qwen3.6-27b-cybersecurity-lora API & Inference Endpoint

📋 Model Details

Table with columns: Attribute, Value
Attribute	Value
Base Model	Qwen/Qwen3.6-27B
Adapter Type	LoRA (Low-Rank Adaptation)
Fine-tuning Method	QLoRA (4-bit NF4 + LoRA)
Hardware	1× RTX 4090 (24GB VRAM)
Training Time	~5 hours
Dataset Size	5,000 rows (sampled from 99,870)
Max Sequence Length	512 tokens
License	Apache 2.0

🎯 Intended Use

This LoRA adapter enhances Qwen3.6-27B's capability in cybersecurity domains including:

Penetration testing methodology & tools
Vulnerability assessment & analysis
Security operations & incident response
Network security & protocol analysis
Web application security (OWASP Top 10)
Malware analysis & reverse engineering
Cryptography & secure communications
Security policy & compliance frameworks

⚙️ Training Configuration

Quantization (4-bit NF4)

python

BitsAndBytesConfig(    load_in_4bit=True,    bnb_4bit_compute_dtype=torch.bfloat16,    bnb_4bit_use_double_quant=True,    bnb_4bit_quant_type="nf4",)

LoRA Configuration

python

LoraConfig(    r=8,    lora_alpha=16,    target_modules=[        "q_proj", "k_proj", "v_proj", "o_proj",        "gate_proj", "up_proj", "down_proj",    ],    lora_dropout=0.05,    bias="none",    task_type="CAUSAL_LM",)

Training Hyperparameters

python

SFTConfig(    max_length=512,    per_device_train_batch_size=1,    gradient_accumulation_steps=8,    num_train_epochs=1,    learning_rate=2e-4,    bf16=True,    optim="paged_adamw_8bit",    warmup_steps=10,    lr_scheduler_type="cosine",    logging_steps=10,)

Training Metrics

Table with columns: Step, Loss, Accuracy
Step	Loss	Accuracy
0	1.94	56.8%
~30	0.36	90.7%
~60	0.26	93.2%
~100	0.21	94.1%
Final	~0.21	~94%

Loss converged rapidly within the first ~200 steps, indicating successful knowledge injection into the LoRA adapters.

📦 How to Use

1. Install Dependencies

bash

pip install torch transformers accelerate peft bitsandbytes

2. Load Model with LoRA Adapter

python

import torchfrom transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfigfrom peft import PeftModel # Base modelmodel_name = "Qwen/Qwen3.6-27B" # 4-bit quantization configbnb_config = BitsAndBytesConfig(    load_in_4bit=True,    bnb_4bit_compute_dtype=torch.bfloat16,    bnb_4bit_use_double_quant=True,    bnb_4bit_quant_type="nf4",) tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)if tokenizer.pad_token is None:    tokenizer.pad_token = tokenizer.eos_token model = AutoModelForCausalLM.from_pretrained(    model_name,    quantization_config=bnb_config,    device_map="auto",    torch_dtype=torch.bfloat16,    trust_remote_code=True,) # Load LoRA adaptermodel = PeftModel.from_pretrained(model, "hotdogs/qwen3.6-27b-cybersecurity-lora")

3. Inference

python

def generate_response(system_prompt, user_prompt):    messages = [        {"role": "system", "content": system_prompt},        {"role": "user", "content": user_prompt},    ]    text = tokenizer.apply_chat_template(        messages, tokenize=False, add_generation_prompt=True    )    inputs = tokenizer(text, return_tensors="pt").to(model.device)        outputs = model.generate(        **inputs,        max_new_tokens=1024,        temperature=0.7,        top_p=0.9,        do_sample=True,    )    return tokenizer.decode(outputs[0], skip_special_tokens=True) # Exampleresponse = generate_response(    "You are a cybersecurity expert. Provide detailed, accurate technical information.",    "Explain how to identify SQL injection vulnerabilities in a web application.")print(response)

