Ollama's -cloud suffix isn’t a label, it’s a silent instruction. I bypassed it.

Описание и применение

Переименование модели, чтобы избежать суффикса «‑cloud», позволяет Ollama корректно использовать облачную модель, устраняя ошибки 404.

Задача

When creating a local‑first model that uses a cloud‑based base, naming the model with a ‘‑cloud’ suffix causes Ollama to strip the suffix and look for a different model on ollama.com, resulting in a 404 error.

Результат

Renaming the model to avoid the reserved ‘‑cloud’ suffix allows Ollama to correctly resolve the remote model and use cloud compute, eliminating 404 errors and enabling local use of large models.

Рабочий процесс

• 1. Define Modelfile with base model gemma4:31b-cloud and system prompt.
• 2. Create model with name ending in ‘‑cloud’ (e.g., Natuworkguy/flash-onyx-2.2:31b-cloud).
• 3. Observe 404 error for Natuworkguy/flash-onyx-2.2:31b.
• 4. Rename model to avoid ‘‑cloud’ suffix (e.g., flash-onyx-2.2:31b-cloudbase).
• 5. Create model again; Ollama now uses remote_model gemma4:31b and succeeds.
• 6. Update build script to separate base tag and published tag, using a ‘cloud-base’ flag in Modelfile header.
• 7. Implement a HEAD check against the registry to skip non‑existent cloud tags.
• 8. Rebuild and publish models with correct naming convention.
• 9. Verify local model works with cloud compute.
• 10. Document the naming rule for future builds.

Доказательства

When accessing the API directly at ollama.com, use model names without the -cloud suffix.

Источник

The config blob knows exactly where compute lives: {"remote_host":"https://ollama.com:443","remote_model":"gemma4:31b"}

Источник