Loading article content...
← Modelli locali per la GenAI
modelli apertihugging faceinferenza locale
Il panorama dei modelli aperti eseguibili in locale
Come orientarsi tra le famiglie di modelli a pesi aperti (Gemma, Qwen, Llama, Mistral, Phi) e come leggere una model card su Hugging Face prima di scaricare i pesi.
Osservatorio Evolutivo2 min di lettura
Abstract (EN)
Open-weight models have turned local generative AI from a niche into a practical option. This article maps the main families a practitioner meets today (Gemma, Qwen, Llama, Mistral, Phi) and explains how to read a Hugging Face model card before downloading weights: license terms, parameter count, context window, quantized variants, and reported benchmarks. The point is not hype but sober selection: matching a model's size and license to the hardware and the use case at hand. We treat the model card as the primary contract between a research group and its users, and we flag the questions it should answer before any weight leaves the hub.