Module config 

Declarative YAML configuration

This module provides Ludwig-style declarative training configuration via YAML.

Example

model:
  path: base-model.gguf
  layers: [q_proj, v_proj]

data:
  train: train.parquet
  batch_size: 8

optimizer:
  name: adam
  lr: 1e-4

lora:
  rank: 64
  alpha: 16

Structs

ArchitectureOverrides

Architecture override parameters from YAML manifest.

AuditArgs

Arguments for audit command

BenchArgs

Arguments for bench command

BundleArgs

Arguments for bundle command

CiteArgs

Arguments for cite command

Cli

Entrenar: Training & Optimization Library

ColumnStats

Statistics about a column used for type inference

CompletionArgs

Arguments for completion command

CurriculumStage

A curriculum learning stage: a data source active until a given step.

DataConfig

Data configuration

DepositArgs

Arguments for deposit command

ExperimentsArgs

Arguments for the experiments command

ExportArgs

Arguments for export command

FinetuneArgs

Arguments for the finetune command (plan/apply classification training)

InferenceConfig

Configuration for type inference

InferredSchema

Inferred schema for a dataset

InfoArgs

Arguments for the info command

InitArgs

Arguments for the init command

InspectArgs

Arguments for inspect command

LoRASpec

LoRA configuration

MergeArgs

Arguments for the merge command

MergeSpec

Model merging configuration

ModelRef

Model reference and target layers

MonitorArgs

Arguments for monitor command

OptimSpec

Optimizer specification

PreregisterArgs

Arguments for preregister command

PublishArgs

Arguments for the publish command

PublishSpec

Auto-publish configuration for uploading to HuggingFace Hub after training.

QuantSpec

Quantization configuration

QuantizeArgs

Arguments for the quantize command

ResearchArgs

Arguments for the research command

ResearchInitArgs

Arguments for research init command

TrainArgs

Arguments for the train command

TrainSpec

Complete training specification

TrainingParams

Training hyperparameters

ValidateArgs

Arguments for the validate command

VerifyArgs

Arguments for verify command

Enums

ArchiveProviderArg

Archive provider for CLI

ArtifactTypeArg

Artifact type for CLI

AuditType

Audit type

CitationFormat

Citation format for CLI

Command

Available commands

ExperimentsCommand

Experiment store subcommands

ExportFormat

Export format for CLI

FeatureType

Inferred feature type

FinetuneCommand

Finetune subcommands (forjar-style plan/apply)

InitTemplate

Init template type

InspectMode

Inspection mode

LicenseArg

License for CLI

MergeMethod

Merge method

ModelMode

Model execution mode

OutputFormat

Output format for info command

QuantMethod

Quantization method

ResearchCommand

Research subcommands

ShellType

Shell type for completions

TrainingMethod

Training method for –method flag

TrainingMode

Training loss mode

ValidationError

Validation error type

Functions

apply_overrides

Apply command-line overrides to a TrainSpec

build_model

Build a model from configuration by loading from file

build_optimizer

Build optimizer from configuration

collect_stats_from_samples

Collect statistics from sample values (simplified in-memory analysis)

infer_schema

Infer schema from column statistics

infer_schema_from_path

Placeholder: Load stats from Parquet file Real implementation would use arrow-rs/parquet crate

infer_type

Infer feature type from column statistics

is_hf_repo_id

Check if a string looks like a HuggingFace repo ID.

load_config

Load training spec from YAML file (without running training)

parse_args

Parse CLI arguments from a string slice (for testing)

train_from_yaml

Train a model from YAML configuration file

try_load_apr_for_inference

ALB-096: Try to load a model from APR format.

validate_config

Validate a training specification