NorskHelsenett/eti-granite-v1
Sentence Similarity • 0.3B • Updated • 37
ETI-Granite
Granite-embedding models fine-tuned on Norwegian health and welfare data (ETI). ModernBERT-based, 32k native context, multilingual.
Note EmbeddingGemma is the production model; granite is an architecture A/B with much longer native context (32k vs 2k). Same training data and recipe, no prompt prefixes (matches granite-r2 base design). Dev NDCG@10 0.8636, test NDCG@10 0.8611.
-
NorskHelsenett/eti-granite-4096-v1
Sentence Similarity • 0.3B • Updated • 1