File size: 2,461 Bytes
3453f3d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 | #!/usr/bin/env python3
"""
Train SEC1-LLM (ACL security incidents)
Incidents:
- ACL blocking legitimate traffic
- ACL misconfiguration
- Excessive deny entries
Output: ONLY CLI FIX COMMANDS (no explanation)
"""
from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, Trainer
from peft import LoraConfig, get_peft_model
from datasets import load_dataset
import torch, json
BASE_MODEL = r"D:\dKorpesio\git_llm_wazuh\hermes\Hermes-3-Llama-3.1-8B"
DATASET = "datasets/service2_dataset_v3_1500.jsonl"
OUT_DIR = "./service_llm/lora_llm_service2"
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
model = AutoModelForCausalLM.from_pretrained(
BASE_MODEL,
torch_dtype=torch.float16,
device_map="auto"
)
lora_cfg = LoraConfig(
r=8,
lora_alpha=32,
lora_dropout=0.1,
target_modules=["q_proj", "v_proj"],
bias="none",
task_type="CAUSAL_LM"
)
model = get_peft_model(model, lora_cfg)
dataset = load_dataset("json", data_files=DATASET)["train"].train_test_split(
test_size=0.1, seed=42
)
def format_sample(ex):
cli = "\n".join(ex["cli_fix"])
prompt = f"""
### Instruction:
{ex["instruction"]}
Rules:
- Output ONLY valid Cisco IOS/IOS-XE CLI commands
- Do NOT include show/debug commands
- Prefer minimal change (do not remove broad protections unless necessary)
- Do NOT provide explanation
### Incident type:
{ex["incident_type"]}
### Wazuh alert:
{json.dumps(ex["wazuh_alert"], indent=2)}
### Devices:
{json.dumps(ex["devices"], indent=2)}
### Response (CLI FIX COMMANDS ONLY):
{cli}
""".strip()
tok = tokenizer(prompt, truncation=True, max_length=1024, padding="max_length")
tok["labels"] = tok["input_ids"].copy()
return tok
train_ds = dataset["train"].map(format_sample)
eval_ds = dataset["test"].map(format_sample)
training_args = TrainingArguments(
output_dir=OUT_DIR,
num_train_epochs=3,
per_device_train_batch_size=1,
gradient_accumulation_steps=4,
learning_rate=2e-4,
fp16=True,
logging_steps=20,
save_strategy="epoch",
save_total_limit=2,
report_to="none"
)
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_ds,
eval_dataset=eval_ds
)
if __name__ == "__main__":
trainer.train()
model.save_pretrained(OUT_DIR)
print("✅ SERVICE2-LLM training finished:", OUT_DIR)
|