Managing downloaded models

Models take gigabytes of disk and memory, so you need to know what is installed and what is loaded.

Ollama keeps models on disk and loads them into memory when used. The main commands:

ollama pull gemma4:e2b downloads (or updates) a model without starting a chat.
ollama ls (or ollama list) shows downloaded models with their size and when they were changed.
ollama ps shows models currently loaded in memory, how much memory they use, whether they run on GPU or CPU, the context size, and when they will be unloaded.
ollama stop gemma4:e2b unloads a model from memory now (it stays on disk).
ollama show gemma4:e2b prints details: parameter count, quantization, context length and capabilities such as tools or vision. --modelfile shows its recipe (a later lesson).
ollama cp gemma4:e2b my-gemma makes a copy under a new name.
ollama rm gemma4:e2b deletes it from disk.

Names and tags. A model name has a family and a tag after the colon: qwen3:8b, gemma4:e2b. Tags usually mark the size or variant. Without a tag, Ollama uses latest, which is the family's default variant. Pin a specific tag in scripts and apps so they don't change unexpectedly when the default moves.

Loaded models stay in memory for a while after the last request (5 minutes by default), so the next request is fast. That is why ollama ps can show models you're not actively using.

Running ollama pull again on an existing model downloads only the changed parts, so it is the way to update a model.

Example

ollama pull qwen3:8b            # download without running
ollama ls
# NAME          ID              SIZE      MODIFIED
# qwen3:8b      500a1f067a9f    5.2 GB    2 minutes ago
# gemma4:e2b    7c1fa3e5a1b2    7.2 GB    3 days ago

ollama ps                       # what is loaded right now
# NAME        ID       SIZE    PROCESSOR    CONTEXT   UNTIL
# qwen3:8b    500a...  6.9 GB  100% GPU     4096      4 minutes from now

ollama show qwen3:8b            # parameters, quantization, context, capabilities
ollama stop qwen3:8b            # free the memory now
ollama rm gemma4:e2b            # delete from disk

Remember

pull, ls, rm manage models on disk
ps and stop manage models loaded in memory
Names are family:tag; pin tags in apps instead of relying on latest

Watch out for

Deleting a model with rm when you only wanted to free memory; use ollama stop for that.

Check yourself

Which command tells you whether a loaded model is running on the GPU or the CPU?

Official docs: docs.ollama.com/cli

Example

Remember

Watch out for

Check yourself

Builds on

Unlocks