Hugging Face — what creators actually do with it

In these 7 posts Hugging Face is mostly the place models are downloaded from, not something creators demonstrate directly. @codrkai searches huggingface.co for a quantized GGUF build of a 9B Qwen model, downloads the 17 GB 16-bit version and serves it with llama.cpp (GGUF download). Under @edhonour's Ollama walkthrough, a commenter thanks him because they had not known how to pull Hugging Face models through Ollama (Ollama intro).

The most substantive thread is Kronos. @gigaqian surveys time-series foundation models such as TimesFM and Chronos, says generic ones barely beat simple baselines on stocks, then presents Kronos, an open model pretrained on about 12 billion candlesticks with code on GitHub and Hugging Face (Kronos). A follow-up builds an automated system around it with Claude, and its caption admits that a stress test of Kronos "failed on many levels" (Kronos build). The first post advises pairing any forecast with risk and cost controls; both posts feed the predictability guide.

The platform's moderation also shows up. One news clip reports Hugging Face disabling a modified GLM-5.3 listing for a content-policy violation, with a site called Pirate Face claiming a mirror; the video pushes cracked-model links, and its own takeaway is that such mirrors are a security risk (GLM-5.3 takedown). Llama co-occurs in 2 posts, including Nemotron 70B launch coverage (Nemotron). This page is for people starting to run open models locally who need to know where the weights come from.