Instructions to use abdallah3id/rawda-cite-jafari-8b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use abdallah3id/rawda-cite-jafari-8b with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="abdallah3id/rawda-cite-jafari-8b") messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("abdallah3id/rawda-cite-jafari-8b") model = AutoModelForCausalLM.from_pretrained("abdallah3id/rawda-cite-jafari-8b", device_map="auto") messages = [ {"role": "user", "content": "Who are you?"}, ] inputs = tokenizer.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=256) print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use abdallah3id/rawda-cite-jafari-8b with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "abdallah3id/rawda-cite-jafari-8b" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "abdallah3id/rawda-cite-jafari-8b", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/abdallah3id/rawda-cite-jafari-8b
- SGLang
How to use abdallah3id/rawda-cite-jafari-8b with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "abdallah3id/rawda-cite-jafari-8b" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "abdallah3id/rawda-cite-jafari-8b", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "abdallah3id/rawda-cite-jafari-8b" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "abdallah3id/rawda-cite-jafari-8b", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use abdallah3id/rawda-cite-jafari-8b with Docker Model Runner:
docker model run hf.co/abdallah3id/rawda-cite-jafari-8b
ABD3ID AI
Jafari Cite 8B ยท Retrieval-Grounded Source Quotation
ู ุณุงุนุฏ ุงุณุชุดูุงุฏ ุจุงูู ุตุงุฏุฑ ูู ุงูููู ุงูุฌุนูุฑูุ ูุนุชู ุฏ ุนูู ุงูู ูุงุทุน ุงูู ุณุชุฑุฌุนุฉ
Project page and mirror documentation maintained by Abdalla Eid
ูุจุฐุฉ ุจุงูุนุฑุจูุฉ
ABD3ID AI โ Jafari Cite 8B ูู ุงุณู ุนุฑุถ ูู ุฑุขุฉ ุงููู ูุฐุฌ Sadiq-Cite 8B. ุตูู ู ุงููู ูุฐุฌ ูู ุฑุญูุฉ ุงุณุชุดูุงุฏ ุถู ู ูุธุงู ุงุณุชุฑุฌุงุน: ูุณุชูุจู ุณุคุงููุง ูู ูุงุทุน ู ุตุฏุฑูุฉ ุฌุฑู ุงุณุชุฑุฌุงุนูุง ู ุณุจููุงุ ุซู ูุญุงูู ุงุฎุชูุงุฑ ุงูู ูุงุทุน ุฐุงุช ุงูุตูุฉ ูููู ุงูุชุจุงุณุงุชูุง ูุจูุงูุงุช ูุชุจูุง ูุตูุญุงุชูุง. ููู ู ุฎุตุต ูู ูุงุฏ ุงูููู ุงูุฌุนูุฑู/ุงูุฅู ุงู ู ูู ุง ูุตูู ุฅุตุฏุงุฑ ุงูู ุตุฏุฑ.
ูุฐุง ููุณ ุฑูุจูุช ู ุญุงุฏุซุฉ ุนุงู ูุง ููุง ุฌูุฉ ุฅูุชุงุก. ูุง ููุจุบู ุงุณุชุฎุฏุงู ู ูุงุณุชูุจุงุท ุญูู ุ ุฃู ุชุทุจูู ุญูู ุนูู ุญุงูุฉ ุงูุณุงุฆูุ ุฃู ุงุนุชุจุงุฑ ุงูุงูุชุจุงุณ ูุญุฏู ุฌูุงุจูุง ุดุฑุนููุง. ุฅุฐุง ูู ุชูุฏู ุงูู ูุธูู ุฉ ู ูุงุทุน ุฐุงุช ุตูุฉุ ูุงูุณููู ุงูู ูุตูุฏ ูู ุงูุชุตุฑูุญ ุจุนุฏู ุงูุนุซูุฑ ุนูู ู ุงุฏุฉ ู ูุงุณุจุฉ ุจุฏู ุงุฎุชูุงู ุฅุฌุงุจุฉ.
