BGE Multilingual Gemma2
BAAI · BAAI/bge-multilingual-gemma2
Weights license: Publisher terms: Gemma Terms of Use · code and components may use other terms
Quality
Not in the MTEB ranking
The MTEB results dataset covers 24 of 131 tasks of MTEB (Multilingual, v2); a mean over a partial task set is not comparable, so none is computed.
Reported by the publisher
- MTEB NFCorpus (test; main_score)
- 38.114Retrieval
- MTEB MSMARCO (dev; main_score)
- 45.707Retrieval
- MTEB FiQA2018 (test; main_score)
- 60.042Retrieval
- MTEB SCIDOCS (test; main_score)
- 26.932Retrieval
- MTEB FEVER (test; main_score)
- 90.382Retrieval
- MTEB ArguAna (test; main_score)
- 77.366Retrieval
- MTEB SciFact (test; main_score)
- 72.047Retrieval
- MTEB TRECCOVID (test; main_score)
- 64.269Retrieval
Every reported result
- MTEB ClimateFEVER (test; main_score)
- 39.372Retrieval
- MTEB HotpotQA (test; main_score)
- 83.263Retrieval
- MTEB NQ (test; main_score)
- 71.447Retrieval
- MTEB QuoraRetrieval (test; main_score)
- 90.038Retrieval
- MTEB Touche2020 (test; main_score)
- 30.256Retrieval
- MTEB DBPedia (test; main_score)
- 51.369Retrieval
- MTEB CQADupstackRetrieval (test; main_score)
- 47.939Retrieval
- MTEB AmazonCounterfactualClassification (en) (en; test; main_score)
- 89.478Classification
- MTEB AmazonPolarityClassification (test; main_score)
- 96.896Classification
- MTEB AmazonReviewsClassification (en) (en; test; main_score)
- 61.602Classification
- MTEB ArxivClusteringP2P (test; main_score)
- 54.906Clustering
- MTEB ArxivClusteringS2S (test; main_score)
- 50.278Clustering
- MTEB AskUbuntuDupQuestions (test; main_score)
- 64.595Reranking
- MTEB BIOSSES (test; main_score)
- 85.745STS
- MTEB Banking77Classification (test; main_score)
- 92.532Classification
- MTEB BiorxivClusteringP2P (test; main_score)
- 52.641Clustering
- MTEB BiorxivClusteringS2S (test; main_score)
- 49.197Clustering
- MTEB EmotionClassification (test; main_score)
- 92.975Classification
- MTEB ImdbClassification (test; main_score)
- 96.665Classification
- MTEB MTOPDomainClassification (en) (en; test; main_score)
- 98.611Classification
- MTEB MTOPIntentClassification (en) (en; test; main_score)
- 95.513Classification
- MTEB MassiveIntentClassification (en) (en; test; main_score)
- 82.048Classification
- MTEB MassiveScenarioClassification (en) (en; test; main_score)
- 84.401Classification
- MTEB MedrxivClusteringP2P (test; main_score)
- 45.809Clustering
- MTEB MedrxivClusteringS2S (test; main_score)
- 44.107Clustering
- MTEB MindSmallReranking (test; main_score)
- 31.794Reranking
- MTEB RedditClustering (test; main_score)
- 56.033Clustering
- MTEB RedditClusteringP2P (test; main_score)
- 65.832Clustering
- MTEB SICK-R (test; main_score)
- 82.661STS
- MTEB STS12 (test; main_score)
- 77.711STS
- MTEB STS13 (test; main_score)
- 87.447STS
- MTEB STS14 (test; main_score)
- 83.483STS
- MTEB STS15 (test; main_score)
- 87.631STS
- MTEB STS16 (test; main_score)
- 86.702STS
- MTEB STS17 (en-en) (en-en; test; main_score)
- 91.183STS
- MTEB STS22 (en) (en; test; main_score)
- 69.023STS
- MTEB STSBenchmark (test; main_score)
- 87.252STS
- MTEB SciDocsRR (test; main_score)
- 87.603Reranking
- MTEB SprintDuplicateQuestions (test; cos_sim_ap)
- 90.940PairClassification
- MTEB SprintDuplicateQuestions (test; max_ap)
- 91.233PairClassification
- MTEB StackExchangeClustering (test; main_score)
- 66.210Clustering
- MTEB StackExchangeClusteringP2P (test; main_score)
- 45.743Clustering
- MTEB StackOverflowDupQuestions (test; main_score)
- 54.903Reranking
- MTEB SummEval (test; main_score)
- 31.196Summarization
- MTEB ToxicConversationsClassification (test; main_score)
- 87.343Classification
- MTEB TweetSentimentExtractionClassification (test; main_score)
- 78.862Classification
