Downloads · 30 days
45
21% of all-time downloads
Agentic-Arabic-2.6-GGUF
DarkStrox/Agentic-Arabic-2.6-GGUF
Agentic-Arabic-2.6-GGUF is a text generation model from DarkStrox. Use it when you need the model to write or continue text. It is set up for gguf. The card lists the license as apache-2.0.
Agentic Arabic 2.6 GGUF is a state-of-the-art 2.6B parameter model fine-tuned specifically for high-precision Arabic Function Calling, Tool Use, and General Instruction Following, built on top of Liquid AI's LFM 2.5 a…
All-time downloads
211
Public
Repo size
1.7 GB
Likes
2
Public
Repo makeup
Click a slice to open those files.
.gguf1.7 GB · 100%
At a glance
- Task
- Text Generation
- Library
- gguf
- License
- apache-2.0
- Access
- Public
- Created
- Aug 17, 2026
- Updated
- Aug 17, 2026
- SHA
- bcb9b555
Base models
Model card
From the Hugging Face model README
🇸🇦 Agentic Arabic 2.6 (GGUF Q4_K_M)
Agentic Arabic 2.6 GGUF is a state-of-the-art 2.6B parameter model fine-tuned specifically for high-precision Arabic Function Calling, Tool Use, and General Instruction Following, built on top of Liquid AI's LFM 2.5 architecture using Unsloth.
👉 Interactive Playground & Demo: https://huggingface.co/spaces/DarkStrox/Agentic-Arabic-2.6-Demo
📈 Benchmark Performance
1. 🛠️ Arabic Function Calling & Tool Use
Evaluated live on unseen Arabic tool-invocation prompts against foundation baselines:
| Model | Architecture / Size | Quantization | Function Calling Accuracy | VRAM Footprint |
|---|---|---|---|---|
| 🟣 Agentic Arabic 2.6 (Q6_K) | Liquid LFM (2.6B) | Q6_K | 95.0% 🟢 | 2.08 GB |
| 🟣 Agentic Arabic 2.6 (Q4_K_M) | Liquid LFM (2.6B) | Q4_K_M | 85.0% 🟡 | 1.67 GB |
| 🩶 LFM 2.6 Base | Liquid LFM (2.6B) | Q5_K_M | 24.0% 🔴 | 1.94 GB |
| ⬛ Gemma 4 E4B IT | Transformer (7.4B) | Q4_K_XL | 10.0% 🔴 | 4.21 GB |
2. ⚡ General Arabic QA & Inference Speed
Evaluated across standard Arabic general instruction tasks (Math, Geography, Concept Explanation, Poetry, Reasoning):
| Model | General QA Accuracy | Avg Response Latency | Speed Multiplier |
|---|---|---|---|
| 🟣 Agentic Arabic 2.6 (Q4_K_M) | 100% (5/5) 🟢 | 1,493 ms ⚡ | 1.0x (Baseline) |
| 🟣 Agentic Arabic 2.6 (Q6_K) | 100% (5/5) 🟢 | 9,570 ms | 6.4x slower |
| 🩶 LFM 2.6 Base | 100% (5/5) | 6,593 ms | 4.4x slower |
| ⬛ Gemma 4 E4B IT | 100% (5/5) | 13,662 ms | 9.1x slower |
💡 Quantization Insight:
Q4_K_Mdelivers ultra-fast response speeds (~1.49s) and 100% general instruction accuracy in a ultra-lightweight 1.67 GB memory footprint. For strict JSON-schema function calling precision,Q6_Kis recommended.
🚀 Quickstart Usage
1. Using llama-cpp-python
from huggingface_hub import hf_hub_download
from llama_cpp import Llama
# Download model GGUF from Hugging Face
model_path = hf_hub_download(
repo_id="DarkStrox/Agentic-Arabic-2.6-GGUF",
filename="LFM2.5-2.6B.Q4_K_M.gguf"
)
llm = Llama(
model_path=model_path,
n_ctx=4096,
n_gpu_layers=-1
)
system_prompt = """You are a function calling AI model. You are provided with function signatures within <tools> </tools> XML tags.
<tools>
[{"name": "get_weather", "description": "Get current weather", "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}}]
</tools>"""
user_prompt = "ما هي حالة الطقس اليوم في عمان؟"
prompt = f"<|im_start|>system\n{system_prompt}<|im_end|>\n<|im_start|>user\n{user_prompt}<|im_end|>\n<|im_start|>assistant\n"
response = llm(prompt, max_tokens=512, stop=["<|im_end|>"])
print(response['choices'][0]['text'])
2. Using llama-server CLI
llama-server.exe -m LFM2.5-2.6B.Q4_K_M.gguf --ngl 99 --port 8080 --ctx-size 4096
📜 Model Details
- Developed by: Custom Arabic Fine-Tuning Pipeline (Unsloth)
- Base Architecture: Liquid AI LFM 2.5
- Format: GGUF (
Q4_K_M) - File Size: 1.67 GB
- Live Space Demo: DarkStrox/Agentic-Arabic-2.6-Demo
- Task
- Text Generation
- Library
- gguf
- License
- apache-2.0
- Languages
- ar
- Created
- Aug 17, 2026
- Updated
- Aug 17, 2026