ุญุงูุฉ ุงููุณุฎุฉ: ุจุญุณุจ ู ุนููู ุงุช ุงูุฅุตุฏุงุฑุ ูุฐู ู ุฑุขุฉ ููุฃูุฒุงู ุฏูู ุชุฏุฑูุจ ุฅุถุงูู ุฃู ุชูููู ู ุณุชูู ุฃุฌุฑุงู ู ุดุฑูุน ABD3ID AI. Abdalla Eid ูุฏูุฑ ุตูุญุฉ ุงูู ุดุฑูุน ููุซุงุฆู ุงูู ุฑุขุฉุ ููุง ุชูุณุจ ูุฐู ุงูุจุทุงูุฉ ุฅููู ุชุฃููู ุงููู ูุฐุฌ ุงูุฃุตูู ุฃู ุชุฏุฑูุจู.
ููู ูุนู ูุ
- ูุจุญุซ ูุธุงู ุงูุงุณุชุฑุฌุงุน ูู ู ุฌู ูุนุฉ ู ุตุงุฏุฑ ูุญุฏุฏูุง ุงูู ุดุบููุ ููุณุชุฎุฑุฌ ู ูุงุทุน ู ุฑูู ุฉ.
- ูู ุฑุฑ ุงูุชุทุจูู ุฅูู ุงููู ูุฐุฌ ุงูุณุคุงู ูุงููุตูุต ุงูู ุณุชุฑุฌุนุฉ ู ุน ุงุณู ุงููุชุงุจ ูุฑูู ุงูุตูุญุฉ ููู ู ูุทุน.
- ูุญุงูู ุงููู ูุฐุฌ ููู ุงููุตูุต ุงูุชู ูุฑุงูุง ุฐุงุช ุตูุฉุ ูุฅุธูุงุฑ ู ุฑุฌุนูุงุ ุฃู ุงูุชุตุฑูุญ ุจุฃู ุงูุณูุงู ุงูู ูุฏู ูุง ูุชุถู ู ู ุงุฏุฉ ู ูุงุณุจุฉ.
- ูู
ูู ูู
ูููู ุงูุชุญูู
sadiq_fiqh.verify()ู ุทุงุจูุฉ ุงูุงูุชุจุงุณุงุช ุงููุงุชุฌุฉ ู ุน ุงูู ูุงุทุน ุงูู ู ุฑุฑุฉ ูุฅุฑุฌุงุน ุญุงูุฉ ููู ุงูุชุจุงุณ. - ูุฑุงุฌุน ุงูู ุณุชุฎุฏู ุงููุต ูุงูู ุตุฏุฑ ูุงูุตูุญุฉ ูุงูุณูุงู ูุจู ุนุฑุถู ุฃู ุงูุงุนุชู ุงุฏ ุนููู.
ุงูุชุญูู ุงูุจุฑู ุฌู ู ู ูุฌูุฏ ุงูุชุจุงุณ ูู ุงูู ูุงุทุน ูุง ูุซุจุช ุตุญุฉ ุงูู ุตุฏุฑ ุงูุฃุตููุ ููุง ุตุญุฉ ูุณุจุฉ ุงููููุ ููุง ุตุญุฉ ุงูุญูู ุงููููู ุฃู ุงูุทุจุงูู ุนูู ูุงูุนุฉ.
ุจููุฉ ุงูุฅุฏุฎุงู ุงูู ุชููุนุฉ
ุฃุฑุณู ุงูุณุคุงู ู ุน ุงูู ูุงุทุน ุงูู ุณุชุฑุฌุนุฉุ ูุฃุฑูู ุจูุงูุงุช ุงูู ุตุฏุฑ ุจูู ู ูุทุน. ูุง ุชุนุชู ุฏ ุนูู ุงููู ูุฐุฌ ูุงุฎุชุฑุงุน ุจูุงูุงุช ุงููุชุงุจ ุฃู ุงูุตูุญุฉ:
ุงูุณุคุงู:
<ุณุคุงู ุงูุจุญุซ>
ุงูู
ูุงุทุน ุงูู
ุณุชุฑุฌุนุฉ:
[1] (ุงุณู
ุงููุชุงุจุ ุต 120)
<ูุต ุงูู
ูุทุน>
[2] (ุงุณู
ุงููุชุงุจุ ุต 245)
<ูุต ุงูู
ูุทุน>
ุฃุธูุฑุช ุงุฎุชุจุงุฑุงุช ุชูุณูู ุงูุฅุฏุฎุงู ูู ุจุทุงูุฉ ุงูู ุตุฏุฑ ุฃู ุงูุงูุชุจุงุณ ูุฏ ูุณุชู ุฑ ุนุจุฑ ููุงูุจ ู ุฎุชููุฉุ ููู ุจูุงูุงุช ุงูุฅุณูุงุฏ ุชุชุฏููุฑ ุนูุฏ ุญุฐู ุจูุงูุงุช ุงูู ุตุฏุฑ: ุนูุฏ ุชู ุฑูุฑ ู ูุงุทุน ุจูุง ูุชุงุจ ุฃู ุตูุญุฉุ ุฃุจูุบ ุงูู ุตุฏุฑ ุนู 0% ูู ุทุงุจูุฉ ุงูุงุณุชุดูุงุฏุงุช ูู ูุทุน. ุฃุฑุณู ุฏุงุฆู ูุง ุจูุงูุงุช ุงูู ุตุฏุฑ ู ุน ูู ู ูุทุนุ ุฃู ุงุญุฐู ุงูุงุณุชุดูุงุฏุงุช ุบูุฑ ุงูู ุชุญููุฉ ูุจู ุนุฑุถ ุงููุงุชุฌ.
