ollama - 💡(How to fix) Fix ollama load <model> — Dedicated CLI command to preload a model into memory [1 participants]

Feature request: ollama load <model>

Description

It would be very useful to have a dedicated CLI command to load a model into RAM/VRAM without starting an interactive session.

Currently, the main ways to preload a model are:

• ollama run <model> "" (hacky)
• Sending an empty request via the API (/api/generate or /api/chat)

A clean CLI command like ollama load llama3.2 (or ollama load llama3.2 --keepalive -1) would be much more intuitive, especially for server + front-end setups.

Use cases

• Preloading one or more models at server startup
• Scripts and systemd services
• Applications that use Ollama via API but want a simple way to warm up models from the command line
• Making the CLI more complete and consistent (pull, run, push, list, ps → load)

Proposed behavior

ollama load llama3.2                    # Load and keep according to default keep_alive
ollama load llama3.2 --keepalive -1     # Never unload
ollama load llama3.2 -k 30m             # Keep alive for 30 minutes
ollama load llama3.2 --gpu              # Force GPU layers if wanted(And even the opposite with the CPU)

It could also support multiple models:

ollama load llama3.2 mistral:7b qwen2.5