Learn Before
Multi-Round Prediction Problem
A multi-round prediction problem is a scenario where a Large Language Model engages in a dialogue over multiple turns. In this setting, the model must not only generate a response to the initial input but also incorporate subsequent user inputs that refine or expand upon earlier interactions. A conversation with rounds can be formally denoted by the sequence , where denotes the user request and denotes the response for each round . Unlike single-turn interactions, the model must generate responses by processing an incrementally growing conversational history.

0
1
Tags
Ch.4 Alignment - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Related
Mathematical Formulation of LLM Inference
Single-Round Prediction Problem
Token-Level Representation of Input and Output Sequences for a Forward Pass
Multi-Round Prediction Problem
Notation for Concatenated Token Sequences
A language model is given an input sequence of tokens representing the phrase 'The best way to learn a new skill is'. The model then calculates the likelihood for several possible completing sequences. Based on the formal objective of the text generation process, which of the following sequences should the model select to output?
Analyzing Model Output Selection
A language model is given an input context
x. It then evaluates two potential output sequences,y_1andy_2. The model's internal calculations determine thaty_1has a higher probability of occurring afterxthany_2. However, a human evaluator findsy_2to be more creative and detailed. According to the formal objective of the text generation process, what should the model do?
Learn After
Comparison of Single-Round vs. Multi-Round Prediction Problems
Healthcare Assistant Chatbot as a Multi-Round Prediction Problem
Training Objective for Multi-Round Dialogue Models
Conditional Log-Probability of a Response in Multi-Round Dialogue
A user is interacting with a language model to plan a vacation. Analyze the following conversation:
Turn 1:
- User: "I want to book a flight to a warm destination for December."
- Model: "That sounds lovely! To help you, could you tell me which continent you're interested in?"
Turn 2:
- User: "Let's focus on South America."
- Model: "Excellent choice for December. Based on that, I recommend Brazil or Colombia. Do you have a preference?"
To generate its response in Turn 2, which of the following sets of information must the model have processed to ensure its suggestion is both relevant and coherent?
Analysis of a Conversational Failure
A user is interacting with a customer support model for an e-commerce site. Consider the following two-turn conversation:
Turn 1:
- User: "Hi, I ordered a blue t-shirt last week, order #12345. The tracking says it was delivered, but I haven't received it."
- Model: "I'm sorry to hear that. Let me check the details for order #12345. I see it was marked as delivered two days ago. Could you please confirm your shipping address is 123 Main St, Anytown?"
Turn 2:
- User: "Yes, that's the correct address. What should I do next?"
Given this history, which of the following responses from the model would be the most effective and contextually appropriate for the next turn?