smshahbaj/RIFA-CODE-0.6B

🚀 RIFA Code (0.6B)

A coding-focused, multilingual assistant fine-tuned from Qwen3-0.6B

English | বাংলা | Banglish

---

✨ Overview

RIFA Code is a lightweight, coding-focused fine-tune of Qwen/Qwen3-0.6B, built to write, explain, and debug code while fluently switching between English, Bengali (বাংলা), and Banglish.

It is part of the RIFA series of models by SM Shahbaj.

🔧 Model Details

Property	Value
🧠 Base model	Qwen/Qwen3-0.6B
👤 Fine-tuned by	SM Shahbaj
🎯 Focus	Programming / coding help, debugging, explanations + general assistance
🌐 Languages	English, Bengali (বাংলা), Banglish
🧩 Method	LoRA fine-tuning
📐 LoRA rank (r)	32
📐 LoRA alpha	64
💭 Thinking mode	Disabled (enable_thinking=False) — direct answers, no <think> blocks

💡 What can it do?

• ✅ Write and explain code across common programming languages
• ✅ Debug and fix errors with clear reasoning
• ✅ Answer naturally in English, Bengali, or Banglish depending on how you ask
• ✅ Hold general conversations beyond just coding

🚀 Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "smshahbaj/RIFA-CODE-0.6B"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
)

messages = [
    {"role": "user", "content": "Write a Python function to check if a number is prime."}
]
prompt = tok.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=False,
)
inputs = tok(prompt, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=200)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

🗣️ Example Prompts

Prompt	Language
"Write a Python function to reverse a string."	English
"তুমি কে?"	বাংলা
"Ei code ta fix kore dao?"	Banglish

⚠️ Limitations

• Small (0.6B) model — best for lightweight/edge use cases, not a replacement for large frontier models on complex reasoning.
• May occasionally produce incorrect code for complex or unusual tasks — always review generated code before use.

📜 License

Released under the Apache 2.0 license.

---

Made with ❤️ by SM Shahbaj — part of the RIFA model series