NVIDIA · NCA-GENL

The weights thread

Where knowledge actually lives and how it gets changed. Opens in M01 with 'knowledge lives in parameters, not in a lookup table,' runs through embeddings and transformer internals as the shape that knowledge takes, and closes in M11 with RLHF — the only mechanism in the stack that moves human judgment into the weights.

T2 · 40 lessons across 5 modules

  1. M0101-01Next-token prediction: what a language model is trained to do
  2. M0101-02LLM parameters: what they are and where knowledge is stored
  3. M0101-03Tensor shapes in transformers: batch, sequence, and hidden size
  4. M0101-04Vectors, dot products, and cosine similarity
  5. M0101-05Loss functions and cross-entropy explained
  6. M0101-06Gradient descent and backpropagation
  7. M0101-07Train, validation, and test splits
  8. M0101-08How to build an evaluation set for an LLM project
  9. M0303-01What text embeddings are: learned dense vectors
  10. M0303-02Token embeddings vs sentence and document embeddings
  11. M0303-03How to choose an embedding model
  12. M0303-04How to test retrieval quality by hand
  13. M0303-05Vector arithmetic and word analogies (word2vec)
  14. M0404-01Self-attention and why context length costs quadratically
  15. M0404-02Positional encoding: how transformers represent word order
  16. M0404-03Encoder-only vs decoder-only vs encoder-decoder models (BERT, GPT, T5)
  17. M0404-04Autoregressive generation: how an LLM produces text
  18. M0404-05Temperature, top-k, top-p, and greedy decoding
  19. M0404-06The context window: what it is and how to budget it
  20. M0707-01Sparse retrieval and BM25 keyword search
  21. M0707-02Dense retrieval with embeddings
  22. M0707-03Limits of embedding search: negation, recency, and authority
  23. M0707-04Vector databases and ANN indexes (HNSW, IVF)
  24. M0707-05Access control and permissions in RAG retrieval
  25. M0707-06Hybrid search: combining keyword and vector retrieval
  26. M0707-07Reranking with a cross-encoder
  27. M0707-08Assembling context: chunk order and the lost-in-the-middle problem
  28. M0707-09The complete RAG pipeline, stage by stage
  29. M0707-10Debugging RAG: retrieval failure versus generation failure
  30. M0707-11Grounding, citations, and letting a model say I don't know
  31. M0707-12When RAG is the wrong tool
  32. M1111-01Pretraining vs continued pretraining vs instruction tuning
  33. M1111-02Supervised fine-tuning (SFT): what it can and cannot change
  34. M1111-03Catastrophic forgetting when fine-tuning
  35. M1111-04GPU memory requirements for training an LLM
  36. M1111-05LoRA and parameter-efficient fine-tuning (PEFT)
  37. M1111-06RLHF: reinforcement learning from human feedback explained
  38. M1111-07Reward models, reward hacking, and preference data
  39. M1111-08Prompt vs RAG vs fine-tuning: the full decision rule
  40. M1111-09Choosing a model adaptation strategy under real constraints

Part of the throughlines running across the NCA-GENL prep course.