Instructions to use AksaraLLM/AksaraLLM-Qwen-1.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AksaraLLM/AksaraLLM-Qwen-1.5B with Transformers:

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

pipe = pipeline("text-generation", model="AksaraLLM/AksaraLLM-Qwen-1.5B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("AksaraLLM/AksaraLLM-Qwen-1.5B")
model = AutoModelForCausalLM.from_pretrained("AksaraLLM/AksaraLLM-Qwen-1.5B")
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

vLLM

How to use AksaraLLM/AksaraLLM-Qwen-1.5B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AksaraLLM/AksaraLLM-Qwen-1.5B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AksaraLLM/AksaraLLM-Qwen-1.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/AksaraLLM/AksaraLLM-Qwen-1.5B

SGLang

How to use AksaraLLM/AksaraLLM-Qwen-1.5B 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 "AksaraLLM/AksaraLLM-Qwen-1.5B" \
    --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": "AksaraLLM/AksaraLLM-Qwen-1.5B",
		"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 "AksaraLLM/AksaraLLM-Qwen-1.5B" \
        --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": "AksaraLLM/AksaraLLM-Qwen-1.5B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use AksaraLLM/AksaraLLM-Qwen-1.5B with Docker Model Runner:
```
docker model run hf.co/AksaraLLM/AksaraLLM-Qwen-1.5B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

AksaraLLM-Qwen-1.5B

Indonesian-tuned Qwen2 1.5B (1.78B params actual). Production checkpoint of the AksaraLLM-Qwen-1.5B series.

Measured baseline (Devin audit, CPU bf16, 50 short Indonesian sentences)

Metric	Value
Perplexity	8.4
English-stopword ratio in ID-prompted output	0.0%
Indonesian-stopword ratio in ID-prompted output	24.5%
Parameters	1777.1 M
Architecture	Qwen2ForCausalLM

Versions

This repo is the rolling production tag for the 1.5B Qwen2 line. Earlier fixed-snapshot variants:

AksaraLLM/AksaraLLM-Qwen-1.5B-v5-public (PPL 8.4, recommended public download)
AksaraLLM/AksaraLLM-Qwen-1.5B-v3-public (older)
AksaraLLM/AksaraLLM-Qwen-1.5B-v3b-overnight (experimental long-run)
AksaraLLM/aksarallm-1.5b-v2-checkpoint (older 1.5B Qwen2 baseline, PPL 9.9)

Quickstart

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tok = AutoTokenizer.from_pretrained("AksaraLLM/AksaraLLM-Qwen-1.5B")
model = AutoModelForCausalLM.from_pretrained(
    "AksaraLLM/AksaraLLM-Qwen-1.5B",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
inp = tok("Indonesia adalah negara", return_tensors="pt").to(model.device)
print(tok.decode(model.generate(**inp, max_new_tokens=120, do_sample=True, top_p=0.9)[0], skip_special_tokens=True))

Known issues / fix-it list

tie_word_embeddings config / checkpoint mismatch (same as v5-public). Set tie_word_embeddings: false in config.json to silence the warning safely.
Identity uncalibrated (model may say "I'm Qwen").
No chat template in tokenizer_config.json — add Qwen2 ChatML.

License

Apache 2.0

Sample outputs (CPU inference, greedy, repeat_penalty=1.15)

Real generation captured during the Devin audit on 2026-05-02. Greedy decoding (do_sample=False) so these are reproducible.

Continuations (no chat template)

Prompt: Indonesia adalah negara kepulauan terbesar di dunia,
Continuation: dengan 17.508 pulau yang tersebar di seluruh wilayahnya. Indonesia memiliki berbagai jenis tanaman dan hewan, termasuk beberapa spesies unik seperti kelinci kambing (Borneo), burung bali (Sunda), dan buaya laut (Sumatera). Selain itu, Indonesia juga merupakan salah satu negara

Prompt: Resep sederhana untuk membuat nasi goreng adalah
Continuation: dengan memasaknya menggunakan bahan-bahan yang mudah didapatkan di rumah. Bisa juga disajikan sebagai hidangan makan malam atau sarapan. Banyak orang menganggap bahwa nasi goreng hanya bisa dibuat oleh chef, namun sebenarnya tidak demikian. Hanya saja cara pembuatan dan proses penyajian yang ber

Prompt: Ibukota Republik Indonesia adalah
Continuation: Jakarta. Kota ini terletak di bagian tengah pulau Jawa, tepatnya berada pada ketinggian 10 meter dari permukaan laut dan memiliki luas wilayah sekitar 584 kilometer persegi. Jakarta merupakan ibu negara atau Ibukota Republik Indonesia yang menjadi pusat pemerintahan nasional serta

Prompt: Bahasa Indonesia berasal dari rumpun bahasa
Continuation: yang berbeda-beda, yaitu: A. Jawa dan Madura B. Melayu dan Minangkabau C. Aceh dan Sunda D. Sumatera dan Kalimantan

Jawaban: C

Penjelasan:

Bahasa Indonesia adalah sebuah bahasa yang terdiri atas beberapa subbahasa atau var

Downloads last month: 251

Safetensors

Model size

2B params

Tensor type

F32

BF16

Model tree for AksaraLLM/AksaraLLM-Qwen-1.5B

Base model

Qwen/Qwen2-1.5B

Finetuned

(74)

this model

Quantizations

1 model