ุงูุชูููู ุงูู ูุดูุฑ ู ู ู ุตุฏุฑ ุงููู ูุฐุฌ
ุงูุฃุฑูุงู ุงูุชุงููุฉ ู ู ุจุทุงูุฉ Sadiq-Cite ุงูู ุตุฏุฑูุฉุ ูููุณุช ุงุฎุชุจุงุฑุงุช ู ุณุชููุฉ ูู ุฑุขุฉ ABD3ID AI. ุชุฐูุฑ ุงูุจุทุงูุฉ ุงูู ุตุฏุฑูุฉ 282 ุณุคุงููุง ู ุญุฌูุจูุง ู ุน ุงูู ูุงุทุน ุงูู ุณุชุฑุฌุนุฉ ู100 ุงุฎุชุจุงุฑ ุฑูุถ ุจู ูุงุทุน ุบูุฑ ุฐุงุช ุตูุฉ:
| ุงูู ููุงุณ ุงูู ุจูุบ ุนูู ูู ุงูู ุตุฏุฑ | ุงููุชูุฌุฉ | ู ูุงุญุธุฉ |
|---|---|---|
| ุฅุฌุงุจุงุช ุชุถู ูุช ุงูุชุจุงุณูุง ูุงุญุฏูุง ุนูู ุงูุฃูู | 99.6% | ูุง ูููุณ ูุญุฏู ุตุญุฉ ุงูุงุณุชุดูุงุฏ |
| ุงูุชุจุงุณุงุช ุฃู ูู ุชุชุจุนูุง ุฅูู ุงูู ูุงุทุน ุงูู ูุฏู ุฉ | 87.5% | 84.5% ู ุทุงุจูุฉ ุญุฑููุฉ ุจุนุฏ ุชุทุจูุน ุฅู ูุงุฆู ู3.0% ุดุจู ุญุฑููุฉ |
| ุงูุชุจุงุณุงุช ูู ุชูุฌุฏ ูู ุงูู ูุงุทุน ุงูู ูุฏู ุฉ | 10.8% | ุฎุทุฑ ุฌููุฑูุ ูุฏ ูุนูุฏ ุงููู ูุฐุฌ ุจูุงุก ูุต ู ู ุฐุงูุฑุชู |
| ุฅุฌุงุจุงุช ูุงูุช ุฌู ูุน ุงูุชุจุงุณุงุชูุง ู ุชุญููุฉ | 79.4% | ูุจูู ุฃู ุงููุงุชุฌ ุงููุงู ู ููุณ ู ูุซูููุง ุชููุงุฆููุง |
| ุฅุฌุงุจุงุช ุชุถู ูุช ุฅุญุงูุฉ ูุชุงุจ ูุตูุญุฉ | 97.2% | ูุฌูุฏ ุฅุญุงูุฉ ูุง ูุนูู ุฃููุง ุตุญูุญุฉ |
| ุฅุญุงูุงุช ุทุงุจูุช ู ูุทุนูุง ู ูุฏู ูุง | 82.5% | ุชุชุทูุจ ุชูููุฑ ุจูุงูุงุช ุงูู ุตุฏุฑ ู ุน ุงูู ูุงุทุน |
| ุฑูุถ ุตุญูุญ ูู 100 ุงุฎุชุจุงุฑ ุจู ูุงุทุน ุบูุฑ ู ุฑุชุจุทุฉ | 100% | ูุชูุฌุฉ ู ุฌู ูุนุฉ ุงูุงุฎุชุจุงุฑ ุงูู ูุดูุฑุฉ |
