Show HN: UATC-Closed-loop VRAM control and dynamic data pruning for LLM training
Category: library
Tags: llm-training, control-system, memory-management
Score: 6.0/10 (Innovation: 7, Technical: 8, Documentation: 5, Utility: 4)
UATC is a closed-loop control system for LLM training that uses Kalman filters, PID controllers, and dynamic data pruning to prevent out-of-memory crashes on edge GPUs. It adapts batch size and other parameters in real-time, demonstrating a novel application of control theory to stabilize memory-constrained fine-tuning.
Target audience: ml engineers, data engineers
Repository: https://github.com/sajjaddoda72-design/UATC · Python · NOASSERTION · 1 stars
View on Hacker News