BGE-M3 Vietnamese fine-tune
According to the model card, the model is fine-tuned from BAAI/bge-m3 specifically to enhance Vietnamese retrieval capability.
Open Source Model Profile · AITeamVN
Vietnamese_Embedding
Vietnamese_Embedding is a 568M-parameter Vietnamese sentence-embedding model from AITeamVN. According to the model card, it is fine-tuned from BAAI/bge-m3 to strengthen Vietnamese retrieval.
- Publisher
- AITeamVN
- Task
- sentence-similarity
- Model type
- xlm-roberta
- License
- apache-2.0
- Library
- sentence-transformers
- Publication status
- Accepted · not indexed
Model overview
Vietnamese_Embedding is published by AITeamVN as a sentence-similarity embedding model. Captured Safetensors metadata reports 567,754,752 parameters, about 568M. According to the model card, it is a Sentence Transformer fine-tuned from BAAI/bge-m3 on approximately 300,000 Vietnamese query-document triplets with a 2048-token maximum sequence length.
Recorded capabilities
2048-token inputs, 1024 dimensions
According to the model card, it supports a 2048-token maximum sequence length with 1024-dimension output and dot-product similarity.
Triplet-trained retrieval
According to the model card, training used approximately 300,000 triplets of queries with positive and negative Vietnamese documents.
Publisher-reported Legal Zalo scores
According to the model card's table, the public checkpoint reports 0.7274 Accuracy@1 and 0.8181 MRR@10, above the listed BGE-M3 baseline.
Use cases in the source record
- Vietnamese passage retrieval and semantic search using the documented 1024-dimension dot-product similarity.
- Legal-domain retrieval evaluation on the Legal Zalo 2021 training set, which the card states was not used for training.
Limitations and unknowns
- No independently measured evaluation results were extracted; Legal Zalo figures are publisher-reported card claims.
- No VRAM, hardware, or pricing detail was extracted from this record.
- The evaluation table also covers sibling reranker and v2 checkpoints rather than this checkpoint alone.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: AITeamVN/Vietnamese_Embedding
Captured: Unknown. Processed: 2026-09-07T19:35:01.894233+00:00.
Model Card: Vietnamese_Embedding Vietnamese_Embedding is an embedding model fine-tuned from the BGE-M3 model ( https://huggingface.co/BAAI/bge-m3 ) to enhance retrieval capabilities for Vietnamese. The model was trained on approximately 300,000 triplets of queries, positive documents, and negative documents for Vietnamese. The model was trained with a maximum sequence length of 2048. Model Details Model Description Model Type: Sentence Transformer Base model: BAAI/bge-m3 Maximum Sequence Length: 2048 tokens Output Dimensionality: 1024 dimensions Similarity Function: Dot product Similarity Language: Vietnamese Licence: Apache 2.0 Usa…
F001F002F003F004F005F006F007F009F010F011F012F013F014F015F016