Skip to content
Most AI tasks no longer need a frontier model Read launch post

Choosing a Model

SIE supports 150+ models; the catalog’s On Serverless filter lists the ones you can call on SIE Serverless today. This guide helps you pick the right one based on your use case, language requirements, and performance needs.

Use CaseRecommended ModelWhy
English-only, balancedNovaSearch/stella_en_400M_v5Strong MTEB scores, efficient size
English-only, max qualitynvidia/NV-Embed-v2Top MTEB scores, 4096 dims
Speed-optimizedsentence-transformers/all-MiniLM-L6-v222M params, 384 dims, very fast
MultilingualBAAI/bge-m3100+ languages, also supports sparse + multivector
Hybrid searchBAAI/bge-m3 or naver/splade-v3Dense + sparse from one model, or dedicated sparse
Late interaction (ColBERT)jinaai/jina-colbert-v2Best ColBERT quality, multilingual
Vision / image searchgoogle/siglip-so400m-patch14-384Image-text similarity
Multilingual, fastQwen/Qwen3-Embedding-0.6B1024 dims, 32K context, 100+ languages
Document vision (PDF)vidore/colpali-v1.3-hfVisual document retrieval
ColBERT rerankinganswerdotai/answerai-colbert-small-v1Fast MaxSim reranking; also jina-colbert-v2, GTE-ModernColBERT-v1
Reranking (multilingual)BAAI/bge-reranker-v2-m3Strong cross-language reranking
Reranking (English)mixedbread-ai/mxbai-rerank-large-v2High quality, 8192 max length
Entity extractionurchade/gliner_multi-v2.1Zero-shot NER, multilingual

Use CaseScenarioRecommended Models
Semantic Search / RAGEnglish-onlystella_en_400M_v5, NV-Embed-v2, all-MiniLM-L6-v2
MultilingualBAAI/bge-m3
Hybrid (dense + sparse)BAAI/bge-m3 + naver/splade-v3
Image SearchText ↔ ImageSigLIP, CLIP
Visual docsColPali
RerankingMultilingualBAAI/bge-reranker-v2-m3
Englishmixedbread-ai/mxbai-rerank-large-v2
Entity ExtractionNERGLiNER
RelationsGLiREL
ClassificationGLiClass

ModelParamsDimsVRAMRelative SpeedQuality
all-MiniLM-L6-v222M384~200MBFastestGood
stella_en_400M_v5400M1024~1.5GBFastVery good
bge-m3568M1024~2GBFastVery good
NV-Embed-v27B4096~14GBSlowBest

Rule of thumb: For English, start with stella_en_400M_v5. For multilingual or hybrid search, use BAAI/bge-m3. Only move to 7B+ models if benchmarks show a meaningful gap on your data.

Output TypeStorageSearch SpeedQualityBest For
DenseSmall (1024 floats)FastGoodStandard semantic search
SparseVariableFastGood for keywordsHybrid search, keyword matching
Multi-vector (ColBERT)Large (N * 128 floats)SlowerBestWhen accuracy is critical

Recommendation: Use dense for most cases. Add sparse for hybrid search if you need keyword matching. Use multi-vector only when you need the best possible retrieval quality and can afford the storage.


Language NeedModels
English onlyStella, NV-Embed-v2, all-MiniLM, GTE-Qwen2
Multilingual (100+ languages)BGE-M3, multilingual-e5-large, Qwen3-Embedding-0.6B
Chinese-focusedGTE-Qwen2, BGE-M3

GPUVRAMModels That Fit
T416GBMost models up to ~1B params
L424GBAll standard models; several can stay resident, depending on model size and batch settings
A100 40GB40GBLarge models, with more headroom for concurrent residency
A100 80GB80GB7B+ parameter models (NV-Embed-v2, e5-mistral-7b)

With LRU eviction, a single GPU can serve the whole catalog over time: only the most recently used models stay in memory, and each requested model still has to fit in the GPU’s VRAM.


Almost always. Two-stage retrieval (retrieve with embeddings, then rerank with a cross-encoder) consistently improves quality:

  1. Retrieve 20-50 candidates with dense embeddings (fast)
  2. Rerank to top 5-10 with a cross-encoder (more accurate)

The reranker sees both query and document together, enabling deeper semantic matching than embedding similarity alone.

# Stage 1: Fast retrieval
results = vector_db.search(query_embedding, k=20)
# Stage 2: Accurate reranking
reranked = client.score(
"mixedbread-ai/mxbai-rerank-large-v2",
query=Item(text="What is machine learning?"),
items=[Item(text=r.text) for r in results]
)

Add sparse embeddings when your data has:

  • Domain-specific terminology that dense models might miss
  • Exact keyword matching requirements (product codes, identifiers)
  • Mixed content where some queries are keyword-like and others are semantic
# Get both dense and sparse from one model
result = client.encode(
"BAAI/bge-m3",
Item(text="your text"),
output_types=["dense", "sparse"]
)
# Use dense for semantic search, sparse for keyword matching
# Combine scores for hybrid retrieval

Talk to us

Tell us about your use case and we'll get back to you shortly.