Supported Frameworks¶
The CLI inspector detects the artifact framework and selects a prompt template for code generation. All detections use lazy imports — none are required dependencies.
| Framework | Support | Metadata extracted |
|---|---|---|
| sklearn | Full | pipeline steps, feature count, class labels |
PyTorch (nn.Module) |
Full | layer count, first/last layer names |
Transformers (PreTrainedModel) |
Full | model type, num_labels, tokenizer class |
| XGBoost | Full | n_estimators, objective |
| LightGBM | Full | n_estimators, objective |
| CatBoost | Full | feature count, loss function |
ONNX (.onnx file) |
Full | input/output names and shapes |
| sentence-transformers | Full | embedding dimension |
| Generic | Fallback | class name only — LLM fills the gaps |
The inspector runs in an isolated subprocess, so missing frameworks degrade gracefully without affecting the CLI.
LLM provider¶
The CLI uses Groq by default (llama-3.3-70b-versatile). Override with: