|
|
--- |
|
|
base_model: unsloth/Qwen3-0.6B |
|
|
library_name: peft |
|
|
pipeline_tag: text-generation |
|
|
tags: |
|
|
- base_model:adapter:unsloth/Qwen3-0.6B |
|
|
- lora |
|
|
- sft |
|
|
- transformers |
|
|
- trl |
|
|
- unsloth |
|
|
license: mit |
|
|
language: |
|
|
- en |
|
|
datasets: |
|
|
- musaoc/Quran-reasoning-SFT |
|
|
--- |
|
|
|
|
|
# Model Card for Model ID |
|
|
|
|
|
|
|
|
## Model Details |
|
|
|
|
|
This model is a fine-tuned version of Qwen/Qwen3-0.6B on the musaoc/Quran-reasoning-SFT dataset. |
|
|
It is designed to perform reasoning and question-answering tasks related to the Quran, providing structured reasoning steps along with the final answer. |
|
|
|
|
|
### Model Description |
|
|
|
|
|
- **Language(s) (NLP):** English |
|
|
- **License:** MIT |
|
|
- **Fine-tuning method**: Supervised fine-tuning (SFT) |
|
|
- **Finetuned from model:** Qwen3-0.6B |
|
|
- **Dataset:** musaoc/Quran-reasoning-SFT |
|
|
|
|
|
|
|
|
## Uses |
|
|
|
|
|
The model is intended for: |
|
|
|
|
|
- Educational purposes: Assisting with structured reasoning about Quranic content. |
|
|
- Research: Exploring reasoning capabilities of small LLMs fine-tuned on religious text. |
|
|
- QA Systems: Providing answers with reasoning traces. |
|
|
|
|
|
Not intended for: |
|
|
|
|
|
- Authoritative religious rulings (fatwas) |
|
|
- Sensitive or controversial theological debates |
|
|
- High-stakes decision making |
|
|
|
|
|
### Out-of-Scope Use |
|
|
|
|
|
- Scope: The model is limited to the reasoning dataset it was trained on. It may not generalize to broader Quranic studies. |
|
|
|
|
|
## Bias, Risks, and Limitations |
|
|
|
|
|
- Bias: Outputs reflect dataset biases and may not represent all scholarly interpretations. |
|
|
- Hallucination risk: Like all LLMs, it may generate incorrect or fabricated reasoning. |
|
|
- Religious sensitivity: Responses may not align with every sect, school, or interpretation. Use with caution in sensitive contexts. |
|
|
|
|
|
## How to Get Started with the Model |
|
|
|
|
|
Use the code below to get started with the model. |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
from peft import PeftModel |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained("unsloth/Qwen3-0.6B",) |
|
|
base_model = AutoModelForCausalLM.from_pretrained( |
|
|
"unsloth/Qwen3-0.6B", |
|
|
device_map={"": 0} |
|
|
) |
|
|
|
|
|
model = PeftModel.from_pretrained(base_model,"Rustamshry/Quran-R1") |
|
|
|
|
|
question = "How does the Quran address the issue of parental authority and children’s rights?" |
|
|
|
|
|
messages = [ |
|
|
{"role" : "user", "content" : question} |
|
|
] |
|
|
text = tokenizer.apply_chat_template( |
|
|
messages, |
|
|
tokenize = False, |
|
|
add_generation_prompt = True, |
|
|
enable_thinking = True, |
|
|
) |
|
|
|
|
|
from transformers import TextStreamer |
|
|
_ = model.generate( |
|
|
**tokenizer(text, return_tensors = "pt").to("cuda"), |
|
|
max_new_tokens = 512, |
|
|
temperature = 0.6, |
|
|
top_p = 0.95, |
|
|
top_k = 20, |
|
|
streamer = TextStreamer(tokenizer, skip_prompt = True), |
|
|
) |
|
|
``` |
|
|
|
|
|
## Training Data |
|
|
|
|
|
**Dataset**: musaoc/Quran-reasoning-SFT |
|
|
|
|
|
The Quranic Reasoning Question Answering (QRQA) Dataset is a synthetic dataset designed for experimenting purposes and for training and evaluating models capable of answering complex, knowledge-intensive questions about the Quran with a strong emphasis on reasoning. |
|
|
This dataset is particularly well-suited for Supervised Fine-Tuning (SFT) of Large Language Models (LLMs) to enhance their understanding of Islamic scripture and their ability to provide thoughtful, reasoned responses. |
|
|
|
|
|
### Framework versions |
|
|
|
|
|
- PEFT 0.17.0 |