| ุฑูุถ ุฎุงุทุฆ ูุณุคุงู ูุงุจู ููุฅุฌุงุจุฉ | 0.4% (1 ู ู 282) | ุจุญุณุจ ุนููุฉ ุงูู ุตุฏุฑ |
ุงูุฎูุงุตุฉ ุงูุนู ููุฉ: ููู ุงูุฃุฑูุงู ุงูู ูุดูุฑุฉุ ูู ุชูุฌุฏ ุงูุงูุชุจุงุณุงุช ูู ุงูู ูุงุทุน ุงูู ูุฏู ุฉ ูู 10.8% ู ู ุงูุญุงูุงุชุ ููู ุชูู ูู ุงูุงูุชุจุงุณุงุช ู ุชุญููุฉ ูู 20.6% ู ู ุงูุฅุฌุงุจุงุช. ูุฐูู ุดุบูู ุฃุฏุงุฉ ุงูุชุญูู ุนูู ูู ุฅุฌุงุจุฉุ ููุง ุชุนุฑุถ ุงูุชุจุงุณูุง ุบูุฑ ู ุชุญูู ุนูู ุฃูู ู ูููู ู ู ุงูู ุตุฏุฑ. ูุง ุชูุณุฑ ูุฐู ุงูุฃุฑูุงู ุนูู ุฃููุง ูุณุจุฉ ุตุญุฉ ููููุฉ.
ุญุณุงุณูุฉ ุจูุงูุงุช ุงูุฅุญุงูุฉ ููุงูุจ ุงูุฅุฏุฎุงู
ูู ุงุฎุชุจุงุฑ ู ุตุฏุฑ ุงููู ูุฐุฌ ุนูู 60 ุณุคุงููุง ุจุฃุฑุจุนุฉ ุฃุดูุงู ูุฅุฏุฎุงู ุงูู ูุงุทุนุ ุฃูุจูุบ ุนู ุงููุชุงุฆุฌ ุงูุชุงููุฉ:
| ุดูู ุฅุฏุฎุงู ุงูู ูุงุทุน | ุฅุฌุงุจุงุช ูููุง ุงูุชุจุงุณุงุช | ุงูุงูุชุจุงุณ ุงูุญุฑูู | ุฅุญุงูุงุช ุทุงุจูุช ู ูุทุนูุง |
|---|---|---|---|
| ุงููุงูุจ ุงูุฃุตูู ู ุน ุจูุงูุงุช ุงูู ุตุฏุฑ | 100% | 88.8% | 67.8% |
| ูุชูุฉ ุณูุงู ุนูู ูู ุท AnythingLLM | 98.3% | 85.5% | 79.7% |
| ูุณูู XML ููู ุณุชูุฏุงุช | 100% | 87.7% | 66.7% |
| ู ูุงุทุน ุจูุง ูุชุงุจ ุฃู ุตูุญุฉ | 100% | 83.0% | 0% |
ูุฐู ูุชุงุฆุฌ ุจุทุงูุฉ ุงูู ุตุฏุฑ ูููุณุช ุถู ุงููุง ูุชูุฑุงุฑ ุงููุณุจ ุนูู ูู corpus ุฃู ุฅุนุฏุงุฏ ุงุณุชุฏูุงู. ุฏูุฉ ูุธุงู ุงูุงุณุชุฑุฌุงุน ููุณู ูู ุชุซุจุชูุง ูุฐู ุงูุฃุฑูุงู .
