Synthetic data for text embedders yields sparse, task-localized MTEB gains and cross-task trade-offs, not broad robust improvement.
We use the AdamW optimizer with a learning rate of4e−4, linear learn- ing rate warm-up for the first 100 steps, and weight decay with 0.1 coefficient afterwards
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Understanding the Influence of Synthetic Data for Text Embedders
Synthetic data for text embedders yields sparse, task-localized MTEB gains and cross-task trade-offs, not broad robust improvement.