What this does

Launches a pinned variant of a model by appending a parameter-count or capability tag to the model name. This ensures deterministic behavior across runs even when the default upstream version updates.

Steps

Invoke a small parameter variant with a colon separator. Runs the 1-billion-parameter slug of Llama 3.2 if available in the library.
```
ollama run llama3.2:1b
```
Expected output: >>> Send a message (type /bye to exit) if the model pulls successfully; prompts to pull if not yet downloaded.
Try a mid-size variant. Runs the 3-billion-parameter variant for more capable but still lightweight inference.
```
ollama run llama3.2:3b
```
Expected output: Same interactive prompt, with a different digest and memory footprint.
Return to the default variant. Uses an undecorated name when a later default release is preferred.
```
ollama run llama3.2
```
Expected output: Interactive prompt using whatever version the library currently serves as the latest untagged reference.
Exit the session cleanly.
```
>>> /bye
```
Expected output: Shell prompt returns.

Verification

ollama show llama3.2:1b 2>&1 | head -3
# Expected: manifest output with 'parameters' field showing ~1B and 'digest' beginning with a hexadecimal string

Common failures

Error: not found - The specific tag is not published for this model family in the library; run without the tag or try a different family.
pulling manifest - Tag recognized but not yet downloaded; ollama run triggers the pull automatically - an internet connection is required.
parameter mismatch - Host RAM insufficient for the chosen variant; consult ollama show memory estimates in the documentation.
invalid tag syntax - Space or special characters between name and colon; use the strict modelname:tag form with no spaces.
daemon not running - ollama run fails with service error; start the daemon first.

How to run a specific model version by appending the tag

What this does

Steps

Verification

Common failures

Related guides