mostafaasey25/ nile-chat-12b

83 Downloads Updated 2 months ago

Egyptian Arabic AI assistant (Q8 GGUF) — supports Arabic script & Arabizi. Based on MBZUAI-Paris/Nile-Chat-12B.

ollama run mostafaasey25/nile-chat-12b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "mostafaasey25/nile-chat-12b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='mostafaasey25/nile-chat-12b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'mostafaasey25/nile-chat-12b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

2 models

Size / Usage

Context

Input

nile-chat-12b:latest

13GB · 128K context window · Text · 2 months ago

nile-chat-12b:latest

13GB

128K

Text

nile-chat-12b:q6_k

9.7GB · 128K context window · Text · 2 months ago

nile-chat-12b:q6_k

9.7GB

128K

Text

Readme

Nile-Chat 12B (Q8 GGUF)

Q8 quantized version of MBZUAI-Paris/Nile-Chat-12B — an instruction-tuned LLM specialized in Egyptian Arabic, supporting both Arabic script and Arabizi (Latin script).

## About the Model

Nile-Chat is developed by MBZUAI France Lab (Mohamed bin Zayed University of Artificial Intelligence) as part of the JAIS Initiative. It is fine-tuned from Google’s Gemma 3 12B using Egyptian Arabic data.

Base model: google/gemma-3-12b-pt
Training data: ~3.3B tokens (pre-training) + 1.9M instructions (SFT) + 0.2M samples (DPO)
Dataset: MBZUAI-Paris/Egyptian-SFT
Paper: Nile-Chat: Egyptian Language Models for Arabic and Latin Scripts

## Quantization Details

| Version | Size | Quality | |———|——|———| | Original (FP16) | 22 GB | Baseline | | This version (Q8) | 12 GB | ~99.5% of original |

## Capabilities

Egyptian Arabic conversation (Arabic script + Arabizi)
Question answering in Egyptian dialect
Translation (English ↔ Egyptian Arabic ↔ MSA)
Transliteration (Arabic script ↔ Latin/Arabizi)
Multi-turn dialogue

## Usage

”`bash ollama run mostafaasey25/nile-chat-12b

Example

اسمك ايه؟ اسمي نايل-شات، على اسم نهر النيل…

Esmak eh? Esmi Nile-Chat, 3ala esm nahr el Nile…

Benchmarks

Arabic Script Average: 64.11 | Latin Script Average: 53.88

Outperforms Qwen2.5-14B, Llama-3.1-8B, gemma-3-12b-it, and other Arabic models on Egyptian dialect benchmarks.

Credits

Original model by: https://huggingface.co/MBZUAI-Paris (MBZUAI France Lab)
Quantization by: mostafaasey25
License: https://ai.google.dev/gemma/terms

# Nile-Chat 12B (Q8 GGUF)

Q8 quantized version of [MBZUAI-Paris/Nile-Chat-12B](https://huggingface.co/MBZUAI-Paris/Nile-Chat-12B) — an instruction-tuned
  LLM specialized in **Egyptian Arabic**, supporting both Arabic script and Arabizi (Latin script).

## About the Model

Nile-Chat is developed by **MBZUAI France Lab** (Mohamed bin Zayed University of Artificial Intelligence) as part of the [JAIS
  Initiative](https://arxiv.org/abs/2308.16149). It is fine-tuned from Google's Gemma 3 12B using Egyptian Arabic data.

- **Base model:** google/gemma-3-12b-pt
  - **Training data:** ~3.3B tokens (pre-training) + 1.9M instructions (SFT) + 0.2M samples (DPO)
  - **Dataset:** [MBZUAI-Paris/Egyptian-SFT](https://huggingface.co/datasets/MBZUAI-Paris/Egyptian-SFT)
  - **Paper:** [Nile-Chat: Egyptian Language Models for Arabic and Latin Scripts](https://huggingface.co/papers/2507.04569)

## Quantization Details

| Version | Size | Quality |
  |---------|------|---------|
  | Original (FP16) | 22 GB | Baseline |
  | **This version (Q8)** | **12 GB** | ~99.5% of original |

## Capabilities

- Egyptian Arabic conversation (Arabic script + Arabizi)
  - Question answering in Egyptian dialect
  - Translation (English ↔ Egyptian Arabic ↔ MSA)
  - Transliteration (Arabic script ↔ Latin/Arabizi)
  - Multi-turn dialogue

## Usage

```bash
  ollama run mostafaasey25/nile-chat-12b

Example

>>> اسمك ايه؟
  اسمي نايل-شات، على اسم نهر النيل...

>>> Esmak eh?
  Esmi Nile-Chat, 3ala esm nahr el Nile...

Benchmarks

Arabic Script Average: 64.11 | Latin Script Average: 53.88

Outperforms Qwen2.5-14B, Llama-3.1-8B, gemma-3-12b-it, and other Arabic models on Egyptian dialect benchmarks.

Credits

- Original model by: https://huggingface.co/MBZUAI-Paris (MBZUAI France Lab)
  - Quantization by: mostafaasey25
  - License: https://ai.google.dev/gemma/terms

---

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)