This repository is the deployment system for Internet-in-a-Box (IIAB): an offline-first learning server for schools and communities without reliable Internet. It is an Ansible configuration-management application that installs and configures a complete local network stack — nginx reverse proxy, PHP-FPM, MySQL/PostgreSQL, hostapd WiFi access point, dnsmasq DNS/DHCP, Squid proxy, captive portal, Samba, CUPS — alongside a catalog of educational services such as Kiwix (offline Wikipedia/ZIM), Kolibri, WordPress, MediaWiki, Nextcloud, Gitea, JupyterHub, Moodle, Calibre-Web, and an offline OpenStreetMap/vector-tile mapping system with geocoding. Python, PHP, and Jinja2 templates generate systemd units, nginx virtual hosts, and app configs for every deployed service. Architecture is a staged Ansible pipeline: eleven role groups (0-init through 9-local-addons, then network) are driven from top-level playbooks (iiab-stages.yml, iiab-from-cmdline.yml) invoked by shell entry points (iiab-install, iiab-configure, iiab-network). Every role follows consistent conventions: defaults in defaults/main.yml, an install.yml and enable-or-disable.yml, boolean *_install/*_enabled flags validated up front with Ansible assert tasks, install state recorded in /etc/iiab/iiab_state.yml, and disk usage audited into iiab.ini. A custom Ansible module (get_curl) performs resumable/atomic downloads used across roles. Content provisioning is data-driven: maps and ZIM catalogs are generated (roles/maps/generate-catalog.py) and fetched at install time; offline search and geometry utilities (FQR, static_search.js) are bundled into the served front end. The project is actively developed (Ubuntu 24.04 and Raspberry Pi OS CI matrices, expected-state integration tests) but carries heavy legacy baggage: a large 0-DEPRECATED-ROLES tree, numerous .unused files, Python 2 support scripts, hardcoded versions/vendor URLs, plaintext default secrets, and fragile per-distro workarounds. The reusable gems — the custom download module, the tested geographic FQR geometry utility, the client-side static geocoder, the Ansible facts script, and the pdsm proot service manager — are isolated enough to extract into new projects. Overall it is a solid, professional infrastructure-as-code project with a few genuinely reusable components, tempered by significant accumulated technical debt.
This is a research repository from the SCIR-HI lab at Harbin Institute of Technology, focused on instruction-tuning large language models (LLaMA, Alpaca, Bloom, Huozi) with Chinese medical knowledge. The core contribution is the construction of Chinese medical instruction datasets using structured knowledge bases (cMeKG) and medical literature via GPT3.5, followed by LoRA-based fine-tuning. The repository provides scripts for fine-tuning, inference, and model export, along with pre-trained LoRA weights. While it demonstrates a complete research pipeline (data generation, training, evaluation), it is primarily a research artifact rather than a production application. The code is functional and includes a clean, reusable Prompter class for template management, but lacks software engineering rigor (no tests, no CI). Reproducibility is moderate: dependencies are pinned, but the data generation process is not fully automated, and external base models and LoRA weights must be obtained separately.