Inference, post-training, and scaling - how LLMs actually run and get better in production.
⚡ Inference
KV caching, FlashAttention, and continuous batching - how a trained model actually generates tokens fast.
📱 Small Models
Sub-billion parameter models - where the budget actually goes, and what it costs to deploy them.
🎯 Post-Training
RLHF, DPO, and the reward models that turn a base model’s raw capability into a model people actually want to use.