What's New
Ollama Integration — Deploy Models in One Command
Seamless pipeline: train → export → deploy to Ollama.
New commands:
soup deploy ollama --model <gguf> --name <name>— deploy GGUF to local Ollamasoup deploy ollama --list— list Soup-deployed modelssoup deploy ollama --remove <name>— remove a modelsoup export --deploy ollama— export GGUF + auto-deploy in one step
Features:
- Auto-detect chat template from
soup.yaml(chatml, llama, mistral, vicuna, zephyr) - Custom system prompts via
--system - Configurable Ollama parameters via
--parameter key=value(repeatable) - Rich terminal UI with deploy plan panel and progress indicators
Security hardening:
- GGUF path traversal protection +
.ggufextension validation - Model name validation (no path separators, null bytes)
- Modelfile parameter key allowlist prevents directive injection
- Parameter value newline/null sanitization
- Subprocess calls use list args (no shell injection)
- Warning panel before overwriting existing Ollama models
Stats: 1449 tests, 57 test files, all passing
Install / Upgrade
pip install --upgrade soup-cli