ุจุฏุก ุงูุงุณุชุฎุฏุงู
ูู
ูู ุชุดุบูู ุงููู
ูุฐุฌ ุนุจุฑ ุฎุงุฏู
ู
ุญูู ู
ุชูุงูู ู
ุน OpenAIุ ู
ุซู LM Studioุ ุซู
ุงุณุชุฎุฏุงู
ูุงุฌูุฉ sadiq_fiqh ููุชููุฆุฉ ูุงูุชุญูู ููู ุญุฒู
ุฉ ุงูู
ุดุฑูุน ุงูู
ุตุฏุฑ:
pip install sadiq-fiqh
from sadiq_fiqh import Passage, Sadiq, verify
assistant = Sadiq(
base_url="http://localhost:1234/v1",
model="sadiq-cite-8b",
)
passages = [
Passage(
text="ุถุน ููุง ูุตูุง ู
ุณุชุฑุฌุนูุง ู
ู ู
ุตุฏุฑู.",
book="ุงุณู
ุงููุชุงุจ",
page=120,
)
]
answer = assistant.cite("ุงูุชุจ ุณุคุงู ุงูุจุญุซ ููุง", passages)
verification = verify(answer, passages)
print(answer)
print(verification)
ู ุซุงู ุชููุฆุฉ ุชูุถูุญูุ ููุฒู ุชุดุบูู ุงูุฎุงุฏู ูุชุญู ูู ุงูุฃูุฒุงู ูุถุจุท ุนููุงูู ูุฅุตุฏุงุฑ ุงูุญุฒู ุฉ ุจู ุง ููุงุณุจ ุจูุฆุชู. ูุง ุชุฑุณู ู ูุงุทุน ุฃู ุจูุงูุงุช ุฎุงุตุฉ ุฅูู ุฎุฏู ุฉ ุฎุงุฑุฌูุฉ ุฏูู ุชุตุฑูุญ ู ูุงุณุจ.
ูู
ูู ูุฐูู ุงุณุชุฎุฏุงู
ู ูู ุฎุทูุท ุนู
ู ู
ุซู AnythingLLM ุฃู n8n ุฃู ุนู
ูู ู
ุชูุงูู ู
ุน OpenAIุ ุดุฑุท ุชู
ุฑูุฑ ุงูู
ูุงุทุน ุงูู
ุณุชุฑุฌุนุฉ ูุจูุงูุงุช ู
ุตุงุฏุฑูุง. ุชูุตู ุจุทุงูุฉ ุงูู
ุตุฏุฑ ุจุงุณุชุฎุฏุงู
ูู ุชุฑู
ูุฒ ุญุชู
ู (temperature=0) ูุซุจุงุช ุงูุงูุชุจุงุณุ ููุฐุง ูุง ูุถู
ู ุตุญุฉ ุงููุงุชุฌ.
ุงูุงุณุชุฎุฏุงู ุงูู ุณุคูู ูุงููููุฏ
- ุงูู ุฌุงู ุงูู ุนูู ูู ุงูุงุณุชุดูุงุฏ ุจุงููุตูุต ุงูู ุณุชุฑุฌุนุฉ ูู ู ูุงุฏ ุงูููู ุงูุฌุนูุฑู/ุงูุฅู ุงู ูุ ูุง ุฅุตุฏุงุฑ ุงูุฃุญูุงู ุฃู ุชุฑุฌูุญูุง.
- ุงูุงูุชุจุงุณ ุบูุฑ ุงูู
ุชุญูู ูุฏ ูููู ุฅุนุงุฏุฉ ุจูุงุก ูุง ููููุง ุญุฑูููุงุ ุงูุญุต ูุชูุฌุฉ
verify()ูุงูู ุตุฏุฑ ุงูุฃุตูู. - ุงูุงุณุชุดูุงุฏ ูุญุชุงุฌ ุงุณู ุงููุชุงุจ ูุฑูู ุงูุตูุญุฉ ูู ุจูุงูุงุช ุงูู ูุทุนุ ุฅุฐุง ุบุงุจุช ุงูุจูุงูุงุช ููุฏ ุชุธูุฑ ุฅุญุงูุงุช ู ุฎุชููุฉ.
- ุตุญุฉ ุงูุงูุชุจุงุณ ุฏุงุฎู ุงูู ูุทุน ูุง ุชุซุจุช ุตุญุฉ ุงููุชุงุจ ุฃู ุงูุทุจุนุฉ ุฃู ุงููุณุจุฉ ุฃู ุณูุงู ุฉ ุงูุณูุงู.
- ุฌูุฏุฉ ุงููุชูุฌุฉ ุชุนุชู ุฏ ุฃูุถูุง ุนูู corpusุ ููุธุงู ุงูุงุณุชุฑุฌุงุนุ ูุงูุชูุณูู ุ ูุชุฑุชูุจ ุงูู ูุงุทุน ูุจูุงูุงุชูุง.
