Using Phi-3 Medium 4K on NVIDIA NIM
Implementation guide · Phi-3 · Microsoft Research
Quick Start
- 1
- 2Use the NVIDIA NIM SDK or REST API to call
phi-3-medium-4k— see the documentation for request format. - 3
Code Examples
About NVIDIA NIM
NIM packages inference runtimes and model profiles into containers that expose standard API surfaces such as chat completions, completions, model listing, tokenization, health, and management endpoints. The hosted API path is useful for prototyping and catalog discovery, while the NGC/container path is the self-hosted route for teams that want GPU-hour infrastructure control, private-network deployment, Kubernetes scaling, or NVIDIA AI Enterprise support. Per-token pricing is not a universal provider-level claim in the current seed data; pricing should stay attached to sourced model-provider rows or NVIDIA's current catalog terms.
NVIDIA NIM is NVIDIA's deployment platform for GPU-accelerated inference microservices. Developers can try hosted NIM APIs through the NVIDIA API Catalog on build.nvidia.com, then move the same model families into self-hosted NIM containers on NVIDIA GPUs in a data center, private cloud, public cloud, or workstation. The catalog positions NIM around optimized open and NVIDIA models, including chat, coding, reasoning, retrieval, vision, speech, and safety use cases, with downloadable model cards and API endpoints where NVIDIA exposes them.
Pricing on NVIDIA NIM
| Type | Rate |
|---|---|
| GPU Hour Rate | $1.00/GPU·hr |
| GPU Config | 1xH100 |
Capabilities
About Phi-3 Medium 4K
The Phi-3 Medium 4K, developed by Microsoft, is a state-of-the-art large language model with 14 billion parameters. It is engineered for efficiency across various tasks, particularly excelling in reasoning capabilities. This model is designed to handle 4,096 token context lengths, allowing for the processing of longer input sequences. Leveraging a dense, decoder-only Transformer architecture, it incorporates techniques like supervised fine-tuning and direct preference optimization to align with human preferences and safety standards. The model supports multilingual data, although it is primarily trained in English. Its lightweight nature allows for deployment on diverse hardware platforms, making it accessible and versatile for both commercial and research purposes. Safety measures are embedded, although further precautions are advised for applications with higher risks.