Config	Value
Method	GRPO (TRL `GRPOTrainer`)
Task	`alert_triage` — classify alerts real vs fake
Reward	env score in [0, 1] from `CyberCrisisEnv`
LoRA rank	8
LoRA alpha	32
Target modules	q/k/v/o/gate/up/down proj
Epochs	1
Steps	5
Hardware	T4 GPU (Google Colab)
Step-1 reward	0.80 (above 0.56 heuristic baseline)

Environment

Live OpenEnv API: https://arsheelpatel06-cyber-crisis.hf.space

openenv validate --url https://arsheelpatel06-cyber-crisis.hf.space
# → 6/6 criteria passed

Load the adapter

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2-0.5B-Instruct")
model = PeftModel.from_pretrained(base, "ArsheelPatel06/cyber-crisis-qwen2-lora")
tokenizer = AutoTokenizer.from_pretrained("ArsheelPatel06/cyber-crisis-qwen2-lora")