- MTEB TwentyNewsgroupsClustering (test; main_score)
- 70.443Clustering
- MTEB TwitterSemEval2015 (test; cos_sim_ap)
- 79.639PairClassification
- MTEB TwitterSemEval2015 (test; max_ap)
- 79.677PairClassification
- MTEB TwitterURLCorpus (test; cos_sim_ap)
- 86.952PairClassification
- MTEB TwitterURLCorpus (test; max_ap)
- 87.008PairClassification
- MTEB AFQMC (validation; main_score)
- 47.173STS
- MTEB ATEC (test; main_score)
- 50.750STS
- MTEB AmazonReviewsClassification (zh) (zh; test; main_score)
- 54.340Classification
- MTEB BQ (test; main_score)
- 62.022STS
- MTEB CLSClusteringP2P (test; main_score)
- 54.645Clustering
- MTEB CLSClusteringS2S (test; main_score)
- 63.683Clustering
- MTEB CMedQAv1 (test; main_score)
- 84.615Reranking
- MTEB CMedQAv2 (test; main_score)
- 85.601Reranking
- MTEB CmedqaRetrieval (dev; main_score)
- 42.208Retrieval
- MTEB Cmnli (validation; cos_sim_ap)
- 90.133PairClassification
- MTEB Cmnli (validation; max_ap)
- 90.149PairClassification
- MTEB CovidRetrieval (dev; main_score)
- 77.458Retrieval
- MTEB DuRetrieval (dev; main_score)
- 90.462Retrieval
- MTEB EcomRetrieval (dev; main_score)
- 69.298Retrieval
- MTEB IFlyTek (validation; main_score)
- 49.942Classification
- MTEB JDReview (test; main_score)
- 88.912Classification
- MTEB LCQMC (test; main_score)
- 75.955STS
- MTEB MMarcoReranking (dev; main_score)
- 35.433Reranking
- MTEB MMarcoRetrieval (dev; main_score)
- 84.701Retrieval
- MTEB MassiveIntentClassification (zh-CN) (zh-CN; test; main_score)
- 78.191Classification
- MTEB MassiveScenarioClassification (zh-CN) (zh-CN; test; main_score)
- 82.582Classification
- MTEB MedicalRetrieval (dev; main_score)
- 62.022Retrieval
- MTEB MultilingualSentiment (validation; main_score)
- 78.913Classification
- MTEB Ocnli (validation; cos_sim_ap)
- 83.207PairClassification
- MTEB Ocnli (validation; max_ap)
- 83.602PairClassification
- MTEB OnlineShopping (test; main_score)
- 94.59Classification
- MTEB PAWSX (test; main_score)
- 30.570STS
- MTEB QBQTC (test; main_score)
- 38.981STS
- MTEB STS22 (zh) (zh; test; main_score)
- 68.682STS
- MTEB STSB (test; main_score)
- 80.871STS
- MTEB T2Reranking (dev; main_score)
- 67.479Reranking
- MTEB T2Retrieval (dev; main_score)
- 86.259Retrieval
- MTEB TNews (validation; main_score)
- 50.255Classification
- MTEB ThuNewsClusteringP2P (test; main_score)
- 64.321Clustering
- MTEB ThuNewsClusteringS2S (test; main_score)
- 54.571Clustering
- MTEB VideoRetrieval (dev; main_score)
- 77.397Retrieval
- MTEB Waimai (test; main_score)
- 89.26Classification
- MTEB AlloProfClusteringP2P (test; main_score)
- 71.203Clustering
- MTEB AlloProfClusteringS2S (test; main_score)
- 59.642Clustering
- MTEB AlloprofReranking (test; main_score)
- 78.620Reranking
- MTEB AlloprofRetrieval (test; main_score)
- 58.499Retrieval
- MTEB AmazonReviewsClassification (fr) (fr; test; main_score)
- 55.186Classification
- MTEB BSARDRetrieval (test; main_score)
- 65.766Retrieval
- MTEB HALClusteringS2S (test; main_score)
- 28.187Clustering
- MTEB MLSUMClusteringP2P (fr) (fr; test; main_score)
- 47.750Clustering
- MTEB MLSUMClusteringS2S (fr) (fr; test; main_score)
- 47.461Clustering
- MTEB MTOPDomainClassification (fr) (fr; test; main_score)
- 97.203Classification
- MTEB MTOPIntentClassification (fr) (fr; test; main_score)
- 93.072Classification
- MTEB MasakhaNEWSClassification (fra) (fra; test; main_score)
- 82.488Classification
- MTEB MasakhaNEWSClusteringP2P (fra) (fra; test; main_score)
- 73.857Clustering
- MTEB MasakhaNEWSClusteringS2S (fra) (fra; test; main_score)
- 67.240Clustering
- MTEB MassiveIntentClassification (fr) (fr; test; main_score)
- 79.600Classification
- MTEB MassiveScenarioClassification (fr) (fr; test; main_score)
- 82.182Classification
- MTEB MintakaRetrieval (fr) (fr; test; main_score)
