DriveTwinLLM is an instruction-tuned, physics-grounded automotive diagnostic language model built on top of Qwen/Qwen2.5-3B-Instruct. It was fine-tuned using LoRA on NVIDIA B200 GPU hardware (180 GB VRAM) to achieve 98.13% evaluation token accuracy on real-world vehicle trouble codes, live ECU telemetry, and OEM service manuals. The model enforces strict zero-hallucination guardrails and features a specialized Dual-Persona Architecture: - Diagnostic computer scan fee ($90 – $130) - Certified shop labor ($120/hr flat-rate standard) - Strictly refuses unmonitored sensors (transmission fluid temperature, TPMS, cabin humidity) and invalid codes (P9999), redirecting the user to monitored engine…
Independent publisher
Kungumapriyaa M
Kungumapriyaa
AI , ML, CV, DL
Models in Library1
Datasets in Library0
Models on Hugging Face4
Followers—