storm icon indicating copy to clipboard operation
storm copied to clipboard

[FEATURE] Define model used for embeddings (esp. also azure endpoints)

Open danieldekay opened this issue 9 months ago • 0 comments

Describe the bug Currently the embedding model is hard-coded in https://github.com/stanford-oval/storm/blob/e80d9bbea7362141a479940dabb751c1f244e4b6/knowledge_storm/encoder.py#L83

This can be an issue in many ways:

  • You want to use different model providers
  • You don't have enough quota on text-embedding-3-small
  • You use an Azure region where text-embedding-3-small is not available

Describe the feature

  • Use an Environment Variable to specify the Embedding model, e.g. 'azure/text-embedding-ada-002' as well as the azure endpoint - which could be different to the completion model endpoints.

danieldekay avatar Mar 19 '25 11:03 danieldekay