Rifa-Nano-0.5B

RIFA Nano (0.5B)

The smallest model in the RIFA series
Fine-tuned by SM Shahbaj

---

About

RIFA Nano is a carefully fine-tuned version of Qwen/Qwen2.5-0.5B-Instruct.
It is designed to be a lightweight, helpful assistant with strong Bangla support and reduced hallucination on unanswerable questions.

---

What's new in v2

• Improved skill mix (Bangla + English + coding + light math + general knowledge)
• Dedicated anti-hallucination training ("I don't know" examples)

---

Downloads

Recommended quant: Q5_K_M (best balance of quality & size)

---

Quick Start

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "smshahbaj/Rifa-Nano-0.5B"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, device_map="auto")

messages = [{"role": "user", "content": "তুমি কে?"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(out[0], skip_special_tokens=True))

GGUF (LM Studio / Ollama / llama.cpp)

Download any .gguf file from the table above and load it directly.

---

Limitations

• This is a 0.5B model. It will still make mistakes on complex reasoning, very recent events, and obscure facts.
• The anti-hallucination training reduces confident wrong answers on clearly unanswerable questions, but it does not add knowledge beyond the model's size.
• Always double-check important information (code, numbers, facts).

---

RIFA Series

---

Maintainer

SM Shahbaj
huggingface.co/smshahbaj

License: Apache 2.0

<!-- RIFA-DOCS-START -->

Documentation

• Series hub (GitHub): https://github.com/smshahbaj/RIFA-Text-Series
• Model page: Rifa-Nano docs
• HF collection: https://huggingface.co/collections/smshahbaj/rifa-text-series

<!-- RIFA-DOCS-END -->