Instructions to use Keanual/MiRAI-4B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Keanual/MiRAI-4B with Transformers:

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

pipe = pipeline("text-generation", model="Keanual/MiRAI-4B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Keanual/MiRAI-4B")
model = AutoModelForCausalLM.from_pretrained("Keanual/MiRAI-4B", 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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Keanual/MiRAI-4B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Keanual/MiRAI-4B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Keanual/MiRAI-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Keanual/MiRAI-4B

SGLang

How to use Keanual/MiRAI-4B 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 "Keanual/MiRAI-4B" \
    --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": "Keanual/MiRAI-4B",
		"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 "Keanual/MiRAI-4B" \
        --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": "Keanual/MiRAI-4B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use Keanual/MiRAI-4B with Docker Model Runner:
```
docker model run hf.co/Keanual/MiRAI-4B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

MiRAI-4B (v0.1)

MiRAI is a model from the K-AI family. It is a small multilingual AI assistant that chats in English, Bahasa Indonesia, and Japanese (日本語).

MiRAI-4B is a fine-tune of Qwen3-4B.

Contoh / Examples

Prompt	MiRAI
Who are you?	I'm MiRAI, a small multilingual AI assistant that speaks English, Indonesian, and Japanese.
Kamu siapa?	Aku MiRAI, asisten AI multibahasa yang bisa berbicara bahasa Inggris, Indonesia, dan Jepang.
あなたは誰ですか？	私はMiRAIです。英語、インドネシア語、日本語で話せる多言語AIアシスタントです。

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

name = "Keanual/MiRAI-4B"
tok = AutoTokenizer.from_pretrained(name)
model = AutoModelForCausalLM.from_pretrained(name, torch_dtype="auto", device_map="auto")

msgs = [{"role": "user", "content": "Kamu siapa?"}]
inp = tok.apply_chat_template(msgs, add_generation_prompt=True,
                              enable_thinking=False,
                              return_tensors="pt", return_dict=True).to(model.device)
out = model.generate(**inp, max_new_tokens=256, temperature=0.7, top_p=0.9, do_sample=True)
print(tok.decode(out[0][inp["input_ids"].shape[1]:], skip_special_tokens=True))

Training details

Base model: Qwen3-4B
Method: QLoRA (4-bit) with Unsloth, LoRA r=16, merged to 16-bit
Hardware: Google Colab, free T4 GPU
Steps: 400 (batch 2, gradient accumulation 4, learning rate 2e-4)
Data (about 1,500 examples per language, plus identity data):
- English: yahma/alpaca-cleaned
- Indonesian: FreedomIntelligence/alpaca-gpt4-indonesian
- Japanese: FreedomIntelligence/alpaca-gpt4-japanese
- Hand-written identity examples (EN / ID / JA) so the model introduces itself as MiRAI

Limitations

This is an early version (v0.1) and a small 4B model, so it can make mistakes and state wrong facts with confidence.
Training data is relatively small and partly machine-generated, so answers can sound generic and quality varies by language.
It was not trained for tool calling.
Not suitable for medical, legal, or financial decisions.

License

The model weights follow the base model license (Apache 2.0). Check the licenses of the datasets listed above before commercial use.

Credits

Part of the K-AI model family.

About K-AI

MiRAI is built by K-AI, an Indonesian AI Startup. Learn more about K-AI, our team, and our other projects on our website: kai-official.netlify.app

Ringkasan (Bahasa Indonesia)

MiRAI-4B adalah model dari keluarga K-AI: asisten AI multibahasa (Inggris, Indonesia, Jepang) hasil fine-tuning dari Qwen3-4B. Ini versi awal (v0.1): ukurannya kecil, jadi bisa keliru dan jawabannya belum selalu sempurna. Jangan dipakai untuk keputusan medis, hukum, atau keuangan.

Downloads last month: 470

Safetensors

Model size

4B params

Tensor type

BF16

Model tree for Keanual/MiRAI-4B

Base model

Qwen/Qwen3-4B-Base

Finetuned

Qwen/Qwen3-4B

Finetuned

(1204)

this model

Quantizations

1 model

Keanual
/

MiRAI-4B