amarshiv86/p07-sre-lora-phi3

P07 · SRE Runbook Q&A — LoRA Fine-tuned Phi-3-mini

Fine-tuned Phi-3-mini-4k-instruct on SRE runbook Q&A pairs using QLoRA + PEFT. Part of the Staff SRE · AI Engineer Portfolio.

Model details

Field	Value
Base model	microsoft/Phi-3-mini-4k-instruct
Fine-tuning method	QLoRA (4-bit) + PEFT LoRA
LoRA rank	16
Target modules	q_proj, v_proj
Training epochs	3
Task	SRE Runbook Q&A

Before vs After (ROUGE scores)

See eval_results.json for full before/after comparison on the test set.

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

base = AutoModelForCausalLM.from_pretrained(
    "microsoft/Phi-3-mini-4k-instruct",
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True,
)
model = PeftModel.from_pretrained(base, "amarshiv86/p07-sre-lora-phi3")
tokenizer = AutoTokenizer.from_pretrained("amarshiv86/p07-sre-lora-phi3")

prompt = "<|user|>\nWhat steps should I take when a pod is in CrashLoopBackOff?<|end|>\n<|assistant|>\n"
inputs = tokenizer(prompt, return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(output[0], skip_special_tokens=True))

Training data

SRE runbook Q&A pairs covering: incident response, Kubernetes troubleshooting, SLO/SLI definitions, on-call procedures, and post-mortem templates.

Links

• GitHub Repo
• HF Space Demo
• Training Dataset