CatsuCatsu Docs

Mixed Bread

Mixed Bread embedding provider documentation

Mixed Bread (mxbai) provides multilingual embedding models with comprehensive quantization support.

Overview

  • Models: 4 models (large variants, German-specific models)
  • Key Features: Multilingual, quantization (float/int8/binary), custom prompts
  • API Docs: Mixed Bread Documentation

Environment Variable

export MIXEDBREAD_API_KEY="your-mixedbread-api-key"

Supported Parameters

ParameterTypeRequiredDescription
modelstrYesModel identifier
inputstr | List[str]YesText(s) to embed
input_typestrNo"query" or "document" (prompt injection)
dimensionsintNoOutput dimensions (Matryoshka)
normalizedboolNoL2 normalize embeddings (default: True)
encoding_formatstrNo"float", "binary", "int8", etc.
promptstrNoCustom prompt for embedding
api_keystrNoOverride API key

Examples

Basic Usage

response = client.embed(
    model="mxbai-embed-large-v1",
    input="Hello, Mixed Bread!"
)

With input_type

# Query embeddings
query_response = client.embed(
    model="mxbai-embed-large-v1",
    input="What is NLP?",
    input_type="query"
)

# Document embeddings
doc_response = client.embed(
    model="mxbai-embed-large-v1",
    input="NLP stands for...",
    input_type="document"
)

With Quantization

# Int8 quantization
int8_response = client.embed(
    model="mxbai-embed-large-v1",
    input="Text",
    encoding_format="int8"
)

# Binary quantization
binary_response = client.embed(
    model="mxbai-embed-large-v1",
    input="Text",
    encoding_format="binary"
)

German-Specific Models

# German language model
response = client.embed(
    model="deepset-mxbai-embed-de-large-v1",
    input="Hallo, wie geht es dir?"
)

With Custom Dimensions (Matryoshka)

response = client.embed(
    model="mxbai-embed-large-v1",
    input="Sample text",
    dimensions=512
)

Model Variants

  • mxbai-embed-large-v1 - General-purpose, 1024d
  • mxbai-embed-2d-large-v1 - 2D variant, 768d
  • deepset-mxbai-embed-de-large-v1 - German-specific, 1024d
  • deepset-mxbai-embed-de-large-en-v1 - German-English bilingual

For pricing, visit catsu.dev.

Special Notes

  • ✅ Comprehensive quantization support (float, int8, binary, and more)
  • ✅ Matryoshka embeddings
  • input_type used for prompt injection
  • Normalized embeddings by default
  • German language support

Next Steps

On this page