Lensa ML
Track 5 of 6

AI Engineering

How LLMs are trained at scale and served fast, plus vector search, calibration, and drift in production systems.

You build with LLMs or ship ML models and want to know what drives speed, memory, cost, and reliability. 3 paths · 16 explainers.

Path 01 · 4 explainers

How LLMs Are Trained

Mixed precision, distributed training, scaling laws, and distillation

  1. 1Mixed Precision Training
  2. 2Distributed Training
  3. 3Scaling Laws
  4. 4Knowledge Distillation
Path 02 · 8 explainers

How LLMs Are Served

Sampling, the KV cache, quantization, MoE, and fast inference techniques

  1. 1Sampling: Temperature, Top-k & Top-p
  2. 2KV Cache
  3. 3Quantization
  4. 4Mixture of Experts
  5. 5Speculative Decoding
  6. 6FlashAttention
  7. 7Continuous Batching
  8. 8Structured Outputs
EmbedAttentionAdd & NormFeed-ForwardAdd & Norm
Path 03 · 4 explainers

ML in Production

Vector search, chunking and reranking, calibration, and data drift

  1. 1Vector Search & HNSW
  2. 2Chunking & Reranking for RAG
  3. 3Model Calibration
  4. 4Data Drift & Model Monitoring