Test_bge_raw_finmodern_tri_merged
a model for financial semantic search
1 min readupdated
This model is a specialized version of the popular BAAI/bge-m3 language architecture, fine-tuned for text matching and document retrieval in the financial sector. It converts sentences and paragraphs into dense 1024-dimensional vectors, using cosine similarity to evaluate relevance. Developer finvector trained it on a dataset of financial document triplets. It is an excellent choice for analysts and developers building search systems for company reports and financial documentation.
Key features
- Built on the powerful BAAI/bge-m3 base with the XLMRobertaModel architecture.
- Maximum sequence length is 512 tokens.
- Output vector dimensionality is 1024.
- Trained using triplet matching (query, positive, negative).
What you can do
- Set up semantic search across financial reports, contracts, and corporate documentation.
- Integrate the model into an enterprise RAG (retrieval-augmented generation) pipeline for precise document-based question answering.
- Compare the similarity of financial statements and paragraphs using cosine distance.
Example
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("finvector/Test_bge_raw_finmodern_tri_merged")
queries = ['Adobe reports revenue outside the Americas...']
documents = ['As a global business that generates...']
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)Read more
DetailsMetrics: cosine_accuracy — 0.8667 on the test dataset.
Limitations: maximum text length is limited to 512 tokens.
License: apache-2.0.
Model size: ~0.57 billion parameters.










Comments