- ูุง ุชุณุชุฎุฏู ุงููุงุชุฌ ูุญุฏู ูููุชูู ุฃู ูุชุทุจูู ุญูู ุนูู ุดุฎุตุ ุงุฑุฌุน ุฅูู ู ุตุฏุฑ ู ูุซูู ูู ุฎุชุต ู ุคูู.
- ูุง ุชุนู ู ุงูู ุฎุฑุฌุงุช ุนูู ู ุฐุงูุจ ุฃู ุชูุงููุฏ ููููุฉ ุฃุฎุฑู.
- ูุชุงุฆุฌ ุงูู ุตุฏุฑ ู ุญุฏูุฏุฉ ุจู ุฌู ูุนุฉ ุงุฎุชุจุงุฑ ู ุญุฏุฏุฉุ ูู ูููุดุฑ ูู ูุฐู ุงูุจุทุงูุฉ ุชูููู ู ุณุชูู ููู ุฑุขุฉ.
ุงูู ุตุฏุฑ ูุงูุชุฑุฎูุต ูุงูุฅุณูุงุฏ
ูุฐู ุงูู ุฑุขุฉ ุชุนุชู ุฏ ุนูู Sadiq-Cite 8Bุ ุงูู ุจูู ุนูู Qwen3-8B. ุจุญุณุจ ุจุทุงูุฉ ุงูู ุตุฏุฑุ ุงูุฃูุฒุงู ูู ู ุฑุขุฉ RAWDA AI ุบูุฑ ู ุนุฏูุฉุ ููู ููุฌุฑู ุนูููุง ุชุฏุฑูุจ ุฅุถุงูู ุฃู ุชูููู ู ุณุชูู ุถู ู ุงูู ุฑุขุฉ.
ุงูู ุณุชูุฏุน ุงูู ุตุฏุฑ ูุนูู ุชุฑุฎูุต MIT ูููู ูุฐุฌ ุงูู ุดุชูุ ููุฐูุฑ ุงูุชุฒุงู ุงุช ุชุฑุฎูุต Apache-2.0 ูููู ูุฐุฌ ุงูุฃุณุงุณู. ุฑุงุฌุน ู ููุงุช ุงูุชุฑุฎูุต ูุงูุฅุดุนุงุฑุงุช ุงูุฃุตููุฉ ูุงุญุชูุธ ุจุงูุฅุณูุงุฏุงุช ุงูู ุทููุจุฉ ูุจู ุงูุงุณุชุฎุฏุงู ุฃู ุฅุนุงุฏุฉ ุงูุชูุฒูุน. ูุง ูู ูุญ ุงุณู ABD3ID AI ุฃู ุฅุฏุงุฑุฉ Abdalla Eid ููู ุฑุขุฉ ู ูููุฉ ุงูุฃูุฒุงู ุฃู ุงูู ูุงุฏ upstream.
English Overview
ABD3ID AI โ Jafari Cite 8B is a project display name for a mirror of Sadiq-Cite 8B. The model is designed as the citation stage of a retrieval pipeline: it receives a question and passages retrieved beforehand, then attempts to select relevant excerpts and provide their book and page metadata. The source release describes it for Ja'fari (Imami) jurisprudential material.
This is not a general chatbot or a fatwa system. It should not derive rulings, apply them to a person's circumstances, or present a quotation alone as a religious answer. When supplied passages do not contain relevant material, the intended behavior is to say so rather than improvise.
Release status: According to the source release description, this is an unchanged weight mirror with no additional training or independent evaluation by the ABD3ID AI project. Abdalla Eid maintains this project's page and mirror documentation; this card does not claim authorship or training of the upstream model.
How it works
- A retrieval system searches a corpus selected by the operator and returns passages.
- The application provides the question and retrieved text, including book and page metadata for each passage.
- The model attempts to quote relevant supplied text and identify its source, or state that the supplied context contains no relevant material.
- The
sadiq_fiqh.verify()component can match output quotations against the supplied passages and return a per-quote status. - A user reviews the wording, source, page, and context before displaying or relying on the result.
Programmatic matching does not verify the original source, attribution, jurisprudential correctness, or applicability to a real case.
