FINCH: Prompt-guided Key-Value Cache Compression for Large Language Models
The paper introduces FINCH, a cache compression algorithm that avoids the need to train large language models (LLMs) from scratch.
Blog
FACTBENCH: A Dynamic Benchmark for in-the-wild Language Model Factuality Evaluation
The paper presents two contributions: (a) FactBench, a benchmark for evaluating hallucination in LLMs, and (b) VERIFY, an evaluation framework designed to ch...
Navigating the Shortcut Maze: A Comprehensive Analysis of Shortcut Learning in Text Classification by Language Models
The paper examines how language models use spurious correlations (shortcuts) to make decisions, introduces a benchmark to test these shortcuts, and finds tha...
Instances Need More Care: Rewriting Prompts for Instances with LLMs in the Loop Yields Better Zero-Shot Performance
This paper presents an approach where the task model generates an initial output, which is refined by a meta model to improve the input prompt, resulting in ...
A Study About Retrieval Augmented Generation
“Why Was Sam Altman Fired As CEO of OpenAI?” this is one of the questions that ChatGPT fails to answer by default. Though ChatGPT is not wrong, it says, “I d...