script-kiddie 1.0 Qwen 3 0.6B

Made with love by whatever

What is script-kiddie?

script-kiddie is a model trained on tool-usage, bash-script-writing, python-coding, and kali-linux tools. It is intented to be an educational example of small model that can assist in light pen-testing.

Chat Template

We are using Qwen's format for conversations and function calling. Here's an example:

print(tokenizer.apply_chat_template(ds["train"][7500]["messages"], tokenize=False))

<|im_start|>system
You are a function calling AI model. You are provided with function signatures within <tools></tools> XML tags.You may call one or more functions to assist with the user query. Don't make assumptions about what values to plug into functions.Here are the available tools:<tools> [{'type': 'function', 'function': {'name': 'get_sunrise_sunset_time', 'description': 'Get the sunrise and sunset times for a specific location', 'parameters': {'type': 'object', 'properties': {'location': {'type': 'string', 'description': 'The city and state, e.g. San Francisco, CA'}, 'date': {'type': 'string', 'description': "The desired date in format 'YYYY-MM-DD'"}}, 'required': ['location', 'date']}}}, {'type': 'function', 'function': {'name': 'calculate_distance', 'description': 'Calculate the distance between two locations', 'parameters': {'type': 'object', 'properties': {'location1': {'type': 'string', 'description': 'The first location'}, 'location2': {'type': 'string', 'description': 'The second location'}}, 'required': ['location1', 'location2']}}}] </tools>Use the following pydantic model json schema for each tool call you will make: {'title': 'FunctionCall', 'type': 'object', 'properties': {'arguments': {'title': 'Arguments', 'type': 'object'}, 'name': {'title': 'Name', 'type': 'string'}}, 'required': ['arguments', 'name']}For each function call return a json object with function name and arguments within <tool_call></tool_call> XML tags as follows:
<tool_call>
{tool_call}
</tool_call><|im_end|>
<|im_start|>user
Hi, I am planning a trip to New York City on 2022-12-25. Can you tell me the sunrise and sunset times for that day?<|im_end|>
<|im_start|>assistant
<tool_call>
{'name': 'get_sunrise_sunset_time', 'arguments': {'location': 'New York City', 'date': '2022-12-25'}}
</tool_call><|im_end|>
<|im_start|>user
<tool_response>
<tool_response>
{'sunrise': '07:16 AM', 'sunset': '04:31 PM'}
</tool_response>
</tool_response><|im_end|>
<|im_start|>assistant
<think>

</think>

On December 25, 2022, in New York City, the sun will rise at 07:16 AM and set at 04:31 PM.<|im_end|>

Evaluation

Evaluations are on part with Qwen3:

hf (pretrained=pool-water/script-kiddie,dtype=bfloat16), gen_kwargs: (None), limit: None, num_fewshot: 2, batch_size: auto (40)
|  Tasks  |Version|Filter|n-shot| Metric |   |Value |   |Stderr|
|---------|------:|------|-----:|--------|---|-----:|---|-----:|
|boolq    |      2|none  |     2|acc     |_  |0.6939|_  |0.0081|
|hellaswag|      1|none  |     2|acc     |_  |0.3961|_  |0.0049|
|         |       |none  |     2|acc_norm|_  |0.4963|_  |0.0050|
|piqa     |      1|none  |     2|acc     |_  |0.6757|_  |0.0109|
|         |       |none  |     2|acc_norm|_  |0.6741|_  |0.0109|
|rte      |      1|none  |     2|acc     |_  |0.6751|_  |0.0282|

Usage

Suggested use is:

• serve with vllm
• use agent qwen_agent

Example Qwen Agent Usage

agent = Assistant(
    llm={
        "model": "pool-water/script-kiddie",
        "model_server": base_url,
        "api_key": "EMPTY",
        "generate_cfg": {
            "max_tokens": 1000,
            "temperature": 0.0,
            "top_p": 0.9,
            "frequency_penalty": 0.5,
            "presence_penalty": 0.0,
            "extra_body": {
                "chat_template_kwargs": {
                    "enable_thinking": False,
                },
            },
        },
    },
    function_list=["nmap", "gobuster"],
)

stream = agent.run(
    [
        {
            "role": "user",
            "content": query,
        },
    ],
)

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

• Developed by: @whatever
• Model type: text-generation
• Language(s) (NLP): en
• License: ???
• Finetuned from model [optional]: Qwen/Qwen3-0.6B

Uses

This software is provided strictly for educational and research purposes only. It is intended to help users learn, experiment, and study relevant concepts. The authors and contributors do not endorse or condone any misuse of this software. Use of this software for malicious, unlawful, or unauthorized activities is strictly prohibited, and users assume full responsibility for compliance with all applicable laws and regulations.

Training Hyperparameters

• Training regime: fp32

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Environmental Impact

• Hardware Type: A100
• Hours used: 0.75 hours
• Cloud Provider: [RunPod
• Compute Region: KS-2
• Carbon Emitted: ~0.08 kg

Compute Infrastructure

• Trained for 45 minutes on a single A100 on RunPod

Hardware

A100

Software

HuggingFace SFT