Gemini

Google's Gemini embedding model offers long context support and Matryoshka embeddings.

Overview

Models: 1 model (gemini-embedding-001)
Key Features: Long context (2048 tokens), Matryoshka (128-3072 dimensions), task-specific optimization
API Docs: Gemini Embeddings

Environment Variable

export GEMINI_API_KEY="your-gemini-api-key"

Supported Parameters

Parameter	Type	Required	Description
`model`	str	Yes	Model identifier (gemini-embedding-001)
`input`	str \| List[str]	Yes	Text(s) to embed
`input_type`	str	No	`"query"` or `"document"`
`task_type`	str	No	Specific task type (see below)
`dimensions`	int	No	Output dimensions: 128-3072
`api_key`	str	No	Override API key for this request

Task Types

Gemini supports specific task types for optimization:

RETRIEVAL_QUERY - Search queries
RETRIEVAL_DOCUMENT - Documents for retrieval
SEMANTIC_SIMILARITY - Semantic similarity tasks
CLASSIFICATION - Text classification
CLUSTERING - Document clustering
QUESTION_ANSWERING - Q&A systems
FACT_VERIFICATION - Fact checking
CODE_RETRIEVAL_QUERY - Code search queries

Examples

Basic Usage

import catsu

client = catsu.Client()

response = client.embed(
    model="gemini-embedding-001",
    input="Hello, Gemini!"
)

print(f"Dimensions: {response.dimensions}")  # 3072 (default)

With task_type

# For search queries
query_response = client.embed(
    model="gemini-embedding-001",
    input="What is quantum computing?",
    task_type="RETRIEVAL_QUERY"
)

# For documents
doc_response = client.embed(
    model="gemini-embedding-001",
    input="Quantum computing uses quantum mechanics...",
    task_type="RETRIEVAL_DOCUMENT"
)

# For classification
classification_response = client.embed(
    model="gemini-embedding-001",
    input="This product is amazing!",
    task_type="CLASSIFICATION"
)

With Custom Dimensions (Matryoshka)

# Smaller dimensions for faster search
response = client.embed(
    model="gemini-embedding-001",
    input="Sample text",
    dimensions=512  # vs default 3072
)

print(f"Dimensions: {response.dimensions}")  # 512

# Minimum dimensions
tiny_response = client.embed(
    model="gemini-embedding-001",
    input="Text",
    dimensions=128
)

Combined Parameters

response = client.embed(
    model="gemini-embedding-001",
    input="AI research paper abstract...",
    input_type="document",
    task_type="SEMANTIC_SIMILARITY",
    dimensions=1024
)

Long Context Example

# Gemini supports up to 2048 tokens
long_text = "..." * 1000  # Very long document

response = client.embed(
    model="gemini-embedding-001",
    input=long_text,
    task_type="RETRIEVAL_DOCUMENT"
)

Async Usage

import asyncio

async def main():
    client = catsu.Client()

    response = await client.aembed(
        model="gemini-embedding-001",
        input="Async Gemini embedding",
        task_type="RETRIEVAL_QUERY",
        dimensions=512
    )

    print(response.embeddings)

asyncio.run(main())

Special Notes

✅ Supports both input_type and task_type parameters
✅ Matryoshka embeddings: 128-3072 dimensions
Long context: up to 2048 tokens per input
Single model (gemini-embedding-001)
task_type provides fine-grained control beyond input_type

Next Steps

Common Parameters: dimensions - Learn about Matryoshka embeddings
Common Parameters: input_type - Query vs document embeddings

On this page