| # LifeStack documentation |
|
|
| All project documentation lives under **`docs/`** (except the repository **[README.md](../README.md)** at the root, which is the HuggingFace Space and GitHub entry point). |
|
|
| **Repository:** [https://github.com/oki-dokii/Meta-R2](https://github.com/oki-dokii/Meta-R2) |
|
|
| --- |
|
|
| ## Quick links |
|
|
| | Doc | Use | |
| |-----|-----| |
| | [training_guide.md](training_guide.md) | Install, train, resume, push to Hub | |
| | [train_trl.md](train_trl.md) | `scripts/train_trl.py` CLI reference | |
| | [reward.md](reward.md) | All 10 GRPO reward functions and anti-hacking design | |
| | [model_card.md](model_card.md) | Hugging Face model card (post to Hub) | |
| | [blog.md](blog.md) | Engineering retrospective / HF blog draft | |
|
|
| --- |
|
|
| ## Core system |
|
|
| | Doc | Code | Description | |
| |-----|------|-------------| |
| | [lifestack_env.md](lifestack_env.md) | `core/lifestack_env.py` | `LifeStackEnv`, `WorldEngine`, `reset/step/rollout`, OpenEnv shim | |
| | [reward.md](reward.md) | `core/reward.py`, `scripts/train_trl.py` | `compute_reward`, `compute_task_reward`, 10 GRPO reward heads | |
| | [task.md](task.md) | `core/task.py` | `Task`, `Route`, `Milestone`, `ExoEvent` dataclasses | |
| | [memory.md](memory.md) | `agent/memory.py` | `LifeStackMemory`, ChromaDB collections, few-shot injection | |
| | [conflict_generator.md](conflict_generator.md) | `agent/conflict_generator.py` | Crisis scenario templates, `TaskGenerator`, 8 domains | |
|
|
| --- |
|
|
| ## Training & configuration |
|
|
| | Doc | Description | |
| |-----|-------------| |
| | [training_guide.md](training_guide.md) | End-to-end guide: Colab, server, resume, common errors | |
| | [train_trl.md](train_trl.md) | `LifeStackGRPOTrainer`, JSON-boundary masking, 5-stage curriculum | |
| | [configuration.md](configuration.md) | `GRPOConfig` via `_make_grpo_config`, reward weights per stage | |
| | [kaggle_train.md](kaggle_train.md) | Notebook-specific setup (T4, session timeouts, OOM tips) | |
|
|
| --- |
|
|
| ## Scripts & evaluation |
|
|
| | Doc | Description | |
| |-----|-------------| |
| | [eval.md](eval.md) | `scripts/eval.py` random-policy baseline | |
| | [scripts.md](scripts.md) | All scripts under `scripts/` with usage examples | |
|
|
| --- |
|
|
| ## Application & deployment |
|
|
| | Doc | Description | |
| |-----|-------------| |
| | [app.md](app.md) | Flask demo `app_flask.py` on port 7860, routes, `MODEL_REGISTRY` | |
| | [DEPLOYMENT.md](DEPLOYMENT.md) | `start.sh` dual-service, ports 7860 + 8000, env vars, Docker | |
|
|
| --- |
|
|
| ## Meta & pitch |
|
|
| | Doc | Description | |
| |-----|-------------| |
| | [implementation_summary.md](implementation_summary.md) | Engineering summary, training timeline, code map | |
| | [mentor_pitch.md](mentor_pitch.md) | 60–90 second pitch with honest numbers | |
| | [CONTRIBUTING.md](CONTRIBUTING.md) | Doc-first contribution rules, commit format | |
|
|
| --- |
|
|
| ## Maintenance rule |
|
|
| When you add a feature or a new doc file, add a one-line entry in the appropriate table above. See [CONTRIBUTING.md](CONTRIBUTING.md) for the full doc-first rule. |
|
|
| *Last updated: 2026-04-26.* |
|
|
