zoom_in Click to enlarge

Granite Code

▲ ▼

language

description Granite Code Overview

Granite Code is IBM's family of open-source large language models designed for code generation and software development tasks, notable for being released under an Apache 2.0 license for self-hosted deployment.

help Granite Code FAQ

Which Granite Code model sizes can be self-hosted?

IBM released the original Granite Code family in 3B, 8B, 20B, and 34B parameter sizes, with base and instruction-tuned variants. Smaller 3B or quantized 8B models are the practical starting points for machines with limited GPU memory.

Can Granite Code be used commercially?

IBM released the Granite Code model weights under the Apache License 2.0 for research and commercial use. Deployers still need to comply with that license and assess generated code for third-party licensing or security issues.

What programming tasks is Granite Code designed to handle?

The instruction models target code generation, explanation, translation, debugging, and repository-oriented assistance. IBM trained the family across 116 programming languages, although performance varies by language and model size.

How can Granite Code be run locally?

IBM publishes model weights through Hugging Face, and compatible releases can be served with tools such as Transformers, vLLM, llama.cpp, or Ollama. The exact runtime depends on the chosen checkpoint and quantization format, so a 34B model requires far more memory than the 3B edition.

swap_horiz

Looking for Granite Code alternatives? Compare top competitors ranked & scored

arrow_forward