4. Merge with Base Model (Optional)

bash

python -m peft merge_and_save \    --model_name Qwen/Qwen3.6-27B \    --peft_model hotdogs/qwen3.6-27b-cybersecurity-lora \    --output_dir ./qwen3.6-27b-cybersecurity-merged

📊 Training Details

Dataset

Source: AlicanKiraz0/Cybersecurity-Dataset-Fenrir-v2.1
Format: Chat template (system/user/assistant)
Sample size: 5,000 rows
Topics: Penetration testing, vulnerability assessment, network security, web security, cryptography, malware analysis, forensics, security policies

Training Environment

Table with columns: Component, Specification
Component	Specification
GPU	1× NVIDIA RTX 4090 (24GB)
CPU	8 vCPU
RAM	32 GB
Framework	PyTorch 2.12 + Transformers 5.12 + TRL 1.6
Quantization	bitsandbytes 0.49 (4-bit NF4)
Precision	bfloat16 mixed precision

Memory Usage

Table with columns: Stage, VRAM Usage
Stage	VRAM Usage
Model Load (4-bit)	~17.65 GB
After LoRA	~17.81 GB
During Training	~18-20 GB

🔗 References

⚠️ Limitations

Trained on a subset (5K/99K) of the full dataset — may not cover all cybersecurity topics exhaustively
LoRA rank 8 — limited representation capacity compared to full fine-tune
4-bit quantization introduces minor precision loss vs half-precision
Domain knowledge is limited to the training dataset coverage

📝 Notes

This adapter was trained using CPU offload-free 4-bit NF4 quantization on a single RTX 4090
Gradient checkpointing was used to reduce VRAM consumption during training
The prepare_model_for_kbit_training step was skipped to avoid OOM from fp32 casting of large tensors — gradients are computed directly in bf16

Created by: hotdogs
Training Date: June 17, 2026

💖 Support / สนับสนุน

If you find this model useful, please consider supporting my work!
หากคุณคิดว่าโมเดลนี้มีประโยชน์ กรุณาสนับสนุนผลงานของฉันด้วยนะคะ! 🙏

₿ Bitcoin — BTC:

text

bc1qf27cyk3vmugcdyv9xdtuv5jwz37863crpj5c9v

Thank you for your support! 🙏✨
ขอบคุณมากๆ สำหรับการสนับสนุนค่า! 💖🤗

📋 Model Details

Table with columns: Attribute, Value
Attribute	Value
Base Model	Qwen/Qwen3.6-27B
Adapter Type	LoRA (Low-Rank Adaptation)
Fine-tuning Method	QLoRA (4-bit NF4 + LoRA)
Hardware	1× RTX 4090 (24GB VRAM)
Training Time	~5 hours
Dataset Size	5,000 rows (sampled from 99,870)
Max Sequence Length	512 tokens
License	Apache 2.0

🎯 Intended Use

This LoRA adapter enhances Qwen3.6-27B's capability in cybersecurity domains including:

Penetration testing methodology & tools
Vulnerability assessment & analysis
Security operations & incident response
Network security & protocol analysis
Web application security (OWASP Top 10)
Malware analysis & reverse engineering
Cryptography & secure communications
Security policy & compliance frameworks

⚙️ Training Configuration

Quantization (4-bit NF4)

python

BitsAndBytesConfig(    load_in_4bit=True,    bnb_4bit_compute_dtype=torch.bfloat16,    bnb_4bit_use_double_quant=True,    bnb_4bit_quant_type="nf4",)

BitsAndBytesConfig(    load_in_4bit=True,    bnb_4bit_compute_dtype=torch.bfloat16,    bnb_4bit_use_double_quant=True,    bnb_4bit_quant_type="nf4",)

LoRA Configuration

python

LoraConfig(    r=8,    lora_alpha=16,    target_modules=[        "q_proj", "k_proj", "v_proj", "o_proj",        "gate_proj", "up_proj", "down_proj",    ],    lora_dropout=0.05,    bias="none",    task_type="CAUSAL_LM",)

