pastel gull
Greetings! I'm working on a larger system where two different embedding models are currently in use (like OpenAI text-embedding-3-small → 1536-D, and Google Gemini embeddings → either up to 768-D with text-embedding-005, or around 3000-D with the newer model). I'm wondering if it's worth trying to align these embeddings somehow (e.g., dimension matching, normalization, mapping), or if in practice the only reliable option is to standardize on a single model family and re-embed everything for consistency.
The catch: full standardization sounds like a huge lift too (multiple vendors, lots of data, etc.), so I'm curious if anyone has gone through this and found a clever middle ground—or do I just have to bite the bullet and re-embed the whole hell?
quaint dove