DeepHat-V1-7B

DeepHat-V1-7B

A 7B coding model tuned for cybersecurity and DevOps work, fine-tuned from Qwen2.5-Coder-7B. For security and operations engineers who want a small local model, under its added usage restrictions.

Made by DeepHat on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: DeepHat-V1-7B is a fine-tune of Qwen2.5-Coder-7B with 7.6B parameters, from a model series made for offensive and defensive cybersecurity. Its sample system message describes it as an expert in cybersecurity and DevOps. It keeps the base model's 131,072-token context.

Who it suits: Security and operations engineers who want a small coding and security assistant they can run themselves.

Industry | Who uses it | How they use it | Example
Security | Security engineers | Get help with defensive and authorized testing tasks | Explain a configuration risk
DevOps | Platform engineers | Ask for scripts and pipeline help | Draft a deployment script
Software | Developers | Use a coding model with a security bent | Review a function for unsafe input handling
Education | Students | Learn security concepts with a local model | Walk through a common vulnerability class

You can skip it if your use would fall under the license restrictions below, or if you want a general chat model.

Where to use it:

• Python, through Hugging Face Transformers
• Windows, macOS or Linux, with Python for the download step
• Other tools that read Qwen2-style Safetensors weights

Use case | Who | What they get back
Code writing | Developers | Code such as a sorting routine
Security questions | Security engineers | Explanations and example code
Long inputs | Analysts | Longer context after enabling YaRN in config.json

How to set it up:

• Python and pip on your computer
• A recent Transformers version, because versions below 4.37.0 give a qwen2 key error
• A GPU or enough memory for about 15.2 GB of BF16 weights, or about 3.8 GB at 4-bit by estimate

1. Install or update Transformers.
2. Download the model files with the hf command.
3. Run the Python quickstart from the card.

pip install -U huggingface_hub
hf download DeepHat/DeepHat-V1-7B

How to use it: Load the model and tokenizer, build a chat with a system message and your question, then generate. The card caps the example at 512 new tokens.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "DeepHat/DeepHat-V1-7B"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "write a quick sort algorithm."
messages = [
    {"role": "system", "content": "You are DeepHat, created by Kindo.ai. You are a helpful assistant that is an expert in Cybersecurity and DevOps."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]

Example: You run the card's script as written, asking for a quick sort algorithm. The model returns the code in the response variable, which you can print.

For newcomers and experts: If you are new, run the quickstart unchanged first, then swap in your own question. Use it only on systems and code you are allowed to work on.

If you know the field, the shipped config.json allows up to 32,768 tokens. For longer inputs the card adds a YaRN rope_scaling entry with a factor of 4.0. The architecture follows Qwen2.5 with 28 layers and grouped-query attention, 28 heads for queries and 4 for keys and values.

{
  ...,
  "rope_scaling": {
    "factor": 4.0,
    "original_max_position_embeddings": 32768,
    "type": "yarn"
  }
}

Qwen2.5-Coder-7B base
Fine-tune for cybersecurity and DevOps
Load with Transformers
Chat reply

Good to know:

• License: Apache 2.0 with the DeepHat extended usage restrictions
• Cost to run: your own hardware, with no paid service needed for the quickstart
• Data it reaches: the files you download and the prompts you give it
• Not for: uses that break the law, military use, harming minors, harassment, or deceiving people, per the license list

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 7.6B
• Weights: about 15.2 GB in BF16; about 3.8 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 4,315
• Likes: 309
• License: apache-2.0
• Languages: en
• Based on: Qwen/Qwen2.5-Coder-7B
• Quantized versions: 17
• Fine-tunes: 4
• Tags: transformers, safetensors, qwen2, text-generation, code, qwen-coder, cybersecurity, devops, conversational

Links

• Model: https://huggingface.co/DeepHat/DeepHat-V1-7B
• Files: https://huggingface.co/DeepHat/DeepHat-V1-7B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download DeepHat/DeepHat-V1-7B

Then follow the usage in its model card. Nothing in this kit runs by itself.