LoraConfig(    r=8,    lora_alpha=16,    target_modules=[        "q_proj", "k_proj", "v_proj", "o_proj",        "gate_proj", "up_proj", "down_proj",    ],    lora_dropout=0.05,    bias="none",    task_type="CAUSAL_LM",)

Training Hyperparameters

python

SFTConfig(    max_length=512,    per_device_train_batch_size=1,    gradient_accumulation_steps=8,    num_train_epochs=1,    learning_rate=2e-4,    bf16=True,    optim="paged_adamw_8bit",    warmup_steps=10,    lr_scheduler_type="cosine",    logging_steps=10,)

SFTConfig(    max_length=512,    per_device_train_batch_size=1,    gradient_accumulation_steps=8,    num_train_epochs=1,    learning_rate=2e-4,    bf16=True,    optim="paged_adamw_8bit",    warmup_steps=10,    lr_scheduler_type="cosine",    logging_steps=10,)

Training Metrics

Table with columns: Step, Loss, Accuracy
Step	Loss	Accuracy
0	1.94	56.8%
~30	0.36	90.7%
~60	0.26	93.2%
~100	0.21	94.1%
Final	~0.21	~94%

Loss converged rapidly within the first ~200 steps, indicating successful knowledge injection into the LoRA adapters.

📦 How to Use

1. Install Dependencies

bash

pip install torch transformers accelerate peft bitsandbytes

pip install torch transformers accelerate peft bitsandbytes

2. Load Model with LoRA Adapter

python

import torchfrom transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfigfrom peft import PeftModel # Base modelmodel_name = "Qwen/Qwen3.6-27B" # 4-bit quantization configbnb_config = BitsAndBytesConfig(    load_in_4bit=True,    bnb_4bit_compute_dtype=torch.bfloat16,    bnb_4bit_use_double_quant=True,    bnb_4bit_quant_type="nf4",) tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)if tokenizer.pad_token is None:    tokenizer.pad_token = tokenizer.eos_token model = AutoModelForCausalLM.from_pretrained(    model_name,    quantization_config=bnb_config,    device_map="auto",    torch_dtype=torch.bfloat16,    trust_remote_code=True,) # Load LoRA adaptermodel = PeftModel.from_pretrained(model, "hotdogs/qwen3.6-27b-cybersecurity-lora")

import torchfrom transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfigfrom peft import PeftModel # Base modelmodel_name = "Qwen/Qwen3.6-27B" # 4-bit quantization configbnb_config = BitsAndBytesConfig(    load_in_4bit=True,    bnb_4bit_compute_dtype=torch.bfloat16,    bnb_4bit_use_double_quant=True,    bnb_4bit_quant_type="nf4",) tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)if tokenizer.pad_token is None:    tokenizer.pad_token = tokenizer.eos_token model = AutoModelForCausalLM.from_pretrained(    model_name,    quantization_config=bnb_config,    device_map="auto",    torch_dtype=torch.bfloat16,    trust_remote_code=True,) # Load LoRA adaptermodel = PeftModel.from_pretrained(model, "hotdogs/qwen3.6-27b-cybersecurity-lora")

3. Inference

python

def generate_response(system_prompt, user_prompt):    messages = [        {"role": "system", "content": system_prompt},        {"role": "user", "content": user_prompt},    ]    text = tokenizer.apply_chat_template(        messages, tokenize=False, add_generation_prompt=True    )    inputs = tokenizer(text, return_tensors="pt").to(model.device)        outputs = model.generate(        **inputs,        max_new_tokens=1024,        temperature=0.7,        top_p=0.9,        do_sample=True,    )    return tokenizer.decode(outputs[0], skip_special_tokens=True) # Exampleresponse = generate_response(    "You are a cybersecurity expert. Provide detailed, accurate technical information.",    "Explain how to identify SQL injection vulnerabilities in a web application.")print(response)

def generate_response(system_prompt, user_prompt):    messages = [        {"role": "system", "content": system_prompt},        {"role": "user", "content": user_prompt},    ]    text = tokenizer.apply_chat_template(        messages, tokenize=False, add_generation_prompt=True    )    inputs = tokenizer(text, return_tensors="pt").to(model.device)        outputs = model.generate(        **inputs,        max_new_tokens=1024,        temperature=0.7,        top_p=0.9,        do_sample=True,    )    return tokenizer.decode(outputs[0], skip_special_tokens=True) # Exampleresponse = generate_response(    "You are a cybersecurity expert. Provide detailed, accurate technical information.",    "Explain how to identify SQL injection vulnerabilities in a web application.")print(response)

