Small Model-Based Data Selection
This data selection technique uses a smaller, auxiliary model to filter or curate a dataset that will be used to train a larger model. The process involves the small model performing 'Data Selection' to create a refined dataset of high-quality samples. This curated dataset is then fed to the larger model for its training phase, where loss is computed and parameters are updated based on the selected data.
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
Small Model-Based Data Selection
Heuristics-Based Data Filtering for Fine-Tuning
Prioritizing Influential Data for Fine-Tuning
A development team fine-tunes a large language model on a massive, newly-generated dataset of 1 million instruction-response pairs. After training, they find the model's performance is poor, often generating repetitive, nonsensical, or factually incorrect answers. Which of the following is the most likely root cause of this issue and the best initial strategy to address it?
Evaluating a Data Filtering Strategy
A team is preparing a large, synthetically-generated dataset for fine-tuning a language model. They suspect the dataset has several quality issues. Match each potential data quality problem with the primary goal of a filtering method designed to address it.
Your company is rolling out an instruction-tuned L...
You lead an LLM enablement team building an instru...
You’re leading an LLM platform team building an in...
Your company is building an internal IT helpdesk a...
Deciding Whether (and How) to Use Weak-Model Synthetic Data for Instruction Fine-Tuning
Diagnosing and Fixing a Synthetic Instruction-Tuning Data Flywheel That Degrades Model Behavior
Designing a Synthetic Instruction Fine-Tuning Pipeline Under Budget and Quality Constraints
Stabilizing an Instruction-Tuned Support Assistant When Synthetic Data Conflicts with Human Policy
Selecting and Filtering Self-Generated Instruction Data When Bootstrapping a Strong Model from a Weak Supervisor
Choosing a Weak-Model + Self-Instruct Data Strategy for Instruction Fine-Tuning Without Regressions
Efficiency Benefits of Data Selection in Fine-Tuning
Alpagasus Data Selection System
Learn After
Ensemble of Small Models for Data Selection
A research team is fine-tuning a very large, computationally expensive language model on a massive, noisy dataset. To optimize their limited budget, they first perform a single pass with the large model over the dataset to calculate the training loss for each data sample. They then train a much smaller, faster model to predict the loss values that the large model assigned. Finally, they use this trained small model to filter the dataset, keeping only the samples predicted to have high loss. Which statement best evaluates the effectiveness of this data selection strategy?
Visual Diagram of Data Selection with a Small Model
You are tasked with curating a high-quality dataset for fine-tuning a large, computationally expensive model from a massive, unfiltered data source. You decide to use a smaller, auxiliary model to help with the selection process. Arrange the following steps into the correct logical sequence for this data selection workflow.
Optimizing Training with a Limited Budget