| NOVA-1 |
| Copyright 2026 NOVA AI |
|
|
| This product includes model weights, tokenizer files and software developed by |
| the Qwen team at Alibaba Cloud. |
|
|
| Base model : Qwen/Qwen2.5-0.5B-Instruct (and optionally Qwen/Qwen2.5-1.5B-Instruct) |
| Copyright : 2024 Alibaba Cloud |
| License : Apache License 2.0 |
| Source : https: |
|
|
| Modifications made by NOVA AI ("NOVA-1"): |
| - supervised fine-tuning with LoRA (Low-Rank Adaptation) on a self-authored |
| English / Hindi / Hinglish instruction dataset; |
| - a new system prompt and assistant identity ("NOVA-1", developed by NOVA AI); |
| - optional merging of the LoRA adapter into the base weights; |
| - inference, evaluation, dataset-generation and API tooling in this repository. |
|
|
| The NOVA-1 training dataset in `data/` was written for this project. No |
| copyrighted books, scraped private data or third-party instruction datasets are |
| included. Mathematical problems are generated by code and their answers are |
| computed by Python, not copied. |
|
|
| NOVA AI is not affiliated with, endorsed by, or sponsored by Alibaba Cloud, |
| the Qwen team, Cloudflare, Hugging Face, Kaggle or Google. "Qwen" and |
| "Qwen2.5" are names of the upstream project and are used only for attribution. |
|
|
| If you redistribute a merged NOVA-1 model you must keep: |
| 1. this NOTICE file, |
| 2. the Apache License 2.0 text (LICENSE), |
| 3. a copy of the model card describing the modifications. |
|
|
| Apache License 2.0: https: |
|
|
