BGE Reranker Large
dense 0.578B
Fields
id
models/bge-reranker-largeFamily
bge-reranker
Vendor
BAAI
Release year
2023
Architecture
dense
Params (B)
0.578
Active params (B)
0.578
Hidden size
1,024
Layers
24
Max context
514
Context scaling method
native
Activation
gelu
Attention variant
cross-encoder (XLM-RoBERTa), full bidirectional attention over query+passage
KV cache (bytes/token)
0
FLOPs per active token
1,160,000,000
Open weights
yes
License
MIT
notes
RERANKER: cross-encoder, kv_cache_bytes_per_token = 0, KV formula does not apply. XLM-RoBERTa-large backbone (hidden 1024, 24 layers, 16 heads, vocab 250002, 514 max_position_embeddings). N passages = N independent forward passes over (query+passage) pairs with no KV reuse. Cost scales linearly with number of passages. The 514-token limit is the binding constraint — long passages must be truncated before reranking. MIT license. GGUF and ONNX ports exist.
task
reranking
Notes
RERANKER: cross-encoder, kv_cache_bytes_per_token = 0, KV formula does not apply. XLM-RoBERTa-large backbone (hidden 1024, 24 layers, 16 heads, vocab 250002, 514 max_position_embeddings). N passages = N independent forward passes over (query+passage) pairs with no KV reuse. Cost scales linearly with number of passages. The 514-token limit is the binding constraint — long passages must be truncated before reranking. MIT license. GGUF and ONNX ports exist.
Cross-references
This record references no other records — it is a leaf. It is cited by 0.