from raghilda.embedding import EmbeddingNVIDIA, EmbedInputType
# Using NVIDIA's hosted API
provider = EmbeddingNVIDIA(model="nvidia/llama-nemotron-embed-1b-v2")
# Or using a self-hosted NIM container
provider = EmbeddingNVIDIA(
model="nvidia/llama-nemotron-embed-1b-v2",
base_url="http://localhost:8000/v1",
)
# Embed documents for indexing
doc_embeddings = provider.embed(
["Hello world", "Testing embeddings"],
input_type=EmbedInputType.DOCUMENT,
)
# Embed a query for search
query_embedding = provider.embed(
["How do I test embeddings?"],
input_type=EmbedInputType.QUERY,
)embedding.EmbeddingNVIDIA
Creates an embedding function provider backed by NVIDIA’s NIM embedding models.
Usage
embedding.EmbeddingNVIDIA(
model="nvidia/llama-nemotron-embed-1b-v2",
base_url="https://integrate.api.nvidia.com/v1",
api_key=None,
batch_size=20,
truncate="NONE",
)Implements the EmbeddingProvider interface.
NVIDIA’s embedding API is compatible with the OpenAI client library. It supports differentiated query vs document embeddings via the input_type parameter.
Browse available models and get API keys at NVIDIA NIM.
NVIDIA also provides Docker images for self-hosting NIM models. When running a model locally or on your own infrastructure, set the base_url parameter to point to your self-hosted endpoint (e.g., "http://localhost:8000/v1").
Parameters
model: str = "nvidia/llama-nemotron-embed-1b-v2"-
The NVIDIA embedding model to use. Default is “nvidia/llama-nemotron-embed-1b-v2”.
base_url: str = "https://integrate.api.nvidia.com/v1"-
The base URL for the NVIDIA API. Default is “https://integrate.api.nvidia.com/v1”. Change this to point to a self-hosted NIM container (e.g.,
"http://localhost:8000/v1"). api_key: str | None = None-
The API key for authenticating with NVIDIA. If None, it will use the NVIDIA_API_KEY environment variable if set. Not required when using a self-hosted NIM container.
batch_size: int = 20-
The number of texts to process in each batch when calling the API.
truncate: str = "NONE"- Truncation strategy. Default is “NONE”. Set to “END” to truncate inputs that exceed the model’s max token length.