How ales27pm/quec-fr-ca-llm-training- works
Quec-fr-CA-llm-training is a data and evaluation pipeline for building Québécois French training datasets and checking their release readiness. Users invoke the qfr CLI to validate source and policy manifests, ingest and curate corpus material, split accepted records, and build/export a training pack. Separate branches acquire modern sources and generate minimal pairs; evaluation diagnostics and release gates provide quality controls. The graph reflects documented CLI capabilities and sampled imports/calls; some unsampled implementations are represented only at their exact module paths, so uncertain connections are omitted.
Main entry points: cli.py qfr commands · release_candidate.py Release candidate · schemas.py Domain schemas · io.py File I/O
Components 26
CLI and orchestration
- qfr commands
src/qfr_pipeline/cli.py - Release candidate
src/qfr_pipeline/release_candidate.py
Corpus acquisition
- Corpus ingestion
src/qfr_pipeline/corpus_sources.py - Modern acquisition
src/qfr_pipeline/modern_corpus.py - Corpus curation
src/qfr_pipeline/curation_policy.py - Corpus readiness
src/qfr_pipeline/corpus_readiness.py
Dataset preparation
- Curated splitting
src/qfr_pipeline/curated_split.py - Training pack
src/qfr_pipeline/training_pack.py - Training export
src/qfr_pipeline/training_export.py - Minimal pairs
src/qfr_pipeline/minimal_pairs.py - Pair quality
src/qfr_pipeline/minimal_pair_quality.py - LP9 micro-pack
src/qfr_pipeline/lp9_micro_pack.py - Legacy pipeline
src/qfr_pipeline/data_pipeline.py - LP9 failure mining
src/qfr_pipeline/lp9_failure_mining.py
Evaluation and release
- LP9 evaluation
src/qfr_pipeline/lp9_eval.py - Eval diagnostics
src/qfr_pipeline/diagnostics.py - Legacy diagnostics
src/qfr_pipeline/legacy_diagnostics.py - Contamination checks
src/qfr_pipeline/contamination.py - Release report
src/qfr_pipeline/release_report.py - LP7 monitor
src/qfr_pipeline/lp7_monitor.py
Shared runtime
- Manifest validation
src/qfr_pipeline/validation.py - Domain schemas
src/qfr_pipeline/schemas.py - File I/O
src/qfr_pipeline/io.py - Repository paths
src/qfr_pipeline/paths.py
Other components
- Pipeline operator
- External corpora
Connections 48
- Pipeline operator → qfr commands: invokes
README.md - qfr commands → Release candidate: runs
src/qfr_pipeline/cli.py - qfr commands → Corpus ingestion: ingests
src/qfr_pipeline/cli.py - qfr commands → Modern acquisition: acquires
README.md - qfr commands → Corpus curation: curates
src/qfr_pipeline/cli.py - qfr commands → Curated splitting: splits
README.md - qfr commands → Training pack: builds
README.md - qfr commands → Training export: exports
src/qfr_pipeline/cli.py - qfr commands → Minimal pairs: generates
src/qfr_pipeline/cli.py - qfr commands → Pair quality: validates
src/qfr_pipeline/cli.py - qfr commands → LP9 evaluation: generates
src/qfr_pipeline/cli.py - qfr commands → Eval diagnostics: diagnoses
src/qfr_pipeline/cli.py - qfr commands → Legacy diagnostics: diagnoses
README.md - qfr commands → Contamination checks: checks
src/qfr_pipeline/cli.py - qfr commands → Release report: reports
src/qfr_pipeline/cli.py - qfr commands → Manifest validation: validates
src/qfr_pipeline/cli.py - qfr commands → LP9 micro-pack: builds
src/qfr_pipeline/cli.py - qfr commands → LP9 failure mining: mines
src/qfr_pipeline/cli.py - qfr commands → Legacy pipeline: runs
src/qfr_pipeline/cli.py - qfr commands → LP7 monitor: monitors
src/qfr_pipeline/cli.py - Release candidate → Corpus ingestion: ingests
src/qfr_pipeline/release_candidate.py - Release candidate → Corpus readiness: checks
src/qfr_pipeline/release_candidate.py - Release candidate → Corpus curation: curates
src/qfr_pipeline/release_candidate.py - Release candidate → Curated splitting: splits
src/qfr_pipeline/release_candidate.py - Release candidate → Eval diagnostics: summarizes
src/qfr_pipeline/release_candidate.py - Release candidate → Minimal pairs: summarizes
src/qfr_pipeline/release_candidate.py - Release candidate → Training pack: builds
src/qfr_pipeline/release_candidate.py - Release candidate → Training export: exports
src/qfr_pipeline/release_candidate.py - Release candidate → Release report: reports
src/qfr_pipeline/release_candidate.py - Corpus ingestion → File I/O: uses
src/qfr_pipeline/corpus_sources.py - Corpus ingestion → Domain schemas: uses
src/qfr_pipeline/corpus_sources.py - Corpus ingestion → Repository paths: resolves via
src/qfr_pipeline/corpus_sources.py - Corpus curation → Domain schemas: uses
src/qfr_pipeline/curation_policy.py - Corpus curation → File I/O: uses
src/qfr_pipeline/curation_policy.py - Corpus curation → Repository paths: resolves via
src/qfr_pipeline/curation_policy.py - Corpus readiness → File I/O: uses
src/qfr_pipeline/corpus_readiness.py - Corpus readiness → Repository paths: resolves via
src/qfr_pipeline/corpus_readiness.py - Curated splitting → File I/O: uses
src/qfr_pipeline/curated_split.py - Curated splitting → Domain schemas: uses
src/qfr_pipeline/curated_split.py - Curated splitting → Repository paths: resolves via
src/qfr_pipeline/curated_split.py - Training pack → File I/O: uses
src/qfr_pipeline/training_pack.py - Training pack → Domain schemas: uses
src/qfr_pipeline/training_pack.py - Training export → File I/O: uses
src/qfr_pipeline/training_export.py - Training export → Domain schemas: uses
src/qfr_pipeline/training_export.py - Minimal pairs → Domain schemas: uses
src/qfr_pipeline/minimal_pairs.py - Minimal pairs → File I/O: uses
src/qfr_pipeline/minimal_pairs.py - Pair quality → Minimal pairs: validates
src/qfr_pipeline/minimal_pair_quality.py - Pair quality → Domain schemas: uses
src/qfr_pipeline/minimal_pair_quality.py