200 synthetic conversations generated with Claude Opus covering common e-commerce CS scenarios. Dataset: https://www.kaggle.com/datasets/rafikusuma/cs-dataset

How to Use

With Unsloth

python

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "rafkus/gemma4-cs-q8_0",
    max_seq_length = 2048,
    load_in_4bit = True,
)

FastLanguageModel.for_inference(model)

messages = [
    {"role": "system", "content": "Kamu adalah asisten customer service yang profesional dan ramah bernama Purwa."},
    {"role": "user", "content": "Halo, saya mau tanya status pesanan saya nomor #INV20240115"},
]

inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")
outputs = model.generate(inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0]))

With Ollama (GGUF)

ollama run rafkus/gemma4-cs-q8_0

Limitations

Trained on synthetic data only
Domain specific: e-commerce CS
Not tested on production traffic