Model Description

This model is a fine-tuned version of Mistral-7B-Instruct-v0.2, specialised for automotive vehicle diagnostics using OBD-II and UDS fault code protocols.

The model acts as an expert automotive diagnostics assistant that:

• Interprets DTC (Diagnostic Trouble Code) fault codes
• Identifies the most likely root causes
• Recommends step-by-step repair procedures
• Issues safety warnings for critical faults
• Advises consulting a certified mechanic when uncertain

Training Details

DTC Code Coverage

• P codes — Powertrain (engine, fuel, transmission, timing)
• B codes — Body (airbags, HVAC)
• C codes — Chassis (ABS, brakes, wheel speed sensors)
• U codes — Network (CAN bus, ECU communication)

Usage

python
Copy code
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

base_model = "mistralai/Mistral-7B-Instruct-v0.2"
peft_model = "RRK1987/automotive-diagnostics-llm"

tokenizer = AutoTokenizer.from_pretrained(peft_model)
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    load_in_4bit=True,
    device_map="auto",
)
model = PeftModel.from_pretrained(model, peft_model)

prompt = "<s>[INST] You are an expert automotive diagnostics assistant. My 2019 Toyota Camry is showing fault code P0300. I am experiencing rough idle and engine shaking. Can you diagnose this? [/INST]"
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=300)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Safety

This model includes safety guardrails that refuse requests to:

• Disable airbag systems
• Bypass ABS braking systems
• Deactivate any safety-critical vehicle systems

Disclaimer

This model is for educational purposes only. Always consult a certified mechanic for actual vehicle repairs, especially for safety-critical systems.

Dataset

Training data available at: https://huggingface.co/datasets/RRK1987/automotive-dtc-finetune

Project Pipeline

This model is part of a 5-phase GenAI capstone project:

• Phase 1 ✅ Data preparation (119 DTC instruction pairs)
• Phase 2 ✅ ChromaDB vector store (35 documents embedded)
• Phase 3 ✅ QLoRA fine-tuning (this model)
• Phase 4 ⏳ RAG pipeline integration
• Phase 5 ⏳ Evaluation + HuggingFace Spaces deployment