Zum Inhalt springen
Kitbay

huggingface-best

Apache-2.0Aktualisiert am 11. Nov. 202611.334GitHub-Sterne

A Hugging Face skill for choosing a model for a given task.

Was der Skill vermittelt

A Hugging Face skill for choosing a model for a given task. When a user asks which model to use for coding, chat, OCR, retrieval, speech, vision or similar work, the agent reads benchmark scores from the official leaderboards hosted on the Hub, adds parameter counts, and, if the user names their hardware, filters out models that will not fit by applying simple memory rules for full-precision and 4-bit quantised weights. The answer is a comparison table of candidates with their scores. It works through Hub web pages and APIs, so no account is needed for public data.

Funktionen

  • Find candidate models for a task from Hub leaderboards
  • Filter models by the memory of the user's device
  • Compare candidates in a table of benchmark scores