{"slug": "fitcheck-estimate-llm-training-and-serving-vram-before-you-run", "title": "FitCheck — Estimate LLM Training and Serving VRAM Before You Run", "summary": "Developer Anassbzdd released FitCheck, an open-source tool that estimates peak VRAM for LoRA, QLoRA, and full LLM fine-tuning and for serving based on model weights and KV cache, reading a model's Hugging Face config.json without downloading weights or requiring PyTorch, CUDA, or a GPU. Across 57 calibration and repeat runs on a single Tesla T4, FitCheck's full-process estimate showed 2.4% mean absolute error and 13.9% worst absolute error, while a separate 12-run holdout showed 5.4% mean absolute error and 12/12 correct fit-boundary verdicts. FitCheck is installable via pip install fitcheck-llm and is available as a Hugging Face Space and on GitHub.", "body_md": "Hi everyone,\n\nI built **FitCheck** to answer a common question before starting an LLM job:\n\n**Will this configuration fit in my GPU’s memory?**\n\nFitCheck estimates peak VRAM for LoRA, QLoRA, and full fine-tuning. It reports the memory breakdown, usable GPU capacity, headroom, fit verdict, and largest estimated micro-batch that fits.\n\nIt also supports serving estimates based on model weights and KV cache, plus an advisor that explores batch size, sequence length, and LoRA rank.\n\nFitCheck reads the model’s Hugging Face `config.json` and parameter-count metadata. It does not download model weights or require PyTorch, CUDA, or a GPU to run the estimate.\n\nI tested the estimator against real GPU measurements. Across 57 calibration and repeat runs, the full-process estimate has a **2.4% mean absolute error** and a **13.9% worst absolute error**. A separate 12-run holdout has a **5.4% mean absolute error** and produced **12/12 correct fit-boundary verdicts** for the tested setup.\n\nThe current measurements are all from one Tesla T4, so this is not a universal accuracy claim. FitCheck is a sizing tool, not a guarantee that every workload will avoid an OOM.\n\n**Try the Space:** [FitCheck - a Hugging Face Space by mlanvvs](https://huggingface.co/spaces/mlanvvs/fitcheck)\n\n**GitHub and technical details:** [GitHub - Anassbzdd/fitcheck: See if your LLM fits in your GPU's VRAM before loading it — no more guessing or OOM errors. · GitHub](https://github.com/Anassbzdd/fitcheck)\n\n**Install:** `pip install fitcheck-llm`\n\nI would appreciate feedback, especially:\n\nIs the result easy to understand?\n\nWhich models or configurations should I test next?\n\nDoes the estimate match your measured GPU usage?\n\nPositive or critical feedback is welcome. My goal is to find where the estimator is wrong and improve it with reproducible measurements.", "url": "https://wpnews.pro/news/fitcheck-estimate-llm-training-and-serving-vram-before-you-run", "canonical_source": "https://discuss.huggingface.co/t/fitcheck-estimate-llm-training-and-serving-vram-before-you-run/180737#post_1", "published_at": "2026-09-25 13:57:33+00:00", "updated_at": "2026-09-25 14:02:03.810823+00:00", "lang": "en", "topics": ["ai-infrastructure", "large-language-models", "mlops", "ai-tools", "developer-tools"], "entities": ["FitCheck", "Anassbzdd", "Hugging Face", "GitHub", "Tesla T4", "LoRA", "QLoRA", "PyTorch"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/fitcheck-estimate-llm-training-and-serving-vram-before-you-run", "markdown": "https://wpnews.pro/news/fitcheck-estimate-llm-training-and-serving-vram-before-you-run.md", "text": "https://wpnews.pro/news/fitcheck-estimate-llm-training-and-serving-vram-before-you-run.txt", "jsonld": "https://wpnews.pro/news/fitcheck-estimate-llm-training-and-serving-vram-before-you-run.jsonld"}}