Models

Bigger Isn't Better: How a 9GB Model Beat 120B Parameters

I benchmarked 17 local LLMs across 13 dimensions with 39 tests. The results destroyed my assumptions about model size.