- 62.532Retrieval
- MTEB OpusparcusPC (fr) (fr; test; main_score)
- 100PairClassification
- MTEB PawsXPairClassification (fr) (fr; test; main_score)
- 70.142PairClassification
- MTEB SICKFr (test; main_score)
- 79.392STS
- MTEB STS22 (fr) (fr; test; main_score)
- 83.281STS
- MTEB STSBenchmarkMultilingualSTS (fr) (fr; test; main_score)
- 85.088STS
- MTEB SummEvalFr (test; main_score)
- 31.260Summarization
- MTEB SyntecReranking (test; main_score)
- 91.833Reranking
- MTEB SyntecRetrieval (test; main_score)
- 90.367Retrieval
- MTEB XPQARetrieval (fr) (fra-fra; test; main_score)
- 77.425Retrieval
- MTEB AllegroReviews (test; main_score)
- 65Classification
- MTEB ArguAna-PL (test; main_score)
- 59.713Retrieval
- MTEB CBD (test; main_score)
- 84.130Classification
- MTEB CDSC-E (test; main_score)
- 78.517PairClassification
- MTEB CDSC-R (test; main_score)
- 90.968STS
- MTEB DBPedia-PL (test; main_score)
- 43.189Retrieval
- MTEB 8TagsClustering (test; main_score)
- 50.288Clustering
- MTEB FiQA-PL (test; main_score)
- 46.121Retrieval
- MTEB HotpotQA-PL (test; main_score)
- 77.028Retrieval
- MTEB MSMARCO-PL (test; main_score)
- 72.695Retrieval
- MTEB MassiveIntentClassification (pl) (pl; test; main_score)
- 79.415Classification
- MTEB MassiveScenarioClassification (pl) (pl; test; main_score)
- 81.933Classification
- MTEB NFCorpus-PL (test; main_score)
- 36.723Retrieval
- MTEB NQ-PL (test; main_score)
- 56.854Retrieval
- MTEB PAC (test; main_score)
- 67.243Classification
- MTEB PSC (test; main_score)
- 99.286PairClassification
- MTEB PolEmo2.0-IN (test; main_score)
- 90.416Classification
- MTEB PolEmo2.0-OUT (test; main_score)
- 77.773Classification
- MTEB PPC (test; cos_sim_ap)
- 95.433PairClassification
- MTEB PPC (test; max_ap)
- 95.433PairClassification
- MTEB Quora-PL (test; main_score)
- 84.471Retrieval
- MTEB SCIDOCS-PL (test; main_score)
- 19.528Retrieval
- MTEB SICK-E-PL (test; main_score)
- 85.972PairClassification
- MTEB SICK-R-PL (test; main_score)
- 78.159STS
- MTEB STS22 (pl) (pl; test; main_score)
- 42.791STS
- MTEB SciFact-PL (test; main_score)
- 74.434Retrieval
- MTEB TRECCOVID-PL (test; main_score)
- 82.749Retrieval
Values as printed on the card, with the publisher's own settings. Not comparable across publishers and not used in any ranking. Model card
Quality × memory
No comparable quality × memory pair
Embedding memory estimates cluster at a few GB, so a quality × memory chart would not separate these models; MTEB is ranked above instead.
Runs on
Fast memory holds it 82
The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.
Show all 82
One memory tier short 11
The same machines with the next memory size up would hold it in fast memory.
Show all 11
Hardware profiles grouped by the embedding planning estimate (8-bit weights, 20% allowance, 2 GB reserve). Not a measured fit.
Consider BGE Multilingual Gemma2 for retrieval and semantic search: it turns text into 3,584-dimension vectors.
The publisher states no memory figure; the hardware list uses a ~14 GB planning estimate, a ceiling and not a measurement. See the prompt format under How to run. PT-BR is not confirmed by a source.
Measured speed
No published measurement for this checkpoint
Throughput (texts per second) is shown only when a source publishes it with hardware, build and setup. It is never estimated.
How to run
The planning estimate uses 8-bit weights, 20% working allowance and a 2 GB reserve; it is a ceiling, not a measurement.
- Runs in
- Sentence Transformers, Transformers, FlagEmbeddingsource
- Query/document prompts
- Queries: prepend the instruction in the form '<instruct>{task}\n<query>{query}' (for example 'Given a web search query, retrieve relevant passages that answer the query.'). Documents: no instruction.
