bge-large-en-v1.5

by BAAI

BAAI General Embedding, state-of-the-art text retrieval

7.1Mdl/month

643likes

335Mparams

HuggingFace Use in Pipeline

Identifiers

Model ID

BAAI/bge-large-en-v1.5

Feature URI

mixpeek://text_extractor@v1/baai_bge_large_v1

Overview

BGE (BAAI General Embedding) is a family of text embedding models that achieve top performance on the MTEB benchmark. The large-en-v1.5 variant produces 1024-dimensional embeddings optimized for English text retrieval and semantic similarity.

On Mixpeek, BGE powers text-based semantic search over extracted text content, transcriptions, captions, OCR results, and document text.

Architecture

BERT-Large architecture (24 layers, 1024-dim hidden, 16 attention heads) with task-specific training using contrastive learning on curated text pairs. Uses [CLS] token pooling with optional instruction prefix.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "my-collection",
  source: { url: "https://example.com/report.pdf" },
  feature_extractors: [{
    name: "text_embedding",
    version: "v1",
    params: {
      model_id: "BAAI/bge-large-en-v1.5"
    }
  }]
});

Capabilities

1024-dimensional dense text embeddings
Top-ranked on MTEB retrieval benchmarks
Instruction-aware embedding with task prefixes
Optimized for asymmetric retrieval (query vs. passage)

Use Cases on Mixpeek

Semantic search over transcribed audio/video content

Document similarity and deduplication

RAG pipeline embedding backend

Cross-document concept matching

Benchmarks

Dataset	Metric	Score	Source
MTEB (56 datasets)	Avg Score	64.23	MTEB Leaderboard — bge-large-en-v1.5
MS MARCO (Passage)	MRR@10	41.2	Xiao et al., 2024 — Table 3
NLI (STS)	Spearman	86.4	MTEB Leaderboard