eland icon indicating copy to clipboard operation
eland copied to clipboard

Add support for downloading deployed models

Open bryceamacker opened this issue 7 months ago • 0 comments

We are using open source huggingface models, but are applying dynamic quantization using eland's --quantize flag. This creates a unique, dynamically quantized version of the model before uploading to Elastic.

As part of our HA/DR infrastructure, we would like to have a static backup of this model so that we can be sure the exact same binary can be saved offline and deployed across all of our multi-region ES clusters. There doesn't seem to be anyway to download or access the models once they have been deployed to Elastic, and the quantization process happens in memory while you are deploying.

bryceamacker avatar May 05 '25 16:05 bryceamacker