- Languages · PT-BR
- Multilingual; the card names English, Chinese, Japanese, Korean, French 'and more' (benchmarks: MIRACL, MTEB-pl, MTEB-fr)PT-BR unconfirmed
Documented ways to run it
Copied as printed in the linked source (read 2026-10-18); not run by this site.
from sentence_transformers import SentenceTransformer
import torch
# Load the model, optionally in float16 precision for faster inference
model = SentenceTransformer("BAAI/bge-multilingual-gemma2", model_kwargs={"torch_dtype": torch.float16})
# Prepare a prompt given an instruction
instruction = 'Given a web search query, retrieve relevant passages that answer the query.'
prompt = f'<instruct>{instruction}\n<query>'
# Prepare queries and documents
queries = [
'how much protein should a female eat',
'summit define',
]
documents = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments."
]
# Compute the query and document embeddings
query_embeddings = model.encode(queries, prompt=prompt)
document_embeddings = model.encode(documents)
# Compute the cosine similarity between the query and document embeddings
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
# tensor([[ 0.5591, 0.0164],
# [-0.0026, 0.4993]], dtype=torch.float16)git clone https://github.com/FlagOpen/FlagEmbedding.git
cd FlagEmbedding
pip install -e .from FlagEmbedding import FlagLLMModel
queries = ["how much protein should a female eat", "summit define"]
documents = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments."
]
model = FlagLLMModel('BAAI/bge-multilingual-gemma2',
query_instruction_for_retrieval="Given a web search query, retrieve relevant passages that answer the query.",
use_fp16=True) # Setting use_fp16 to True speeds up computation with a slight performance degradation
embeddings_1 = model.encode_queries(queries)
embeddings_2 = model.encode_corpus(documents)
similarity = embeddings_1 @ embeddings_2.T
print(similarity)
# [[ 0.559 0.01654 ]
# [-0.002575 0.4998 ]]import torch
import torch.nn.functional as F
from torch import Tensor
from transformers import AutoTokenizer, AutoModel
def last_token_pool(last_hidden_states: Tensor,
attention_mask: Tensor) -> Tensor:
left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0])
if left_padding:
return last_hidden_states[:, -1]
else:
sequence_lengths = attention_mask.sum(dim=1) - 1
batch_size = last_hidden_states.shape[0]
return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths]
def get_detailed_instruct(task_description: str, query: str) -> str:
return f'<instruct>{task_description}\n<query>{query}'
task = 'Given a web search query, retrieve relevant passages that answer the query.'
queries = [
get_detailed_instruct(task, 'how much protein should a female eat'),
get_detailed_instruct(task, 'summit define')
]
# No need to add instructions for documents
documents = [
"As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
"Definition of summit for English Language Learners. : 1 the highest point of a mountain : the top of a mountain. : 2 the highest level. : 3 a meeting or series of meetings between the leaders of two or more governments."
]
input_texts = queries + documents
tokenizer = AutoTokenizer.from_pretrained('BAAI/bge-multilingual-gemma2')
model = AutoModel.from_pretrained('BAAI/bge-multilingual-gemma2')
model.eval()
max_length = 4096
# Tokenize the input texts
batch_dict = tokenizer(input_texts, max_length=max_length, padding=True, truncation=True, return_tensors='pt', pad_to_multiple_of=8)
with torch.no_grad():
outputs = model(**batch_dict)
embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
# normalize embeddings
embeddings = F.normalize(embeddings, p=2, dim=1)
scores = (embeddings[:2] @ embeddings[2:].T) * 100
print(scores.tolist())
# [[55.92064666748047, 1.6549524068832397], [-0.2698777914047241, 49.95653533935547]]Technical details
What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.
- Checkpoint / source
- BAAI/bge-multilingual-gemma2
- Task
- embeddings for search (feature extraction)
- Library in source
- Sentence Transformers, Transformers, FlagEmbedding
- Weight license
- Publisher terms: Gemma Terms of Use
- Commercial use
- Allowed Gemma Terms of Use (last modified 2025-03-24; verbatim copy in the ScanCode LicenseDB, homepage ai.google.dev/gemma/terms) · read 2026-10-18license text
- Parameter basis
- Hub safetensors total (9,241,713,152 parameters, read 2026-10-09 from the Hub API).
- Announced
- 2024-07-26
Sources
Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.
- Publisher model card
- Hugging Face model page / card · 2026-10-09
- Hugging Face API (parameters, licence, library) · 2026-10-09
- MTEB results: BAAI/bge-multilingual-gemma2 model_meta.json (GitHub) · 2026-10-09
- BAAI/bge-multilingual-gemma2 config.json (Hugging Face) · 2026-10-09
- FlagEmbedding README (GitHub, FlagOpen): 7/26/2024 Release a new embedding model bge-multilingual-gemma2 · 2026-10-10
Curated facts checked: 2026-10-09. Each value cites its source (the publisher's card or the MTEB results dataset); estimates are labelled as estimates, and a value no source states stays empty. A repository commit is not a model release date.
















































