eTPS Leaderboard

TL;DR

A benchmarking site that cuts through the marketing fluff of local LLMs by measuring effective tokens per second versus raw generation speed.

Who is this actually for?

Self-hosters and developers tired of benchmarks that don't reflect how a model actually feels during a multi-turn conversation.

The Good

• The Effectiveness Index exposes models that spit out tokens fast but fail to maintain quality in real workflows.
• TTFT (Time to First Token) is front and center, which is the only metric that matters for UX snobs.

The Catch (Potential Downsides)

The 'Effectiveness Index' sounds like a proprietary black box that might be hard to replicate or trust without seeing the full testing framework. Keeping a leaderboard updated as fast as new models drop on Hugging Face is a massive time sink that usually leads to project abandonment.