NVIDIA NIM API (Inference Microservices)

NVIDIA NIM provides optimized inference for a wide variety of open-source models, hosted by NVIDIA or deployable on your own hardware.

🚀 Key Specs

  • Model Support: Llama 3.3, Mistral-Large-2, Nemotron-4, Phi-3, etc.
  • Free Allowance: 40 RPM (Requests Per Minute) via the NVIDIA Developer Program.
  • Cost: Free for prototyping and research; requires enterprise license for production scale.
  • Style: Fully OpenAI API compatible.

🗝 How to get an API Key

  1. Join the NVIDIA Developer Program (developer.nvidia.com).
  2. Visit the NVIDIA API Catalog.
  3. Select a model and click Get API Key.

🛠 Usage Example (Python/OpenAI Library)

from openai import OpenAI
 
client = OpenAI(
  base_url="https://integrate.api.nvidia.com/v1",
  api_key="$NVIDIA_API_KEY"
)
 
completion = client.chat.completions.create(
  model="nvidia/llama-3.1-nemotron-70b-instruct",
  messages=[{"role":"user","content":"Explain how NIM works."}],
  temperature=0.5,
  top_p=1,
  max_tokens=1024,
  stream=False
)
 
print(completion.choices[0].message.content)