# Pick a model for your computer

Start with a small chat model whose complete weight files use no more than 60% of your RAM; this is a rough filter, not a promise that it will run.

Sources checked: 2026-10-17

## 1. Find your memory

On a Mac, open Apple menu → About This Mac and read Memory. On Windows, open Settings → System → About and read Installed RAM. On Linux, open your system information app or run `free -h`.

RAM is working space. Free disk space is where downloads live. You need both. On Apple silicon the CPU and GPU share unified memory. A separate graphics card has its own VRAM; do not add VRAM to RAM and call it one budget.

## 2. Use our file-size rule

The Finder uses 60% of the stated memory as its file-size ceiling, leaving 40% for other use. The table uses decimal GB, like the Finder. Choose below the ceiling for a first try. Other apps, context and the runner still need space.

| Your RAM | File-size ceiling |
| --- | --- |
| 8 GB | 4.8 GB |
| 16 GB | 9.6 GB |
| 24 GB | 14.4 GB |
| 32 GB | 19.2 GB |
| 64 GB | 38.4 GB |

## 3. Read the exact file size

On 17 October 2026, our [Qwen 0.5B file record](/artifact/hf-model-qwen-qwen2-5-0-5b-instruct-gguf) listed `qwen2.5-0.5b-instruct-q4_k_m.gguf` at 491,400,032 bytes, about 0.49 GB. That passes the file-size screen for 8 GB RAM. It does not prove speed or compatibility on your computer.

Use the size of the chosen complete model, not the total of every alternative in a repository. For a split model, count every required part. Leave disk space for the app, runtime and download too.

## 4. Check these before downloading

- Task: choose a text chat or instruct model for conversation.
- Format: choose one your app explicitly supports. For a first GGUF download, prefer a single complete file.
- Memory: look at actual bytes, not only a name such as 7B or Q4.
- License: open the model's license before choosing how to use it.
- Availability: use the current download action; a catalog record alone does not mean the file is available.

[Open the model Finder](/finder) to compare files. If it has no suitable result, use the small-model example in the Ollama guide.

## Sources

- [Hugging Bay Finder memory rule](https://huggingbay.xyz/llms.txt)
- [Ollama settings and privacy](https://docs.ollama.com/faq)
- [Ollama context and memory](https://docs.ollama.com/context-length)
- [Qwen 0.5B file record](https://huggingbay.xyz/api/artifacts/hf-model-qwen-qwen2-5-0-5b-instruct-gguf)

[Next: choose your computer](/learn#computer)