tests** | unexecuted source derived scope | [tests](#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](#behavior-complement) |
| `reverse_reconstruction` | reverse_reconstruction | partial source derived model | [reverse_reconstruction](#reverse-reconstruction) |
| `schedule` | schedule | partial source derived model | [schedule](#schedule) |
| `sampling` | sampling | partial source derived model | [sampling](#sampling) |
| `checkpoint` | checkpoint | partial source derived model | [checkpoint](#checkpoint) |
| `partial_effect` | partial_effect | partial source derived model | [partial_effect](#partial-effect) |

## Model text


### descriptant

#### Source-bound description

Training composes filtered data, optional initialization, sampled plans, joint cross entropy and AdamW; saves strict best full-dev mean.

##### 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.

##### Epoch 2026-11-04

The loss now runs over keep()'s rows: arc loss on rows with VoteHead ≥ GP_PA, relation loss on rows with Vote ≥ GP_PA (all rows without GP_PA), tags on all rows; optional soft arc targets (GP_SOFT) and a random-subset control (GP_PA_CONTROL=random, its own xorshift seeded from GP_SEED). Batches come from batch_full with the model's ngram. A new model may take the subword input (GP_SUBWORD=1).

### approximant

#### Forward approximant

Training composes filtered data, optional initialization, sampled plans, joint cross entropy and AdamW; saves strict best full-dev mean.

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

Quick probes do not select checkpoints; output files can be partially updated on later failure.

##### Cost

Epochs times sampled batch network/backward/AdamW plus dev inference; O(parameters+corpus+batch graph) memory.

Allocation, finite arithmetic, platform dimensions and external failures are not universally certified.

### models

## Typed model catalog

### behavior_complement

Quick-dev proxy cannot select checkpoint when dev sets exist.

### reverse_reconstruction

Saved weights alone omit optimizer moments, sample RNG and checkpoint trajectory; antecedents are nonunique.

### schedule

warm=min(est,100 init else2000); total=est*epochs; lr floors decay at0.02.

### sampling

Always sentences included; others sample each epoch, first plan reused for epoch1.

### checkpoint

Strict mean>best and full_dev/no-dev guard; last epoch always full.

### partial_effect

Creates metadata before epochs, saves weights on improved epoch and rewrites log afterward; no transaction.

### tests

#### Checks and limits

This exact fragment remains source-derived and unexecuted in this epoch. Parent/sibling selected evidence, not runtime coverage of this fragment: wave30::score_zip_and_plan_failure_precedence; train/eval/annotate external effects remain unexecuted

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.