ZamAI Labs

ZamAI Labs is where we develop and publish foundational work that supports the ZamAI ecosystem — including Pashto language resources, datasets, processing pipelines, and model work.

• Website: https://zamai.dev
• Organization: https://github.com/ZamAI-ORG

---

What we publish

• Pashto datasets and curated language resources
• Text processing and normalization pipelines
• Training experiments and model artifacts
• Evaluation tools and reproducible research assets

---

Repositories

Area	Repository	Description	Link
Processing	pashto-processing	Main repository for processing Pashto texts (normalization, cleaning, preprocessing).	https://github.com/ZamAI-ORG/pashto-processing
Datasets	pashto-datasets	Curated and processed Pashto datasets with source attribution and dataset notes.	https://github.com/ZamAI-ORG/pashto-datasets
Models	zamai-models	Model artifacts and experiments published by ZamAI Labs.	https://github.com/ZamAI-ORG/zamai-models
Pipelines	zamai-training-pipelines	Central hub for data workflow and training/fine-tuning pipelines.	https://github.com/ZamAI-ORG/zamai-training-pipelines
Training	training-spaces	Reusable training/experiment spaces (templates, scripts, reproducible runs).	https://github.com/ZamAI-ORG/training-spaces
Fine-tuning	phi3-mini-pashto-lora	Pashto instruction-tuned LoRA adaptation of microsoft/Phi-3-mini-4k-instruct.	https://github.com/ZamAI-ORG/phi3-mini-pashto-lora
mT5	mt5-pashto	Pashto-focused mT5 experiments and fine-tuning work.	https://github.com/ZamAI-ORG/mt5-pashto
Templates	zamai-pashto-template	Starter template for Pashto language projects in ZamAI Labs.	https://github.com/ZamAI-ORG/zamai-pashto-template

---

Conventions

Dataset structure (recommended)

For dataset repositories/folders, we aim to keep:

• raw/ (optional): small samples used for testing/validation
• processed/: cleaned/normalized outputs ready for training
• SOURCE.md: source URL + attribution + license/terms
• notes.md: processing notes (what ZamAI changed)

Licensing & attribution

Licensing depends on the base datasets/models and their terms. Each repo should document:

• original sources
• required attribution
• any restrictions (if applicable)

---

Contact

• Email: hello@zamai.dev
• GitHub: https://github.com/ZamAI-ORG

© 2026 ZamAI. Home of Zeerak.