- formatting
- images
- links
- math
- code
- blockquotes
- external-services
•
•
•
•
•
•
-
From LunarLander to String-Stable Platooning
An LLM-guided handwritten controller suppresses string amplification under delay, noise, and heterogeneity; a tuned robust CACC baseline then clarifies the real boundary.
-
RL fundamental-part 1
Notes on reinforcement learning basics, policy gradients, PPO, DDPG, SAC, RLHF, and GRPO.
-
When Reward Chasing Meets Robustness
A follow-up LunarLander lab note: an LLM-guided handwritten controller reaches PPO-like reward, then exposes the tradeoff between nominal performance and robustness.
-
Heuristic Learning for Control: Building a Harness Beyond Gradients
Notes from building a small heuristic-learning control harness where an LLM reads experiment logs, edits non-neural controllers, tunes parameters, and refines failure modes without gradient training.
-
KV Cache Management: Quantization and Compression
A walk-through of representative KV cache compression (Scissorhands, Top-K attention, SnapKV, PyramidKV) and quantization (TurboQuant, RaBitQ) methods for long-context LLM inference.