andreaschari
/

bge-m3-RU_MMARCO_50_MIXED

Sentence Similarity

sentence-transformers

feature-extraction

text-embeddings-inference

Model card Files Files and versions

BGE-m3 RU mMARCO/v2 50/50 Native Transliterated Queries

This is a BGE-M3 model post-trained on the Russian dataset from MMARCO/v2. The queries are a 50/50 split between native Russian and transliterated Russian to English text using uroman.

The model was used for the SIGIR 2025 Short paper: Lost in Transliteration: Bridging the Script Gap in Neural IR.

Model Details

Model Description

Model Type: Sentence Transformer
Maximum Sequence Length: 8192 tokens
Output Dimensionality: 1024 tokens
Similarity Function: Cosine Similarity

Training Details

Framework Versions

Python: 3.10.13
Sentence Transformers: 3.1.1
Transformers: 4.45.1
PyTorch: 2.4.1
Accelerate: 0.34.2
Datasets: 3.0.1
Tokenizers: 0.20.3

Downloads last month: 1

Safetensors

Model size

0.6B params

Tensor type

F32

·

Model tree for andreaschari/bge-m3-RU_MMARCO_50_MIXED

Base model

Finetuned

(447)

this model

Dataset used to train andreaschari/bge-m3-RU_MMARCO_50_MIXED

Collection including andreaschari/bge-m3-RU_MMARCO_50_MIXED

SIGIR2025: Lost in Transliteration

Model Checkpoints and Data for the SIGIR Short Paper: Lost in Transliteration: Bridging the Script Gap in Neural IR • 19 items • Updated Apr 24, 2025