model_id- baobab-ga
model_version- 2026.10.national
phase- research
status- research_slot_shipped
as_of- 2026-10-22T04:42:21.374Z
research
id- baobab-ga-generative-lm
status- research_planned
foundation_llm_weights- false
generative_inference_live- false
ships_trained_baobab_foundation- false
separate_variety_checkpoints_required- true
training_recipe- docs/baobab/GENERATIVE_LM_RECIPE.md
note- Research slot + harness + training recipe only. Do not brand phrase MT or speech corridors as a full national LLM. Generative inference unsupported until gates are met and Lugemi-owned weights exist.
proposed_checkpoint_ids
accra- baobab-ga-accra-lm-*
never_conflate_dangme- true
never_irish_ga- true
prerequisites
- en-gaa-accra corridor + national API surfaces remain the product wedge
- Mix / Fidelity / Live Ga–English evals and rights clear
- Consented parallel + speech data for Accra Ga with held-outs
acceptance_gates
held_out_speakers
id- held_out_speakers
required- true
met- false
description- Held-out evaluation speakers must never appear in training exports; speaker lists remain empty until consented collection.
community_consent_accra_ga
id- community_consent_accra_ga
required- true
met- false
description- trainingEligible stays false until Accra / Ga Mashie community speakers re-confirm for training.
dangme_separation
id- dangme_separation
required- true
met- false
description- Any future LM/adapters must ship as baobab-ga-accra-* — never one blob marketed as Ga+Dangme.
beat_phrase_mt_baselines
id- beat_phrase_mt_baselines
required- true
met- false
description- Generative claims require held-out gains vs corridor phrase-MT baselines on declared Ga-native tasks.
commercial_rights_clear
id- commercial_rights_clear
required- true
met- false
description- Base model + data licenses must clear commercial ministry/enterprise redistribution before promotion.
task_harness
goldens- apps/api/src/baobab/baobab-ga.research.ts#GA_RESEARCH_GOLDENS
runner- apps/api/src/baobab/baobab-ga.research.ts#evaluateBaobabGaNativeTasks
generative_default- unsupported
task_eval_summary
cases- 4
generative_unsupported- 4
corridor_mt_baseline_usable- 2
note- Harness proves generative paths are unsupported and records phrase-MT baselines only.