all-mpnet-base-v2
dense 0.109B
Fields
id
models/all-mpnet-base-v2Family
sentence-transformers
Vendor
Sentence Transformers (Microsoft)
Release year
2021
Architecture
dense
Params (B)
0.109
Active params (B)
0.109
Hidden size
768
Layers
12
Max context
514
Context scaling method
native
Activation
gelu
Attention variant
bidirectional encoder (MPNet), full attention, no causal mask, relative position bias
KV cache (bytes/token)
0
FLOPs per active token
219,000,000
Open weights
yes
License
Apache-2.0
embedding dims
768
notes
Encoder: kv_cache_bytes_per_token = 0, KV formula does not apply. MPNet architecture (not BERT): hidden 768, 12 layers, 12 heads, intermediate 3072, vocab 30527, max_position_embeddings 514. Uses relative position bias (32 buckets) instead of absolute position embeddings. Same parameter count as bge-base and gte-base (109M) but different architecture. Apache-2.0. GGUF and ONNX ports exist. The sentence-transformers library default for higher quality.
Notes
Encoder: kv_cache_bytes_per_token = 0, KV formula does not apply. MPNet architecture (not BERT): hidden 768, 12 layers, 12 heads, intermediate 3072, vocab 30527, max_position_embeddings 514. Uses relative position bias (32 buckets) instead of absolute position embeddings. Same parameter count as bge-base and gte-base (109M) but different architecture. Apache-2.0. GGUF and ONNX ports exist. The sentence-transformers library default for higher quality.
Cross-references
This record references no other records — it is a leaf. It is cited by 0.