LLM Basics

Practice LLM Basics in Python with research-based exercises, implementation guides, and tests on PaperLeet.

• Next-token Loss: Shifted Causal Targets

  Align predictions with the next token and compute a stable average loss.

• SFT Loss Mask: Train on Response Tokens

  Shift targets, mask prompt and padding tokens, and average loss over valid response tokens.

• Multi-turn SFT: Assistant Targets and Conversation Boundaries

  Select assistant targets across turns while excluding padding and cross-conversation predictions.

• Top-k: Temperature and Candidate Probabilities

  Apply temperature, keep exactly k candidates, and renormalize their probabilities.

• Top-p: Cumulative Probability and Nucleus Candidates

  Keep the smallest candidate set that reaches the probability threshold, then restore vocabulary order.

• Gradient Accumulation: Valid-token Weighting

  Weight microbatch gradients by valid token counts to match one combined batch.

• Global Gradient Clipping: One Norm for All Parameters

  Compute one gradient norm across all parameters and apply a shared clipping scale.