4. Merge with Base Model (Optional)

bash

python -m peft merge_and_save \    --model_name Qwen/Qwen3.6-27B \    --peft_model hotdogs/qwen3.6-27b-cybersecurity-lora \    --output_dir ./qwen3.6-27b-cybersecurity-merged

python -m peft merge_and_save \    --model_name Qwen/Qwen3.6-27B \    --peft_model hotdogs/qwen3.6-27b-cybersecurity-lora \    --output_dir ./qwen3.6-27b-cybersecurity-merged

📊 Training Details

Dataset

Source: AlicanKiraz0/Cybersecurity-Dataset-Fenrir-v2.1
Format: Chat template (system/user/assistant)
Sample size: 5,000 rows
Topics: Penetration testing, vulnerability assessment, network security, web security, cryptography, malware analysis, forensics, security policies

Training Environment

Table with columns: Component, Specification
Component	Specification
GPU	1× NVIDIA RTX 4090 (24GB)
CPU	8 vCPU
RAM	32 GB
Framework	PyTorch 2.12 + Transformers 5.12 + TRL 1.6
Quantization	bitsandbytes 0.49 (4-bit NF4)
Precision	bfloat16 mixed precision

Memory Usage

Table with columns: Stage, VRAM Usage
Stage	VRAM Usage
Model Load (4-bit)	~17.65 GB
After LoRA	~17.81 GB
During Training	~18-20 GB

🔗 References

⚠️ Limitations

Trained on a subset (5K/99K) of the full dataset — may not cover all cybersecurity topics exhaustively
LoRA rank 8 — limited representation capacity compared to full fine-tune
4-bit quantization introduces minor precision loss vs half-precision
Domain knowledge is limited to the training dataset coverage

📝 Notes

This adapter was trained using CPU offload-free 4-bit NF4 quantization on a single RTX 4090
Gradient checkpointing was used to reduce VRAM consumption during training
The prepare_model_for_kbit_training step was skipped to avoid OOM from fp32 casting of large tensors — gradients are computed directly in bf16

Created by: hotdogs
Training Date: June 17, 2026

💖 Support / สนับสนุน

₿ Bitcoin — BTC:

text

bc1qf27cyk3vmugcdyv9xdtuv5jwz37863crpj5c9v

Thank you for your support! 🙏✨
ขอบคุณมากๆ สำหรับการสนับสนุนค่า! 💖🤗

qwen3.6-27b-cybersecurity-lora

README

📋 Model Details

🎯 Intended Use

⚙️ Training Configuration

Quantization (4-bit NF4)

LoRA Configuration

Training Hyperparameters

Training Metrics

📦 How to Use

1. Install Dependencies

2. Load Model with LoRA Adapter

3. Inference

4. Merge with Base Model (Optional)

📊 Training Details

Dataset

Training Environment

Memory Usage

🔗 References

⚠️ Limitations

📝 Notes

💖 Support / สนับสนุน

₿ Bitcoin — BTC:

Explore FriendliAI today

README

📋 Model Details

🎯 Intended Use

⚙️ Training Configuration

Quantization (4-bit NF4)

LoRA Configuration

Training Hyperparameters

Training Metrics

📦 How to Use

1. Install Dependencies

2. Load Model with LoRA Adapter

3. Inference

4. Merge with Base Model (Optional)

📊 Training Details

Dataset

Training Environment

Memory Usage

🔗 References

⚠️ Limitations

📝 Notes

💖 Support / สนับสนุน

₿ Bitcoin — BTC: