Learn Before
Applying Context Compression for a Specialized Task
A financial services company wants its chatbot to answer user queries strictly following a proprietary, 50-page risk assessment document. Fully retraining the underlying language model is too costly, and including the entire 50-page document with every user query is computationally infeasible. Based on the technique of approximating a long context with a compact, continuous representation, outline a two-step process this company could implement to solve this problem.
0
1
Tags
Ch.4 Alignment - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Application in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
Optimization Goal for Soft Prompt Learning via Context Compression
Challenge of Context Compression for Long Sequences
Prompt as a Form of Context
A research team is developing a system where a very long, detailed set of instructions is 'compressed' into a compact, learnable set of numerical values. This compact representation is then used to guide a language model in performing a specific task, aiming to replicate the performance that would be achieved if the model had processed the full set of instructions. What is the most significant practical challenge the team will face when implementing this 'compression' process?
Applying Context Compression for a Specialized Task
Sequential Context Compression with an RNN-like Mechanism
The Goal of Context Compression for Soft Prompts