Embedding Models¶

EmbedCache supports 22+ embedding models through the FastEmbed library.

Available Models¶

MiniLM Series¶

Lightweight models suitable for general-purpose use.

Model	Dimensions	Description
`AllMiniLML6V2`	384	Fast, general-purpose
`AllMiniLML6V2Q`	384	Quantized version
`AllMiniLML12V2`	384	Higher quality
`AllMiniLML12V2Q`	384	Quantized version

BGE Series (BAAI)¶

High-quality models from BAAI (Beijing Academy of AI).

Model	Dimensions	Description
`BGESmallENV15`	384	Small, fast, good quality
`BGESmallENV15Q`	384	Quantized version
`BGEBaseENV15`	768	Balanced size/quality
`BGEBaseENV15Q`	768	Quantized version
`BGELargeENV15`	1024	Highest quality BGE
`BGELargeENV15Q`	1024	Quantized version
`BGESmallZHV15`	512	Chinese language

Nomic Series¶

Models from Nomic AI.

Model	Dimensions	Description
`NomicEmbedTextV1`	768	Original Nomic model
`NomicEmbedTextV15`	768	Improved version
`NomicEmbedTextV15Q`	768	Quantized version

Multilingual E5 Series¶

Models supporting multiple languages.

Model	Dimensions	Description
`MultilingualE5Small`	384	Small multilingual
`MultilingualE5Base`	768	Base multilingual
`MultilingualE5Large`	1024	Large multilingual

Paraphrase Series¶

Optimized for paraphrase and similarity detection.

Model	Dimensions	Description
`ParaphraseMLMiniLML12V2`	384	MiniLM paraphrase
`ParaphraseMLMiniLML12V2Q`	384	Quantized version
`ParaphraseMLMpnetBaseV2`	768	MPNet paraphrase

MxbaiEmbed Series¶

Model	Dimensions	Description
`MxbaiEmbedLargeV1`	1024	Large embedding model
`MxbaiEmbedLargeV1Q`	1024	Quantized version

Enabling Models¶

Configure enabled models in your .env file:

ENABLED_MODELS=BGESmallENV15,AllMiniLML6V2,MultilingualE5Base

Memory Usage

Each enabled model consumes memory. Only enable models you need.

Choosing a Model¶

By Use Case¶

Use Case	Recommended Model
General search	`BGESmallENV15`
High accuracy	`BGELargeENV15`
Fast processing	`AllMiniLML6V2`
Multilingual	`MultilingualE5Base`
Low memory	`AllMiniLML6V2Q`

By Language¶

Language	Recommended Model
English	`BGESmallENV15`
Chinese	`BGESmallZHV15`
Multiple	`MultilingualE5Base`

By Resource Constraints¶

Constraint	Recommended Model
Low memory	Quantized models (`*Q`)
Fast inference	`AllMiniLML6V2`
Best quality	`BGELargeENV15`

Model Comparison¶

# Benchmark different models
for model in AllMiniLML6V2 BGESmallENV15 BGEBaseENV15; do
  echo "Testing $model..."
  time curl -s -X POST http://localhost:8081/v1/embed \
    -H "Content-Type: application/json" \
    -d "{
      \"text\": [\"Test sentence for benchmarking.\"],
      \"config\": {
        \"chunking_type\": \"words\",
        \"chunking_size\": 512,
        \"embedding_model\": \"$model\"
      }
    }" > /dev/null
done

Quantized Models¶

Models ending in Q are quantized versions that:

Use less memory
Are slightly faster
Have marginally lower quality

Use quantized models when:

Running on memory-constrained systems
Processing high volumes
Quality requirements are flexible