arXiv benchmark finds int8 quantization nearly matches float32 on text embeddings
int8 quantization retains 94.4% baseline effectiveness at 32x storage reduction across 38 MTEB tasks.
Stories per day · 30 days
1 stories, newest first · one card per story