Instructions to use smshahbaj/Rifa-Nano-0.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use smshahbaj/Rifa-Nano-0.5B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="smshahbaj/Rifa-Nano-0.5B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("smshahbaj/Rifa-Nano-0.5B")
model = AutoModelForCausalLM.from_pretrained("smshahbaj/Rifa-Nano-0.5B", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use smshahbaj/Rifa-Nano-0.5B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "smshahbaj/Rifa-Nano-0.5B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "smshahbaj/Rifa-Nano-0.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/smshahbaj/Rifa-Nano-0.5B

SGLang

How to use smshahbaj/Rifa-Nano-0.5B 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 "smshahbaj/Rifa-Nano-0.5B" \
    --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": "smshahbaj/Rifa-Nano-0.5B",
		"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 "smshahbaj/Rifa-Nano-0.5B" \
        --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": "smshahbaj/Rifa-Nano-0.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use smshahbaj/Rifa-Nano-0.5B with Docker Model Runner:
```
docker model run hf.co/smshahbaj/Rifa-Nano-0.5B
```

🚀 RIFA Nano 0.5B

A compact, multilingual AI assistant — built for speed, efficiency, and everyday usefulness.

Developed by SM Shahbaj

📖 Overview

RIFA Nano is the smallest and fastest model in the RIFA series — a family of language models built to run efficiently across a wide range of hardware, from low-end mobile devices to modern GPUs. Nano is fine-tuned on top of Qwen2.5-0.5B-Instruct using LoRA, combining a strong identity, multilingual fluency, and lightweight coding ability in a package small enough to run almost anywhere.

Model	Parameters	Best For
RIFA Nano ⭐	0.5B	Low-end / mobile devices, fast responses
RIFA Flash	1.5B	Balanced performance and speed
RIFA Pro	3B	Complex reasoning, heavier workloads

✨ Capabilities

🌐 Multilingual — fluent in English, Bengali (বাংলা), and Banglish
💬 General assistant — Q&A, everyday writing, summarization, explanations
💻 Basic coding — short scripts, debugging help, code explanations
🧭 Honest by design — responds "Sorry, I don't have the answer for that right now" instead of making things up
🪪 Stable identity — consistently identifies as RIFA Nano, resistant to prompts trying to make it claim another identity

🚀 Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "smshahbaj/Rifa-Nano-0.5B"

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [
    {"role": "user", "content": "Who are you?"}
]

text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
inputs = tokenizer([text], return_tensors="pt").to(model.device)

with torch.no_grad():
    output = model.generate(
        **inputs,
        max_new_tokens=200,
        temperature=0.5,
        top_p=0.9,
        repetition_penalty=1.15,
    )

response = tokenizer.decode(
    output[0][inputs["input_ids"].shape[-1]:],
    skip_special_tokens=True
)
print(response)

🏋️ Training Details


Base model	Qwen/Qwen2.5-0.5B-Instruct
Method	LoRA (r=16, α=32, dropout=0.05)
Identity dataset	308 samples (English, Bangla, Banglish, adversarial/jailbreak-resistant, 5x repeated)
General data	Bangla Alpaca + English Alpaca + CodeAlpaca
Epochs	4
Hardware	Kaggle GPU (T4 x2)
Precision	4-bit (NF4) training, merged to FP16

⚠️ Limitations

As the smallest model in the RIFA series, Nano is not built for long, multi-step reasoning — use RIFA Flash or RIFA Pro for complex tasks.
Knowledge has a training cutoff and the model has no internet access.
No persistent memory between conversations by default.
Best suited for lightweight, everyday tasks rather than deep technical or research-grade work.

📜 License

Released under the Apache 2.0 license, consistent with the base model.

👤 Developer

Built and fine-tuned by SM Shahbaj — full-stack developer, founder of Smart Web BD and Risha Apps, independent music producer, and author of the sci-fi novel Micro Mayhem (2023).

🌐 Website: smshahbaj.com
💻 GitHub: github.com/smshahbaj
🤗 HuggingFace: huggingface.co/smshahbaj

Part of the RIFA model series — Nano · Flash · Pro

Downloads last month: -

Safetensors

Model size

0.5B params

Tensor type

F16

Model tree for smshahbaj/Rifa-Nano-0.5B

Base model

Qwen/Qwen2.5-0.5B

Finetuned

Qwen/Qwen2.5-0.5B-Instruct

Adapter

(766)

this model