RULER: What's the Real Context Size of Your Long-Context Language Models?
This addresses the need for more comprehensive evaluation benchmarks for long-context language models, which is crucial for researchers and developers in AI and NLP, though it is incremental as it builds upon existing tests.
The paper tackles the problem of evaluating long-context language models by showing that the widely used needle-in-a-haystack test is superficial, and introduces a new synthetic benchmark called RULER with flexible configurations to test diverse tasks like multi-hop tracing and aggregation. The result reveals that despite models claiming context sizes of 32K tokens or greater, only half maintain satisfactory performance at 32K, with large drops as length increases, and analysis of Yi-34B shows significant room for improvement with longer inputs and more complex tasks.
The needle-in-a-haystack (NIAH) test, which examines the ability to retrieve a piece of information (the "needle") from long distractor texts (the "haystack"), has been widely adopted to evaluate long-context language models (LMs). However, this simple retrieval-based test is indicative of only a superficial form of long-context understanding. To provide a more comprehensive evaluation of long-context LMs, we create a new synthetic benchmark RULER with flexible configurations for customized sequence length and task complexity. RULER expands upon the vanilla NIAH test to encompass variations with diverse types and quantities of needles. Moreover, RULER introduces new task categories multi-hop tracing and aggregation to test behaviors beyond searching from context. We evaluate 17 long-context LMs with 13 representative tasks in RULER. Despite achieving nearly perfect accuracy in the vanilla NIAH test, almost all models exhibit large performance drops as the context length increases. While these models all claim context sizes of 32K tokens or greater, only half of them can maintain satisfactory performance at the length of 32K. Our analysis of Yi-34B, which supports context length of 200K, reveals large room for improvement as we increase input length and task complexity. We open source RULER to spur comprehensive evaluation of long-context LMs.