benchmarks/hackathon_report.md

MAAS-LLM: End-to-End Video Benchmark

Test Configuration

• Video: Pella Home Fire Drone Footage (720p)
• Environment: Local PC (x86 CPU Baseline)
• Vision Model: YOLOv10 (PyTorch CPU)
• Reasoning Model: Phi-3 4k Instruct (llama.cpp)

Stage 1: Vanilla Baseline (No LoRA)

• Vision Speed (YOLO): 8.77 FPS
• LLM Reasoning Latency: 26.19s per command

Stage 2: Fine-Tuned LoRA (Disaster Analyst)

• Vision Speed (YOLO): 10.33 FPS
• LLM Reasoning Latency: 29.86s per command

---

Generated automatically by MAAS-LLM Evaluator for the Arm AI Optimization Challenge.