Ankit Meda
49 posts

Ankit Meda
@cringeyburger
Wanna try this AI thingy | AI @ Abundant (YC) | @IITKgp


Working in ML starts out as a math problem and very rapidly becomes a containerization problem










How is DeepSeek V4 so INSANELY cheap? 🤔 Compared to a GQA baseline, it's new *compressed attention* mechanism (CSA and HCA) slashes the KV cache memory cost by 98% 🤯 at a 1M-token context! Here’s how: youtu.be/q8holiIirgo












Problem: RL alignment is costly, unstable & needs retraining for reward changes. How: Reformulate decoding as Monte Carlo energy estimation + importance sampling acceleration. Outperforms post-trained RL & TTS baselines on reasoning, coding, and science tasks. ETS: Energy-Guided Test-Time Scaling for Training-Free RL Alignment Paper: arxiv.org/abs/2601.21484 Code: github.com/sheriyuo/ETS #AI #LLM #RL #Inference #TTS #Sampling


















