GTE Reranker ModernBERT
dense 0.149B
Fields
id
models/gte-reranker-modernbertFamily
gte-reranker
Vendor
Alibaba (Alibaba-NLP)
Release year
2024
Architecture
dense
Params (B)
0.149
Active params (B)
0.149
Hidden size
768
Layers
22
Max context
8,192
Context scaling method
native
Activation
gelu
Attention variant
cross-encoder (ModernBERT), full bidirectional attention over query+passage with local+global attention pattern
KV cache (bytes/token)
0
FLOPs per active token
298,000,000
Open weights
yes
License
Apache-2.0
notes
RERANKER: cross-encoder, kv_cache_bytes_per_token = 0, KV formula does not apply. ModernBERT architecture (not BERT): hidden 768, 22 layers, 12 heads, intermediate 1152, vocab 50368, 8192 max_position_embeddings. Uses local attention (128 tokens) with global attention every 3 layers — this is the ModernBERT efficiency innovation. 149M params, much smaller than bge-reranker-large (578M) but with 8192 context vs 514. Apache-2.0. GGUF and ONNX ports exist.
task
reranking
Notes
RERANKER: cross-encoder, kv_cache_bytes_per_token = 0, KV formula does not apply. ModernBERT architecture (not BERT): hidden 768, 22 layers, 12 heads, intermediate 1152, vocab 50368, 8192 max_position_embeddings. Uses local attention (128 tokens) with global attention every 3 layers — this is the ModernBERT efficiency innovation. 149M params, much smaller than bge-reranker-large (578M) but with 8192 context vs 514. Apache-2.0. GGUF and ONNX ports exist.
Cross-references
This record references no other records — it is a leaf. It is cited by 0.