Instructions to use ecmwf/aifs-single-1.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- AnemoI
How to use ecmwf/aifs-single-1.1 with AnemoI:
from anemoi.inference.runners.default import DefaultRunner from anemoi.inference.config.run import RunConfiguration # Create Configuration config = RunConfiguration(checkpoint = {"huggingface":"ecmwf/aifs-single-1.1"}) # Load Runner runner = DefaultRunner(config) - Notebooks
- Google Colab
- Kaggle
SafeTensors version of model?
#7
by owainkenway - opened
I note that this model is distributed as a pickled pytorch checkpoint, the deserialisation of which can be somewhat risky - would it be possible for a safetensors version to be added to the repository?
A safetensors checkpoint is significantly different to the pickled pytorch checkpoint and not a trivial change. Due to how we store data within the weights and checkpoint, anemoi is not capable of this. We acknowledge that this is less than ideal, and it exists on the roadmap to allow for a safetensors version. However, I cannot give you a definitive timeline.
hcookie129 changed discussion status to closed