Expected input
Pass the question together with retrieved passages, including source metadata for each passage. Do not rely on the model to invent book or page details:
Question:
<research question>
Retrieved passages:
[1] (Book title, p. 120)
<passage text>
[2] (Book title, p. 245)
<passage text>
The source card reports that quotation behavior persisted across several prompt formats, but citation matching degraded when source metadata was omitted: with bare passages and no book or page, reported citation-to-passage matching was 0%. Always include source metadata or remove citations that have not been verified.
Evaluation reported by the upstream project
The following numbers are reported by the Sadiq-Cite source card, not by an independent evaluation of the ABD3ID AI mirror. The source card describes 282 held-out questions with retrieved passages and 100 refusal probes with deliberately unrelated passages:
| Metric reported by source | Result | Note |
|---|---|---|
| Answers containing at least one quotation | 99.6% | Does not by itself measure citation correctness |
| Quotes traceable to supplied passages | 87.5% | 84.5% exact after orthographic normalization; 3.0% near-verbatim |
| Quotes not found in supplied passages | 10.8% | Material risk: text may be reconstructed from model memory |
| Answers where every quote verified | 79.4% | Full answers are not automatically reliable |
| Answers with a book-and-page citation | 97.2% | Presence of a citation does not make it correct |
| Citations matching a supplied passage | 82.5% | Requires source metadata in the passages |
| Correct refusals on 100 unrelated-context probes | 100% | Result from the reported test set |
| False refusals on answerable questions | 0.4% (1 of 282) | As reported by the source card |
Practical takeaway: In the reported evaluation, 10.8% of quotes were absent from the supplied passages, and 20.6% of answers did not have every quote verified. Run the verifier on every answer and do not present unverified text as a source quotation. These figures are not a measure of jurisprudential correctness.
Sensitivity of citations to input format
The source project reports the following results on 60 questions across four passage-injection formats:
| Passage format | Answers with quotes | Verbatim quote rate | Citations matching a passage |
|---|---|---|---|
| Native format with source metadata | 100% | 88.8% | 67.8% |
| AnythingLLM-style context block | 98.3% | 85.5% | 79.7% |
| XML document tags | 100% | 87.7% | 66.7% |
| Bare passages without book or page | 100% | 83.0% | 0% |
These are source-card results, not a guarantee that the figures will reproduce for every corpus or inference setup. They do not evaluate the retrieval system itself.
Quickstart
The model can be served through a local OpenAI-compatible endpoint, such as LM Studio, and used with the source project's sadiq_fiqh interface:
pip install sadiq-fiqh
from sadiq_fiqh import Passage, Sadiq, verify
assistant = Sadiq(
base_url="http://localhost:1234/v1",
model="sadiq-cite-8b",
)
passages = [
Passage(
text="Insert a passage retrieved from your corpus.",
book="Book title",
page=120,
)
]
answer = assistant.cite("Enter the research question", passages)
verification = verify(answer, passages)
print(answer)
print(verification)
This is an illustrative setup. Start a compatible server, load the weights, and configure its endpoint and package version for your environment. Do not send private passages or data to an external service without appropriate authorization.
The model can also be used in workflows such as AnythingLLM, n8n, or other OpenAI-compatible clients, provided retrieved passages and source metadata are included. The source card recommends deterministic decoding (temperature=0) for consistency; this does not guarantee correctness.
Responsible use and limitations
- The stated scope is citation of retrieved Ja'fari/Imami jurisprudential texts, not issuing or deriving rulings.
- An unverified quote may be reconstructed rather than copied; inspect
verify()results and the original source. - Citations require book and page metadata; missing metadata can lead to fabricated references.
- A quote matching a supplied passage does not verify the book, edition, attribution, or context.
- Results also depend on the corpus, retrieval system, chunking, passage ordering, and metadata.
- Do not use the output alone for a fatwa or to apply a ruling to a person; consult reliable sources and qualified specialists.
- Do not generalize the outputs to other jurisprudential traditions.
- The source results cover a defined test set; this mirror has no independent evaluation reported here.
Source, license, and attribution
This mirror is based on Sadiq-Cite 8B, which is built on Qwen3-8B. According to the source card, the RAWDA AI mirror uses unchanged weights; no additional training or independent mirror evaluation is reported.
The source repository declares MIT for the derived model and notes the Apache-2.0 obligations of its base model. Review and retain upstream license files, notices, and required attributions before use or redistribution. The ABD3ID AI name and mirror documentation maintained by Abdalla Eid do not transfer ownership of upstream weights or materials.
- Downloads last month
- 778