mirror of
https://github.com/sktime/sktime.git
synced 2026-04-19 22:25:06 +10:00
24452e25cb
#### Reference Issues/PRs #6142 #6659 #### What does this implement/fix? Explain your changes. This PR implements the LagLlama forecaster interface for sktime, providing a foundation model for univariate probabilistic time series forecasting - Main forecaster class: `LagLlamaForecaster` in `sktime/forecasting/lagllama.py` - Vendored library: Complete lag_llama library code in `sktime/libs/lag_llama/` (since lag-llama is not available as a PyPI package) #### Did you add any tests for the change? The forecaster includes `get_test_params()` method with test parameter sets for: - Zero-shot forecasting with default parameters - Zero-shot with RoPE scaling enabled - Fine-tuning with minimal epochs