ModelFit Measures LLM VRAM Needs From Actual Model Files, Not Estimates
A developer built ModelFit, a site answering whether a large language model can run on a given machine, after finding that sources gave wildly different VRAM figures for Kimi K3. The site fetches public GGUF model files from the Hugging Face API and sums their real sizes instead of relying on the parameters-times-bits formula. It calculates KV cache exactly from each model's config where available, and otherwise applies a flat overhead estimate, labeling which method was used on each page. A daily automated pipeline refreshes, tests, and redeploys the pages, so new releases typically appear within 24 hours. Mixture-of-experts models are counted at full size, since all experts remain loaded in memory.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in