Instructions to use aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1 with Transformers:

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

pipe = pipeline("text-generation", model="aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1")
model = AutoModelForCausalLM.from_pretrained("aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1")
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 aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1

SGLang

How to use aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1 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 "aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1" \
    --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": "aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1",
		"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 "aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1" \
        --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": "aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1 with Docker Model Runner:
```
docker model run hf.co/aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1
```

UB-SR-02 Qwen3.5 9B CPT 200M + SFT 15K

Model ini adalah model bahasa Indonesia berbasis Qwen3.5-9B-Base yang telah melalui dua tahap adaptasi:

Continued Pre-Training / CPT pada korpus pedagogik berbahasa Indonesia sekitar 200 juta token.
Supervised Fine-Tuning / SFT pada sekitar 15.000 dataset instruksi untuk menghasilkan konten edukasi terstruktur dalam format JSON.

Model ini dikembangkan untuk mendukung pembuatan konten pembelajaran bagi konteks Sekolah Rakyat Indonesia, khususnya konten yang selaras dengan prinsip Pembelajaran Mendalam.

Model Details

Base model: Qwen/Qwen3.5-9B-Base
Model type: Causal Language Model
Language: Indonesian
Training stages:
- Continued Pre-Training: ±200M tokens
- Supervised Fine-Tuning: ±15K instruction samples
Fine-tuning method: LoRA
Quantized training: No
QLoRA: No
Primary domain: Education, pedagogy, Indonesian learning materials
Primary output format: Structured JSON
Target users: Educational content generation systems, RAG pipelines, curriculum-aligned content generation

Intended Use

Model ini ditujukan untuk menghasilkan konten edukasi terstruktur, seperti:

Materi pembelajaran
Soal pilihan ganda
Soal esai
Flashcard
Mindmap
Pretest
Konten berbasis sumber belajar atau chunk dari RAG

Model ini paling cocok digunakan dalam sistem yang memberikan:

system prompt berisi aturan pedagogik dan format output
user prompt berisi query, konteks pembelajaran, dan materi sumber
instruksi eksplisit agar model hanya menghasilkan JSON valid

Training Overview

Continued Pre-Training

Tahap CPT dilakukan untuk memperkuat kemampuan model dalam memahami bahasa Indonesia dan domain pedagogik.

Korpus CPT mencakup teks edukatif/pedagogik berbahasa Indonesia dengan total sekitar 200 juta token.

Tujuan CPT:

meningkatkan pemahaman domain pendidikan
memperkuat konteks pedagogik Indonesia
meningkatkan kemampuan memahami materi pembelajaran
mengurangi ketergantungan pada pengetahuan umum base model

Supervised Fine-Tuning

Tahap SFT dilakukan menggunakan sekitar 15.000 dataset instruksi.

Dataset SFT berisi pasangan prompt dan respons yang dirancang untuk menghasilkan output edukasi dalam struktur JSON.

Fokus SFT:

mengikuti instruksi sistem
menghasilkan JSON yang valid
menjaga relevansi dengan materi sumber
menyesuaikan gaya bahasa untuk siswa SMA Indonesia
mengikuti prinsip Pembelajaran Mendalam
menghasilkan konten berbasis konteks, bukan halusinasi bebas

Downloads last month: 143

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for aitf-ub-2026/ub-sr-02-qwen3.5-9b-base-sft-v1

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

(99)

this model