---
language:
- en
license: mit
tags:
- pytorch
- causal-lm
- checkpoints
- sft
- maria-1b
---

# Maria-1B SFT & Alignment Checkpoints

This repository hosts all authentic PyTorch `.pt` checkpoints trained during the SFT and alignment phase of **Maria-1B** (HyperGQE architecture, 1.07B parameters).

## Available Checkpoints

| Filename | Format | Description |
| :--- | :--- | :--- |
| `maria_1b_unified_sft_v6.pt` | PyTorch State Dict | Final Unified SFT v6 checkpoint (Analyst + Dialogue + Coding) |
| `maria_1b_unified_sft_v5.pt` | PyTorch State Dict | Unified SFT v5 checkpoint |
| `maria_1b_20B_analyst_v4.pt` | PyTorch State Dict | Domain reasoning & analytics specialist v4 |
| `maria_1b_20B_analyst_v3.pt` | PyTorch State Dict | Domain reasoning & analytics specialist v3 |
| `maria_1b_20B_instruct_v2.pt` | PyTorch State Dict | Conversational Chat & Instruction tuning v2 |
| `maria_1b_20B_instruct_chat.pt`| PyTorch State Dict | Initial conversational patch checkpoint |

## Architecture & Base Model
* **Base Model**: [Zahid2005/Maria-1B](https://huggingface.co/Zahid2005/Maria-1B) (20B token scratch pretraining)
* **Instruct Model**: [Zahid2005/Maria-1B-Instruct](https://huggingface.co/Zahid2005/Maria-1B-Instruct)
* **Preprint Citation**: [https://doi.org/10.5281/zenodo.22747827](https://doi.org/10.5281/zenodo.22747827)