snapgate-3B

official website snapgate AI : www.snapgate.tech

snapgate-3B is a 3 billion parameter language model fine-tuned from Qwen2.5-3B, designed specifically for customer service, summarization, and task execution like snapclaw AI agent (www.snapgate.tech) in Indonesian and English business contexts.

---

🚀 Key Capabilities

Capability	Description
🎧 Customer Service	Answers customer inquiries in a friendly and solution-oriented manner
📝 Summarization	Condenses long texts into key bullet points
⚙️ Task Execution	Executes structured text-based instructions
🇮🇩 Bilingual	Supports Indonesian and English

---

⚡ Quick Start

Installation

pip install transformers torch accelerate

Inference

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "kadalicious22/snapgate-3B",
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("kadalicious22/snapgate-3B")

def chat(message, system_prompt="You are a Snapgate customer service agent. Respond in a friendly and solution-oriented manner."):
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": message}
    ]
    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer(text, return_tensors="pt").to(model.device)

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=256,
            temperature=0.5,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id,
            eos_token_id=tokenizer.eos_token_id,
            repetition_penalty=1.3,
            no_repeat_ngram_size=4,
            top_p=0.9,
            top_k=50,
        )

    return tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)

# Example usage
print(chat("My order hasn't arrived after 5 days, what should I do?"))

---

💬 Usage Examples

Customer Service

system_cs = "You are a Snapgate customer service agent. Respond in a friendly and solution-oriented manner."

chat("Hi, I'd like to know how to return an item?", system_prompt=system_cs)
# → "Hello! Of course, I'm happy to help with your return process..."

Summarization

system_summary = "You are a summarization agent. Summarize the input into key bullet points."

text = "Snapgate is an AI-powered customer service platform..."
chat(f"Summarize the following text:\n{text}", system_prompt=system_summary)
# → "• Snapgate is an AI-powered CS platform\n• Supports multiple communication channels..."

Agent / Override Role

# Task execution
system_task = "You are an assistant that helps classify support tickets."
chat("Ticket: 'I have not been able to log into my account since yesterday'", system_prompt=system_task)

---

🛠️ Technical Details

Property	Value
Base Model	Qwen/Qwen2.5-3B
Parameters	~3 Billion
Precision	float16
Format	Safetensors
Context Length	32,768 tokens
Framework	Transformers

Recommended Hardware

Setup	Minimum
GPU VRAM	8 GB (float16)
RAM	16 GB
Storage	~7 GB

💡 For CPU usage, apply 4-bit quantization using bitsandbytes.

---

⚠️ Limitations

• Model is optimized for Indonesian and English; performance in other languages is not guaranteed.
• As a 3B parameter model, performance may be lower than larger models on complex tasks.
• Model may produce repetitive output if repetition_penalty is not set appropriately.
• Not recommended for tasks requiring real-time or up-to-date knowledge.
• Always validate outputs before deploying in a production environment.

---

📜 License

This model is released under the Apache 2.0 license. Free to use for commercial purposes with proper attribution.

---

🙏 Credits

• Base model: Qwen2.5-3B by Alibaba Cloud
• Fine-tuning: kadalicious22

---

For business inquiries, open a discussion on this model page.