Hugging Fit

VRAM fit, runtime compatibility, and decluttering for huggingface.co — hugging.fit

Hugging Fit reads every Hugging Face model page and tells you the quant size, VRAM headroom, and runtime compatibility — in the language of vLLM, llama.cpp, Ollama, LM Studio, MLX, Jan, Draw Things, and friends. Built for people who run models on their own GPUs and are tired of downloading 70 GB before learning the architecture isn't supported. ▼ VRAM FIT, PER QUANT Set your card once (8 / 12 / 16 / 24 / 48 / 80 GB or custom). Every quantization on the model page is auto-classified fits / tight / won't fit, with the math shown — weights + KV cache (configurable fp16 or fp8) + activation headroom. Sharded GGUFs (Q4_K_M-00001-of-00005.gguf etc.) are grouped and totaled correctly. ▼ RUNTIME & FORMAT COMPATIBILITY Apps detected: vLLM, llama.cpp, Ollama, LM Studio, Jan, MLX LM, Draw Things, DiffusionBee, Docker Model Runner, Lemonade, TGI, SGLang, ExLlamaV2. Libraries: Transformers, Diffusers, GGUF, MLX, Sentence-Transformers, PEFT, Safetensors. Each surfaced as a small badge on the model page and on every listing card. ▼ SMART LISTINGS Browsing /models, /datasets, /spaces, an org's tab, or a collection? Cards get inline badges (params · size · format · fit · last-updated). Dim or hide non-fits. Tick your runtimes and Hugging Fit redirects to HF's URL filter (?apps=ollama,llama-cpp&library=gguf&pipeline_tag=text-generation&license=apache-2.0&other=4-bit) so the server pre-narrows results before you scroll. ▼ ONE-CLICK DOWNLOADS Per-quant ↓ button on every GGUF row. Whole-repo ↓ for safetensors / AWQ / GPTQ / FP8 / NVFP4 / diffusion checkpoints. Sharded files all land in one tidy folder (Downloads/huggingface/<org>_<repo>/). Confirm dialog for anything > 5 GB or > 3 files — no surprise pulls. ▼ HOVER DOSSIER — PIN, DRAG, DOWNLOAD Hover any model link anywhere on Hugging Face: a dossier appears with parameters, license, base model, formats, downloads, and a fit verdict. Pin it so it doesn't auto-hide. Drag it anywhere on the page. Hit ↓ to download the whole repo without ever clicking through. ▼ PROFILE MINI-PANEL Land on /<user> or /<org>? A quick-nav block jumps to their Models (with your runtime filter pre-applied), Datasets, Spaces, Collections, Papers, Posts. One click each, no scrolling. ▼ THEME HUGGING FACE ITSELF Five themes: Midnight (default cool blue), Aurora (synthwave pink/cyan), Daylight (warm cream/orange), Forest (emerald), Mono (brutalist black/white). Toggle "Theme HF page" and the same palette retints Hugging Face by hijacking its Tailwind dark class. ▼ UI CLEANUP, YOUR WAY Hide HF clutter you never use: promo banners, the inference widget, "Use this model" / Deploy buttons, trending carousels, the footer. Or paste your own CSS selectors. ▼ PRIVACY POSTURE - All settings stored in chrome.storage.sync (or local). Never sent to a server. - No analytics, no telemetry, no third-party scripts. - Network requests only to huggingface.co — the same site you're already on. ▼ PERMISSIONS - storage: save your VRAM target, theme, filters, hide rules - downloads: the ↓ buttons trigger chrome.downloads - host: huggingface.co (read public model API + tree) Hugging Fit is an independent third-party utility. It is not affiliated with, endorsed by, or sponsored by Hugging Face. "Hugging Face" is a trademark of Hugging Face Inc. Free. v1.0.0. More info, FAQ, and source at https://hugging.fit/.