Swarmauri NVIDIA NIM LLM

swarmauri_llm_nvidia_nim connects Swarmauri conversations and toolkits to an NVIDIA NIM for LLMs deployment.

Features

• OpenAI-compatible chat completions through NvidiaNIMModel.
• Tool calling through NvidiaNIMToolModel.
• Synchronous, asynchronous, incremental streaming, and batch workflows inherited from Swarmauri's shared transports.
• Optional discovery from the deployment's /v1/models endpoint.
• Configurable deployment base URL for local, private, and managed NIM endpoints.
• Persistent, injectable Swarmauri HTTP clients with conservative rate-limit retries.

Installation

uv add swarmauri_llm_nvidia_nim

pip install swarmauri_llm_nvidia_nim

Usage

from swarmauri_llm_nvidia_nim import NvidiaNIMModel
from swarmauri_standard.conversations.Conversation import Conversation
from swarmauri_standard.messages.HumanMessage import HumanMessage

conversation = Conversation(messages=[HumanMessage(content="What is accelerated computing?")])
model = NvidiaNIMModel(
    base_url="http://localhost:8000",
    name="meta/llama-3.1-8b-instruct",
)
model.predict(conversation)
print(conversation.get_last().content)

Pass api_key when the deployment requires bearer authentication. Set discover_models=True to populate allowed_models from the running deployment. For tool workflows, instantiate NvidiaNIMToolModel and pass a Swarmauri Toolkit to predict, apredict, stream, or astream.

Pass an IHttpClient as http_client to integrate a custom transport or to test without network access. Pass an IHttpRetryPolicy as retry_policy to replace the default policy, which retries safe reads and rate-limited chat submissions without replaying failed POST requests for generic server errors.

See the NVIDIA NIM LLM API reference for deployment and model-specific capability details.