Instructions to use nuhmanpk/secopsbot-0.6b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use nuhmanpk/secopsbot-0.6b with Transformers:

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

pipe = pipeline("text-generation", model="nuhmanpk/secopsbot-0.6b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("nuhmanpk/secopsbot-0.6b")
model = AutoModelForCausalLM.from_pretrained("nuhmanpk/secopsbot-0.6b", device_map="auto")
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 nuhmanpk/secopsbot-0.6b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "nuhmanpk/secopsbot-0.6b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "nuhmanpk/secopsbot-0.6b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/nuhmanpk/secopsbot-0.6b

SGLang

How to use nuhmanpk/secopsbot-0.6b 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 "nuhmanpk/secopsbot-0.6b" \
    --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": "nuhmanpk/secopsbot-0.6b",
		"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 "nuhmanpk/secopsbot-0.6b" \
        --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": "nuhmanpk/secopsbot-0.6b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use nuhmanpk/secopsbot-0.6b with Docker Model Runner:
```
docker model run hf.co/nuhmanpk/secopsbot-0.6b
```

SecOpsBot-0.6B 🛡️

A 0.6B-parameter grounded cybersecurity controls assistant. Give it an excerpt from a security standard (NIST SP 800-53, 800-171, 800-61, FIPS, ...) and a task, and it produces a summary, key facts, actionable checklist, lifecycle steps, rationale or definition, citing control identifiers only when they appear in the excerpt.

Small enough to run fully offline on a laptop (see the GGUF version), which matters for compliance and security teams who can't send internal material to cloud LLMs. Works well as the "reader" in a RAG pipeline over security documentation.

Results

Evaluated on 60 rows from the test split, whose source documents were never seen in training.

Model	ROUGE-L ↑	Control-ID hallucination ↓	Control-ID recall ↑	Avg words
Qwen3-0.6B (base)	0.279	4.2%	32.7%	98
SecOpsBot-0.6B	0.913	2.0%	95.9%	123

Control-ID hallucination: share of control IDs in the answer (e.g. AC-2, SC-7(3)) that do not appear in the source excerpt.
Control-ID recall: share of control IDs in the reference answer that the model also cites.

Usage

The model is trained with this system prompt. Always use it:

You are SecOpsBot, a cybersecurity controls and incident response assistant. Answer only using information from the provided source excerpt. Be precise, cite control identifiers when the excerpt names them, and be actionable. If the excerpt does not contain the answer, say so.

Example user message format (from the dataset):

Based on this excerpt about assessing security and privacy controls, what should an organization do to PREPARE for, DETECT, and RESPOND to this issue?

SOURCE:
POTENTIAL ASSESSMENT METHODS AND OBJECTS: SA-08(20)-Examine [SELECT FROM: System and services acquision policy; procedures addressing the security design principle of metadata management used in the specificaon, design, development, implementaon, and modificaon of the system; system design documentaon; security and privacy requirements and specificaons for the system; system security and privacy architecture; system security plan; other relevant documents or records]. SA-08(20)-Interview [SELECT FROM: Organizaonal personnel wit
...

from transformers import AutoTokenizer, AutoModelForCausalLM
tok = AutoTokenizer.from_pretrained("nuhmanpk/secopsbot-0.6b")
model = AutoModelForCausalLM.from_pretrained("nuhmanpk/secopsbot-0.6b", device_map="auto")

messages = [
    {"role": "system", "content": SYSTEM_PROMPT},
    {"role": "user", "content": "<source excerpt + task, formatted as above>"},
]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, enable_thinking=False,
                                 return_tensors="pt", return_dict=True).to(model.device)
out = model.generate(**inputs, max_new_tokens=512, do_sample=False, repetition_penalty=1.05)
print(tok.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Run locally with Ollama:

ollama run hf.co/nuhmanpk/secopsbot-0.6b-GGUF:Q4_K_M

Training

Base: unsloth/Qwen3-0.6B, LoRA r=16 on all attention + MLP projections, merged to 16-bit
Data: 13,106 training rows from nuhmanpk/cybersecurity-controls-instructions (train split, 1 epoch), loss on assistant turns only
lr 2e-4 cosine, effective batch 16, max length 4096, trained on a Colab T4 with Unsloth

Limitations

Answers are only as good as the excerpt you provide; this is a reader, not a knowledge base.
Inherits PDF text-extraction noise from the source documents.
Not a substitute for official NIST guidance, a qualified assessor, or legal/compliance advice. Verify before acting.
English only; evaluated on a small held-out sample.

Author

Built by Nuhman PK

🤗 Hugging Face: huggingface.co/nuhmanpk
💻 GitHub: github.com/nuhmanpk
📊 Kaggle: kaggle.com/nuhmanpk
💼 LinkedIn: linkedin.com/in/nuhmanpk
🐦 X: @pk__nuhman
✍️ Medium: @nuhmanpk · Dev.to: dev.to/nuhmanpk

Citation

@misc{secopsbot2026,
  author = {Nuhman PK},
  title  = {SecOpsBot-0.6B: A Tiny Grounded Cybersecurity Controls Assistant},
  year   = {2026},
  url    = {https://huggingface.co/nuhmanpk/secopsbot-0.6b}
}

Downloads last month: 304

Safetensors

Model size

0.6B params

Tensor type

BF16

Model tree for nuhmanpk/secopsbot-0.6b

Base model

Qwen/Qwen3-0.6B-Base

Finetuned

Qwen/Qwen3-0.6B

Finetuned

(1332)

this model

nuhmanpk
/

secopsbot-0.6b