Modelpedia
Work in progress
About
Findings
Models
Concepts
Methods
Datasets
Sources
Light
Dark
Needle-in-a-haystack
anchor
Note
a test harness released as a repository rather than a paper, so the anchor and the artifact are the same URL and only the anchor is recorded
Findings
IC-004
Retrieval heads are sparse, universal, and causally responsible for long-context retrieval in LLMs
[eval]
IC-092
All evaluated LLMs show consistent F1 degradation to at most 0.60 when two or more events match a retrieval cue
[context]
IC-353
Llama-3.1-instruct 8b and 70b fail the harder NIAH test (sandwich needle) but pass the easier passkey retrieval test
[eval]