E5 Large v2
dense 0.335B
Fields
id
models/e5-large-v2Family
e5
Vendor
intfloat (Intel)
Release year
2023
Architecture
dense
Params (B)
0.335
Active params (B)
0.335
Hidden size
1,024
Layers
24
Max context
512
Context scaling method
native
Activation
gelu
Attention variant
bidirectional encoder (BERT), full attention, no causal mask
KV cache (bytes/token)
0
FLOPs per active token
670,000,000
Open weights
yes
License
MIT
embedding dims
1,024
notes
Encoder: kv_cache_bytes_per_token = 0, KV formula does not apply. Identical architecture to bge-large-en-v1.5 and gte-large: BERT-large, hidden 1024, 24 layers, 16 heads, intermediate 4096, vocab 30522, 512 max_position_embeddings. The E5 family's distinguishing feature is the 'query:' and 'passage:' prefix convention required at inference time — failing to add these prefixes degrades retrieval quality significantly. This is a serving gotcha: the prefix must be baked into the embedding pipeline, not added as an afterthought. MIT license. GGUF and ONNX ports exist.
task
embedding
Notes
Encoder: kv_cache_bytes_per_token = 0, KV formula does not apply. Identical architecture to bge-large-en-v1.5 and gte-large: BERT-large, hidden 1024, 24 layers, 16 heads, intermediate 4096, vocab 30522, 512 max_position_embeddings. The E5 family's distinguishing feature is the 'query:' and 'passage:' prefix convention required at inference time — failing to add these prefixes degrades retrieval quality significantly. This is a serving gotcha: the prefix must be baked into the embedding pipeline, not added as an afterthought. MIT license. GGUF and ONNX ports exist.
Cross-references
This record references no other records — it is a leaf. It is cited by 0.