Small vs larger embedding models for cost and accuracy

The current setup runs a small Jina and a Jina in Q8 format inside . The next plan is to stop running the locally and use the instead. That would free local computer resources for larger and s.

The main choice is whether to keep a small 0.6B model or move to a larger 4B quantized model. The intended use is writing and math, so accuracy matters more than simply keeping the setup light.

Key points

  • The current setup uses Jina and s in Q8 format.
  • The may be moved from local hardware to the .
  • Saved local resources could be used for larger and s.
  • The decision is between a small 0.6B model and a larger 4B quantized model.
  • Writing and math tasks make accuracy the main priority.
Read original